One of the key use cases for generative AI involves answering questions over private datasets, with retrieval-augmented generation (RAG) as the go-to framework. As new RAG techniques emerge, there’s a growing need to benchmark their performance across diverse datasets and metrics.  To meet this need, we’re introducing BenchmarkQED, a new suite of tools that automates RAG benchmarking at scale, available on GitHub (opens in new tab). It includes components for query generation, evaluation, and dataset preparation, each designed to support rigorous, reproducible testing.   BenchmarkQED complements the RAG methods in our open-source GraphRAG library, enabling users to run a GraphRAG-style evaluation across models, metrics, and datasets. GraphRAG uses a large language model (LLM) to generate and summarize entity-based knowledge graphs, producing more comprehensive and diverse answers than standard RAG for large-scale tasks.  In this post, we walk through the core components of BenchmarkQED that contribute to the overall benchmarking process. We also share some of the latest benchmark results comparing our LazyGraphRAG system to competing methods, including a vector-based RAG with a 1M-token context window, where the leading LazyGraphRAG configuration showed significant win rates across all combinations of quality metrics and query classes. In the paper, we distinguish between local queries, where answers are found in a small number of text regions, and sometimes even a single region, and global queries, which require reasoning over large portions of or even the entire dataset.  Conventional vector-based RAG excels at local queries because the regions containing the answer to the query resemble the query itself and can be retrieved as the nearest neighbor in the vector space of text embeddings. However, it struggles with global questions, such as, “What are the main themes of the dataset?” which require understanding dataset qualities not explicitly stated in the text.   AutoQ: Automated query synthesis This limitation motivated the development of GraphRAG a system designed to answer global queries. GraphRAG’s evaluation requirements subsequently led to the creation of AutoQ, a method for synthesizing these global queries for any dataset. AutoQ extends this approach by generating synthetic queries across the spectrum of queries, from local to global. It defines four distinct classes based on the source and scope of the query (Figure 1, top) forming a logical progression along the spectrum (Figure 1, bottom). Figure 1. Construction of a 2×2 design space for synthetic query generation with AutoQ, showing how the four resulting query classes map onto the local-global query spectrum.  AutoQ can be configured to generate any number and distribution of synthetic queries along these classes, enabling consistent benchmarking across datasets without requiring user customization. Figure 2 shows the synthesis process and sample queries from each class, using an AP News dataset. Figure 2. Synthesis process and example query for each of the four AutoQ query classes.  About Microsoft Research Advancing science and technology to benefit humanity View our story Opens in a new tab AutoE: Automated evaluation framework  Our evaluation of GraphRAG focused on analyzing key qualities of answers to global questions. The following qualities were used for the current evaluation: Comprehensiveness: Does the answer address all relevant aspects of the question?  Diversity: Does it present varied perspectives or insights?  Empowerment: Does it help the reader understand and make informed judgments?  Relevance: Does it address what the question is specifically asking?   The AutoE component scales evaluation of these qualities using the LLM-as-a-Judge method. It presents pairs of answers to an LLM, along with the query and target metric, in counterbalanced order. The model determines whether the first answer wins, loses, or ties with the second. Over a set of queries, whether from AutoQ or elsewhere, this produces win rates between competing methods. When ground truth is available, AutoE can also score answers on correctness, completeness, and related metrics. An illustrative evaluation is shown in Figure 3. Using a dataset of 1,397 AP News articles on health and healthcare, AutoQ generated 50 queries per class (200 total). AutoE then compared LazyGraphRAG to a competing RAG method, running six trials per query across four metrics, using GPT-4.1 as a judge. These trial-level results were aggregated using metric-based win rates, where each trial is scored 1 for a win, 0.5 for a tie, and 0 for a loss, and then averaged to calculate the overall win rate for each RAG method. Figure 3. Win rates of four LazyGraphRAG (LGR) configurations across methods, broken down by the AutoQ query class and averaged across AutoE’s four metrics: comprehensiveness, diversity, empowerment, and relevance. LazyGraphRAG outperforms comparison conditions where the bar is above 50%. The four LazyGraphRAG conditions (LGR_b200_c200, LGR_b50_c200, LGR_b50_c600, LGR_b200_c200_mini) differ by query budget (b50, b200) and chunk size (c200, c600). All used GPT-4o mini for relevance tests and GPT-4o for query expansion (to five subqueries) and answer generation, except for LGR_b200_c200_mini, which used GPT-4o mini throughout. Comparison systems were GraphRAG (Local, Global, and Drift Search), Vector RAG with 8k- and 120k-token windows, and three published methods: LightRAG (opens in new tab), RAPTOR (opens in new tab), and TREX (opens in new tab). All methods were limited to the same 8k tokens for answer generation. GraphRAG Global Search used level 2 of the community hierarchy. LazyGraphRAG outperformed every comparison condition using the same generative model (GPT-4o), winning all 96 comparisons, with all but one reaching statistical significance. The best overall performance came from the larger budget, smaller chunk size configuration (LGR_b200_c200). For DataLocal queries, the smaller budget (LGR_b50_c200) performed slightly better, likely because fewer chunks were relevant. For ActivityLocal queries, the larger chunk size (LGR_b50_c600) had a slight edge, likely because longer chunks provide a more coherent context. Competing methods performed relatively better on the query classes for which they were designed: GraphRAG Global for global queries, Vector RAG for local queries, and GraphRAG Drift Search, which combines both strategies, posed the strongest challenge overall. Increasing Vector RAG’s context window from 8k to 120k tokens did not improve its performance compared to LazyGraphRAG. This raised the question of how LazyGraphRAG would perform against Vector RAG with 1-million token context window containing most of the dataset. Figure 4 shows the follow-up experiment comparing LazyGraphRAG to Vector RAG using GPT-4.1 that enabled this comparison. Even against the 1M-token window, LazyGraphRAG achieved higher win rates across all comparisons, failing to reach significance only for the relevance of answers to DataLocal queries. These queries tend to benefit most from Vector RAG’s ranking of directly relevant chunks, making it hard for LazyGraphRAG to generate answers that have greater relevance to the query, even though these answers may be dramatically more comprehensive, diverse, and empowering overall. Figure 4. Win rates of LazyGraphRAG (LGR) over Vector RAG across different context window sizes, broken down by the four AutoQ query classes and four AutoE metrics: comprehensiveness, diversity, empowerment, and relevance. Bars above 50% indicate that LazyGraphRAG outperformed the comparison condition.  AutoD: Automated data sampling and summarization Text datasets have an underlying topical structure, but the depth, breadth, and connectivity of that structure can vary widely. This variability makes it difficult to evaluate RAG systems consistently, as results may reflect the idiosyncrasies of the dataset rather than the system’s general capabilities. The AutoD component addresses this by sampling datasets to meet a target specification, defined by the number of topic clusters (breadth) and the number of samples per cluster (depth). This creates consistency across datasets, enabling more meaningful comparisons, as structurally aligned datasets lead to comparable AutoQ queries, which in turn support consistent AutoE evaluations. AutoD also includes tools for summarizing input or output datasets in a way that reflects their topical coverage. These summaries play an important role in the AutoQ query synthesis process, but they can also be used more broadly, such as in prompts where context space is limited. Since the release of the GraphRAG paper, we’ve received many requests to share the dataset of the Behind the Tech (opens in new tab) podcast transcripts we used in our evaluation. An updated version of this dataset is now available in the BenchmarkQED repository (opens in new tab), alongside the AP News dataset containing 1,397 health-related articles, licensed for open release.   We hope these datasets, together with the BenchmarkQED tools (opens in new tab), help accelerate benchmark-driven development of RAG systems and AI question-answering. We invite the community to try them on GitHub (opens in new tab).  Opens in a new tab

In a Bloomberg interview Wednesday night in downtown San Francisco, Alphabet CEO Sundar Pichai pushed back against concerns that AI could eventually make half the company’s 180,000-person workforce redundant. Instead, Pichai stressed the company’s commitment to growth through at least next year. “I expect we will grow from our current engineering phase even into next year, because it allows us to do more,” Pichai said, adding that AI is making engineers more productive by eliminating tedious tasks and enabling them to focus on more impactful work. Rather than replacing workers, he referred to AI as “an accelerator” that will drive new product development, thereby creating demand for more employees. Alphabet has staged numerous layoffs in recent years, though so far, cuts in 2025 appear to be more targeted than in previous years. It reportedly parted ways with less than 100 people in Google’s cloud division earlier this year and, more recently, hundreds more in its platforms and devices unit. In 2024 and 2023, the cuts were far more severe, with 12,000 people dropped from the company in 2023 and at least another 1,000 employees laid off last year. Looking forward, Pichai pointed to Alphabet’s expanding ventures like Waymo autonomous vehicles, quantum computing initiatives, and YouTube’s explosive growth as evidence of innovation opportunities that continually bubble up. He noted YouTube’s scale in India alone, with 100 million channels and 15,000 channels boasting over one million subscribers. At one point, Pichai said trying to think too far ahead is “pointless.” But he also acknowledged the legitimacy of fears about job displacement, saying when asked about Anthropic CEO Dario Amodei’s recent comments that AI could erode half of entry-level white collar jobs within five years, “I respect that . . .I think it’s important to voice those concerns and debate them.” As the interview wrapped up, Pichai was asked about the limits of AI, and whether it’s possible that the world might never achieve artificial general intelligence, meaning AI that’s as smart as humans at everything. He quickly paused before answering. “There’s a lot of forward progress ahead with the paths we are on, not only the set of ideas we are working on today, [but] some of the newer ideas we are experimenting with,” he said. “I’m very optimistic on seeing a lot of progress. But you know,” he added, “you’ve always had these technology curves where you may hit a temporary plateau. So are we currently on an absolute path to AGI? I don’t think anyone can say for sure.” Techcrunch event Save now through June 4 for TechCrunch Sessions: AI Save $300 on your ticket to TC Sessions: AI—and get 50% off a second. Hear from leaders at OpenAI, Anthropic, Khosla Ventures, and more during a full day of expert insights, hands-on workshops, and high-impact networking. These low-rate deals disappear when the doors open on June 5. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 REGISTER NOW

In a Bloomberg interview Wednesday night in downtown San Francisco, Alphabet CEO Sundar Pichai pushed back against concerns that AI could eventually make half the company’s 180,000-person workforce redundant. Instead, Pichai stressed the company’s commitment to growth through at least next year. “I expect we will grow from our current engineering phase even into next year, because it allows us to do more,” Pichai said, adding that AI is making engineers more productive by eliminating tedious tasks and enabling them to focus on more impactful work. Rather than replacing workers, he referred to AI as “an accelerator” that will drive new product development, thereby creating demand for more employees. Alphabet has staged numerous layoffs in recent years, though so far, cuts in 2025 appear to be more targeted than in previous years. It reportedly parted ways with less than 100 people in Google’s cloud division earlier this year and, more recently, hundreds more in its platforms and devices unit. In 2024 and 2023, the cuts were far more severe, with 12,000 people dropped from the company in 2023 and at least another 1,000 employees laid off last year. Looking forward, Pichai pointed to Alphabet’s expanding ventures like Waymo autonomous vehicles, quantum computing initiatives, and YouTube’s explosive growth as evidence of innovation opportunities that continually bubble up. He noted YouTube’s scale in India alone, with 100 million channels and 15,000 channels boasting over one million subscribers. At one point, Pichai said trying to think too far ahead is “pointless.” But he also acknowledged the legitimacy of fears about job displacement, saying when asked about Anthropic CEO Dario Amodei’s recent comments that AI could erode half of entry-level white collar jobs within five years, “I respect that . . .I think it’s important to voice those concerns and debate them.” As the interview wrapped up, Pichai was asked about the limits of AI, and whether it’s possible that the world might never achieve artificial general intelligence, meaning AI that’s as smart as humans at everything. He quickly paused before answering. “There’s a lot of forward progress ahead with the paths we are on, not only the set of ideas we are working on today, [but] some of the newer ideas we are experimenting with,” he said. “I’m very optimistic on seeing a lot of progress. But you know,” he added, “you’ve always had these technology curves where you may hit a temporary plateau. So are we currently on an absolute path to AGI? I don’t think anyone can say for sure.” Techcrunch event Save now through June 4 for TechCrunch Sessions: AI Save $300 on your ticket to TC Sessions: AI—and get 50% off a second. Hear from leaders at OpenAI, Anthropic, Khosla Ventures, and more during a full day of expert insights, hands-on workshops, and high-impact networking. These low-rate deals disappear when the doors open on June 5. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 REGISTER NOW

If you’ve ever been locked out of your work computer, you know the urgent need to reach IT support. Unfortunately, help desk staff are often busy assisting others, which can mean a significant delay before you regain access. Andrei Serban realized the critical importance of IT’s manual work when he was a product lead on Rippling’s apps and integrations team. Serban, who joined Rippling after it acquired his coding security startup Fuzzbuzz in 2023, saw an opportunity to automate many of the help desk’s basic tasks with AI, from resetting passwords, to granting access to apps like Figma and Miro, to routine troubleshooting. Serban couldn’t wait to get started on his vision. He left Rippling last year to found Console, a startup on a mission to help IT teams reduce mundane, repetitive tasks, thereby freeing up time for help desk professionals to work on more strategic and sophisticated projects. While trying to automate help desk functions isn’t new, Console differentiates itself from existing competitors — including Moveworks, which was acquired by ServiceNow in March for $2.85 billion — by forgoing long, complex installation processes. Thanks to its easy integration with Slack, Console’s AI assistant can be made available to everyone in the company in just a few weeks. “We’re able to get there so fast because we don’t require you to replace your help desk,” he said. Console sees itself as an AI co-worker that up-levels existing help desk professionals. Serban even suggested that as companies digitize more of their operations, the IT function will become increasingly vital. Employees message Console on Slack and the startup’s AI agent quickly responds to requests because it knows everything about the user, from their specific laptop model to the applications they have permission to use. Console’s AI can resolve over 50% of the tasks on its own, and it will loop in someone from IT for more complex issues, Serban said. Techcrunch event Save now through June 4 for TechCrunch Sessions: AI Save $300 on your ticket to TC Sessions: AI—and get 50% off a second. Hear from leaders at OpenAI, Anthropic, Khosla Ventures, and more during a full day of expert insights, hands-on workshops, and high-impact networking. These low-rate deals disappear when the doors open on June 5. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 REGISTER NOW The fast install time attracted customers, including Scale AI, Flock Safety, and Calendly. These customer wins have led Console to secure $6.2 million in seed funding from Thrive Capital. Vince Hankes, a partner at Thrive, said that the firm, which has backed many of the leading AI companies, including OpenAI, Cursor, and Scale AI, has had a thesis about artificial intelligence’s potential to assist with IT tasks since ChatGPT was first released in the fall of 2022. While Hankes acknowledged that Console’s fast integration might make it seem easily replaceable, he believes its AI technology will improve so quickly with user adoption that it will become an indispensable, deeply integrated system essential to everyone at the customer’s company. Console’s goal is to eventually start supporting other functions, including answering employee requests about HR, finance, and legal. “We want Console to be an employee’s first call for help,” Serban said.  Topics

/May 30, 2025/4:28 p.m. ETTrump Attacks Harvard With Social Media Screening for All VisasThis pilot program will soon be expanded across the country.Spencer Platt/Getty ImagesThe Trump administration has begun carrying out its expanded vetting for student visa applicants, surveilling their social media accounts to make sure they aren’t posting anything in support of Palestine, which the administration considers antisemitic. This vetting will start with Harvard visa applicants but is expected to be adopted nationwide.Secretary of Stato Marco Rubio sent a cable to all U.S. embassies and consulates on Thursday ordering them to “conduct a complete screening of the online presence of any nonimmigrant visa applicant seeking to travel to Harvard University for any purpose.” That would apply not just to students but also to faculty, staff, and researchers visiting the university.The Trump administration is taking particular interest in people who have their social media accounts on “private,” an obvious, ominous crossing of boundaries.The State Department has ordered officers to examine “whether the lack of any online presence, or having social media accounts restricted to ‘private’ or with limited visibility, may be reflective of evasiveness and call into question the applicant’s credibility.”This is yet another instance of Harvard serving as a test subject for the administration’s larger crackdown on free speech and international students at American universities. Trump has already revoked billions of dollars in research funding from the Massachusetts school, and even banned it from admitting any international students at all, although the latter policy was temporarily revoked by a judge. Most Recent Post/May 30, 2025/3:53 p.m. ETStephen Miller Grilled on Musk’s Drug Use as Wife Lands New GigTrump’s chief adviser seems desperate to avoid questions on Elon Musk. Does that have anything to do with his wife’s new job? Francis Chung/Politico/Bloomberg/Getty ImagesStephen Miller had a dismissive response Friday to new reports of Elon Musk’s drug use during Trump’s campaign last year. CNN’s Pamela Brown asked the far-right Trump adviser if there was “any drug testing or requests for him to drug test when he was in the White House given the fact that he was also a contractor with the government.”  A chuckling Miller ignored the question and said, “Fortunately for you and all of the friends at CNN, you’ll have the opportunity to ask Elon all the questions you want today yourself,” before he then segued into the Trump administration’s anti-immigrant agenda. “The drugs I’m concerned about are the drugs that are coming across the border from the criminal cartels that are killing hundreds of thousands of Americans,” Miller said. Perhaps Miller laughed instead of answering because his wife, Katie Miller, has left her job as adviser and spokesperson for the Department of Government Efficiency to work full-time for Musk and his companies. Miller has probably had enough of Musk, as he has also been subtweeting the tech oligarch, trying to refute Musk’s criticisms that the Republican budget bill would raise the deficit. “The Big Beautiful Bill is NOT an annual budget bill and does not fund the departments of government. It does not finance our agencies or federal programs,” Miller said, in a long X post earlier this week. Is there bad blood between Miller and Musk that has now spiraled because Miller’s wife is working for the tech oligarch and fellow fascism enthusiast? Most Recent Post/May 30, 2025/3:19 p.m. ETOld Man Trump Repeatedly Fumbles in Weird Speech Praising Elon MuskDonald Trump couldn’t keep some of his words straight as he marked the supposed end of Elon Musk’s tenure at the White House.Kevin Dietsch/Getty ImagesHours after reports emerged Friday that Elon Musk had been under the influence of heavy drugs during his time advising the president, Musk and Donald Trump stumbled and fumbled their way through a White House press conference recognizing the end of the tech billionaire’s special government employee status.The wildly unusual joint conference featured Musk’s black eye, a giant gold key that Trump said he only gives to “very special people,” cringe-worthy regurgitations by Musk of Trump’s take on his Pulitzer Board defamation suit, and claims that Musk’s unpopular and controversial time in the White House was not quite over.But as Trump continued to praise Musk and his time atop the Department of Government Efficiency, the president’s verbal gaffes became more apparent. He claimed that DOGE had uncovered $42 million in wasteful spending, referring to expenditures related to Uganda, which Trump pronounced as “oo-ganda.” The 78-year-old also mentioned he would have Musk’s DOGE cuts “cauterized by Congress,” though he quickly corrected himself by saying they would be “affirmed by Congress,” instead. Trump’s on-camera slippage has gotten worse in recent weeks: Earlier this month, Trump dozed off while in a meeting with Crown Prince Mohammed bin Salman in Riyadh, Saudi Arabia. That is despite the fact that the president received a clean bill of health in a medical report released in April that described Trump as being in “excellent health,” including neurological functioning.Musk, meanwhile, refused to acknowledge emerging reports of his alleged drug use. But the news of White House drug use under Trump’s helm is nothing new: In fact, if the reports prove true, it would be little more than a return to form. Last year, a report by the Department of Defense inspector general indicated that the West Wing operated more like a pill mill than the nation’s highest office. Common pills included modafinil, Adderall, fentanyl, morphine, and ketamine, according to the Pentagon report. But other, unlisted drugs—like Xanax—were equally easy to come by from the White House Medical Unit, according to anonymous sources that spoke to Rolling Stone.While other presidents were known to take a mix of drug cocktails to fight off back pain (like JFK) or bad moods (like Nixon), no previous administrations matched the level of debauchery of Trump’s, whose in-office pharmacists unquestioningly handed out highly addictive substances to staffers who needed pick-me-ups or energy boosts—no doctor’s exam, referral, or prescription required.“It was kind of like the Wild West. Things were pretty loose. Whatever someone needs, we were going to fill this,” another source told Rolling Stone in March 2024.Meanwhile, pharmacists described an atmosphere of fear within the West Wing, claiming they would be “fired” if they spoke out or would receive negative work assignments if they didn’t hand pills over to staffers.Read more about the press conference:Trump and Elon Musk Have Ominous Warning About Future of DOGEMost Recent Post/May 30, 2025/3:00 p.m. ETElon Musk Gives Strange Excuse for Massive Black EyeMusk showed up a press conference with Donald Trump sporting a noticeable shiner.Kevin Dietsch/Getty ImagesElon Musk sported what looked like a black eye during his DOGE goodbye press conference with President Trump on Friday. When asked about it, he blamed the bruise on his 5-year-old son punching him in the face. “Mr. Musk … is your eye OK? What happened to your eye; I noticed there’s a bruise there?” one reporter finally asked near the end of the press conference.“Well, I wasn’t anywhere near France,” Musk said, in a weak attempt at a joke regarding footage of French President Emmanuel Macron’s wife slapping him in the face.“I was just horsing around with [my son] little X and said, ‘Go ’head and punch me in the face,’ and he did. Turns out even a 5-year-old punching you in the face actually does—”“That was X that did it? X could do it!” Trump chimed in. “If you knew X …”“I didn’t really feel much at the time; I guess it bruises up. But I was just messing around with the kids.”Musk chose an impeccable time to show up to a press conference with a black eye. Earlier in the day, The New York Times reported on Musk’s rampant drug use on and off the campaign trail, as the world’s richest man frequently mixed ketamine and psychedelics and kept a small box of pills, mostly containing Adderall. The shiner only adds to speculation around his personal habits.More on that Times report:Elon Musk Was on Crazy Combo of Drugs During Trump CampaignMost Recent Post/May 30, 2025/2:51 p.m. ETTrump and Elon Musk Have Ominous Warning About Future of DOGEElon Musk’s time as a government employee has come to an end, but his time with Donald Trump has not.Kevin Dietsch/Getty ImagesDespite the fanfare over Elon Musk’s supposed departure from the Department of Government Efficiency, Donald Trump says that the billionaire bureaucrat isn’t really going anywhere.“Many of the DOGE people are staying behind, so they’re not leaving. And Elon’s not really leaving. He’s gonna be back and forth, I think. I have a feeling. It’s his baby, and he’s gonna be doing a lot of things,” Trump said during a press conference in the Oval Office Friday.The press conference was held to mark the end of Musk’s time as a so-called “special government employee,” a title that allowed him to bypass certain ethics requirements during his 134-day stint in Trump’s administration. The president made sure to give Musk a gaudy golden key—what it actually unlocks went totally unaddressed—to make sure he could get back into the White House. “This is not the end of DOGE, but really the beginning,” Musk said, promising that DOGE’s “influence” would “only grow stronger” over time.Earlier Friday, the billionaire bureaucrat shared a post on X asserting that the legacy of DOGE was more psychological than anything else. Surely, it will take longer than four months to forget the image of Musk running around with a chainsaw. Read more about Musk:Elon Musk Was on Crazy Combo of Drugs During Trump CampaignMost Recent Post/May 30, 2025/1:21 p.m. ETDem Governor Vetoes Ban on Surprise Ambulance Bills in Shocking MoveThe bill had unanimous support in both chambers of the state legislature.Michael Ciaglo/Getty ImagesColorado’s Democratic Governor Jared Polis has vetoed a bill that would ban surprise billing by ambulance companies, over the unanimous objections of both chambers of the state legislature. Why would Polis veto a bill that’s popular with everyone, even Colorado Republicans? The governor wrote in his veto statement that drafting errors in the bill made it “unimplementable” and estimated that it would make insurance premiums go up by as much as $0.73 to $2.15 per person. “I am committed to working with proponents and sponsors to protect Coloradans from surprise bills, but I encourage all parties to work towards a more reasonable reimbursement rate that mitigates premium impacts and nets a better deal for Colorado families,” Polis wrote. In Colorado, if legislators in both chambers repass the bill with a two-thirds majority, they can override the governor’s veto, especially considering that the bill passed with the support of every single legislator. But the legislature adjourned on May 7, meaning that the bill has to be passed again when the legislature reconvenes in January.  For some reason, ending surprise ambulance billing nationally is not the slam-dunk issue it should be. Congress ended most surprise medical bills in 2020 but exempted ground ambulances from the bill. Was Polis’s veto due to badly drafted language and a (seemingly modest) price hike in insurance premiums, as he said, or was it for a different, more nefarious reason? We might not know unless and until the bill is reintroduced next year. More on surprise ambulance bills:Congress Doesn’t Care About Your Surprise Ambulance Bill Most Recent Post/May 30, 2025/12:21 p.m. ETTrump’s Pardons Since Jan 6 Spree Show an Infuriatingly Corrupt TrendSince his January 6 pardon spree, Donald Trump has tended to grant clemency a little closer to home.Saul Loeb/AFP/Getty ImagesA good chunk of the white-collar criminals pardoned by Donald Trump after his massive “Day One” pardoning spree either have a political or financial tie to him.The president has issued 60 pardons since he offered political forgiveness to some 1,600 individuals charged in the January 6, 2021, attack on the U.S. Capitol. But out of those subsequent 60 unrelated to the attack, 12 people—or roughly one in five—were already in Trump’s orbit, according to ABC News.They included several politicos, including former Illinois Governor Rod Blagojevich, who was convicted on several counts of corruption, including for an attempt to sell Barack Obama’s Senate seat after he left the position for the White House; former Republican Representative Michael Grimm, who pleaded guilty to tax fraud; former Nevada gubernatorial candidate Michele Fiore, who allegedly stole public funds intended to commemorate a slain police officer; and former Tennessee state Senator Brian Kelsey, who pleaded guilty to campaign finance fraud in 2022.Trump also pardoned major financiers of his presidential campaigns. Trevor Milton, the founder of the Nikola electric vehicle company, donated nearly $2 million toward Trump’s 2024 campaign. Imaad Zuberi, who has donated to both parties, issued “at least $800,000 to committees associated with Trump and the Republican Party,” ABC reported.Others helped Trump advance his retribution campaign against his political enemies, or helped advance his own image in the broader Republican Party. Devon Archer and Jason Galanis, both former business partners of Hunter Biden, accused the younger Biden of leveraging his father’s name and influence in order to conduct business overseas. Archer had defrauded a Native American tribal entity, while Galanis was serving time for multiple offenses. Trump also forgave Todd and Julie Chrisley—reality TV stars known for their show Chrisley Knows Best who were sentenced to a combined 19 years on fraud and tax evasion charges—after their daughter Savannah Chrisley spoke at the 2024 Republican National Convention.Speaking to press Friday after her parents’ release, Savannah Chrisley said that the “biggest misconception right now is I either paid for a pardon or slept for a pardon—,” but she couldn’t finish her sentence before Todd interjected: “That’s something I would have done,” he said.Read who else Trump is thinking of pardoning:Trump Considering Pardons for Men Who Tried to Kill Gretchen WhitmerMost Recent Post/May 30, 2025/12:04 p.m. ETTrump Knew He Was Deporting Innocent People to El Salvador All AlongMany of the people deported to El Salvador have no criminal record, and Donald Trump knew it.Michael M. Santiago/Getty ImagesDonald Trump’s administration was well aware that many of the 238 Venezuelan immigrants it shipped off to a notorious megaprison in El Salvador had no criminal records at all, according to a Friday report from ProPublica.  While Trump officials claimed that the deportees were brutal gang members and “the worst of the worst,” only 32 of the deportees had actually been convicted of crimes, and most of them were minor offenses such as traffic violations, according to data from the Department of Homeland Security reviewed by ProPublica, The Texas Tribune, and a team of journalists from Venezuelan media outlets. One of the men, 23-year-old Maikol Gabriel López Lizano, faced a misdemeanor charge after he was arrested in 2023 for riding his bike and drinking a can of beer.Little more than half of the deportees, 130 of the 238, were charged only with violating U.S. immigration laws. Twenty of them had criminal records from other countries. The U.S. government data showed that 67 individuals had pending charges, with only six being for violent crimes. In several cases, the government data about the pending charges differed from what ProPublica was able to find. In some cases, the men had actually been convicted, and in one, the charges had been dropped. But in many cases, these individuals were remanded to a foreign prison before their criminal cases were ever resolved. The Trump administration has touted allegations of gang affiliation as a justification for denying the deportees their due process rights. But none of the men’s names appeared on a list of roughly 1,400 alleged Tren de Aragua members kept by the Venezuelan government, ProPublica reported. Trump’s border czar Tom Homan tried desperately in March to downplay reporting that many of these individuals did not have criminal records. “A lot of gang members don’t have criminal histories, just like a lot of terrorists in this world, they’re not in any terrorist databases, right?” Homan said on ABC News. But the methods the government relies on to classify individuals as gang members—such as identification of gang-affiliated tattoos—have been disproven by experts. Not only were many of the men who were deported not proven gang members, they weren’t even criminals, and by denying them the right to due process, they were remanded to a foreign prison notorious for human rights abuses without ever getting to prove it. Trump has continued to pressure the Supreme Court to allow him to sidestep due process as part of his massive deportation campaign, claiming that the judiciary has no right to intrude on matters of “foreign policy.” But immigrants residing on U.S. soil—who are clearly not the bloodthirsty criminals the administration insists they are—are still subject to protections under U.S. law. Read more about the deportations:Trump Asks Supreme Court to Help Him Deport People Wherever He WantsMost Recent Post/May 30, 2025/11:41 a.m. ETJoni Ernst Stoops to Shocking Low When Told Medicaid Cuts Will KillSenator Joni Ernst had a disgusting answer when confronted by a constituent at her town hall about Trump’s budget bill.Drew Angerer/Getty ImagesRepublican Senator Joni Ernst had a particularly unhinged response to questions from her constituents at a town hall in Parkersburg, Iowa, on Friday.Ernst was asked about the GOP’s budget bill kicking people off of Medicaid, and her condescending answer quickly became callous and flippant as the Iowa politician smirked at the audience.“When you are arguing about illegals that are receiving Medicaid, 1.4 million, they’re not eligible, so they will be coming off, so—” Ernst began, before an audience member shouted, “People are going to die!”“People are not—well, we all are going to die,” Ernst responded, as the audience drowned her in loud protests.What was Ernst thinking with that answer? Almost every Republican town hall this year has gone badly for the politician holding it, thanks to President Trump upending the federal government, and Ernst surely knew that choosing death over Medicaid wouldn’t go over well with the crowd. Earlier this week in Nebraska, Representative Mike Flood was heckled after he admitted that he didn’t read the budget bill.Ersnt’s town hall wasn’t even the first one in Iowa to go badly for a Republican. On Wednesday, Representative Ashley Hinson was met with jeers and boos, with audience members in Decorah, Iowa calling her a fraud and a liar. But at least Hinson had the good sense not to seemingly embrace death over a vital, lifesaving government program. More on Trump’s bill:Here Are the Worst Things in Trump’s Big, Beautiful Bill Most Recent Post/May 30, 2025/11:35 a.m. ETKetanji Brown Jackson Blasts “Botched” Supreme Court Ruling on TPSSupreme Court Justice Ketanji Brown Jackson, in a scathing disssent, called out the rest of the court for allowing Trump’s harmful executive order to stand.Anna Moneymaker/Getty ImagesSupreme Court Justice Ketanji Brown Jackson thinks the Supreme Court “botched” a decision to allow the Trump administration to revoke the Temporary Protected Status protections of about 500,000 Haitian, Cuban, Nicaraguan, and Venezuelan immigrants.Jackson and fellow liberal Justice Sonia Sotomayor were the only two dissenters.“The Court has plainly botched this assessment today. It requires next to nothing from the Government with respect to irreparable harm,” Jackson wrote in the dissent. “And it undervalues the devastating consequences of allowing the Government to precipitously upend the lives of and livelihoods of nearly half a million noncitizens while their legal claims are pending.”TPS is a long-standing program that allowed those 500,000 immigrants to stay in the U.S. after they fled violence and risk in their home countries. After the Supreme Court’s ruling, all of them are at high risk of sudden deportation. “It is apparent that the government seeks a stay to enable it to inflict maximum predecision damage,” Jackson wrote.Read the full dissent here.View More Posts

The company behind the Opera browser is launching yet another AI tool with Opera Neon, an agentic AI browser. This basically means that it's a browser with an AI agent built in, which can go beyond answering questions and will purportedly be able to browse the internet for you to help you get various things done. This includes helping you plan trips, booking vacations, and even creating web apps with simple natural language prompts. Oddly enough, this isn't Opera's first go at agentic AI, as it follows the announcement for the standard Opera browser's Browser Operator tool. Technically, Browser Operator isn't released yet, but it seems the difference is that Neon's use cases will be a bit broader, as the AI will supposedly even able to generate content in the cloud while you're offline.The catch is that Neon isn't free, and is currently invite-only. Opera says it'll require a paid subscription when it launches, and while the company hasn't revealed the pricing or the launch date yet, you can join a waitlist to get notified about details closer to release, plus get in line for an invite. Opera says you'll be able to use the integrated AI as a chatbot and it will be able to search the web to find answers for you. It'll also be able to handle repetitive tasks such as filling forms and shopping. The biggest draw seems to be its ability to create content, though. On the Opera Neon website, a sample screenshot shows a someone requesting the AI to make a "retro snake game" for them.One plus going for this product is that it claims to be able to analyze webpages without recording your screen all the time (looking at you, Recall). Opera also claims that your browsing history, website data, and login information will be stored locally on your computer, which is good for anyone with privacy concerns.It goes without saying that all of these features will only be as useful as the AI model is accurate. The last thing I'd want is to have a faceless AI model book an overpriced hotel in a shady location, so I'll be taking all these trip planning claims with a pinch of salt until I see Neon in action. While launching new products always gets more attention, the sheer number of Opera's recent releases means that its browser lineup is getting a bit confusing. Opera currently has the following browsers listed on its website: Opera Browser, Opera GX, Opera Air, and Opera Mini. This makes Opera Neon the fifth product in the lineup. Each has its own specialty, but I'm starting to feel a little choice paralysis here.

Q: My question is how best to set up an Xbox for my kid. I want to know how to control use time and whether I should use my email to set up. – Guillermo from Utah A: You're in luck! Console parental controls have come a long way from the days of hiding NES systems (or their controllers) from unruly kids. Microsoft, in particular, has made it very easy through its Xbox Family Settings app for iOS and Android, which supports Xbox Series X/S and Xbox One systems. Once you've added your kids to your Xbox family group, you'll be able to manage and see reports of their playtime, restrict access to specific content, require parental approval for purchases and sort incoming friend requests. In typical Microsoft fashion, though, setting up Xbox family accounts can be a bit confusing, so let's go over it step by step. Microsoft How to set up an Xbox for your child Turn on your Xbox and sign in with your Microsoft account. Press the Xbox button on your controller, and head to Profile & system > Settings > account > Family settings > Manage family members. Choose Add to family > Add new. Unfortunately, Microsoft requires an email address for new accounts. You can either enter your child's existing email if they have one, or choose the Get a new email option to create an Outlook email address. (This may seem a bit absurd for younger children, so if you don't want to create an email for your 6 year old, you may be better off just physically controlling access to consoles like the old days.) Next, you'll have to verify the addition of a new family group member. You can do so by either choosing This Xbox on the next screen and entering your Microsoft account details, or selecting Their phone/PC to have an email sent to you. In either case, you'll have to consent to the new member. What is a Microsoft child account? In its documentation, Microsoft says "a child account is defined as any Microsoft account that’s affiliated with an adult Microsoft account when the age of the child or teen is less than the age of majority for their country or region." These accounts are broken into "Child" and "Teen" categories, which specify age ranges between eight and 12, and 13 to 17. Microsoft notes there can be differences depending on location, like in South Korea where teen accounts range from 13 to 18. Devindra Hardawar for Engadget Are there any issues with Microsoft child accounts? As useful as the Xbox Family app and child accounts may seem, they can also lead to problems while playing Minecraft, according to comments on Reddit. Reddit user "Microdad_" said they weren't able to gift a game to their child's account, they had to deal with a buggy website to allow them to play Minecraft and after all of that trouble their child still can't play in their realm. User "Electronic_Ocelot825" encountered similar issues, replying, "You are spot on. I go through same [sic] crap and it drives me to the point where I am furious.” Have a tech question you’d like Engadget to answer? As tech reporters, the Engadget staff is always answering questions from readers, friends and family about electronics, software, gaming, big tech policies and more. So we decided to write down our answers. This question came from a coworker from Yahoo, Engadget's corporate parent. If you’ve got a tech-related question you’d like us to answer for you, please email ask@engadget.com.This article originally appeared on Engadget at https://www.engadget.com/gaming/xbox/ask-engadget-how-do-i-set-up-an-xbox-for-my-kid-110022860.html?src=rss

ExpressVPN is good at its job. It's easy to be skeptical of any service with a knack for self-promotion, but don't let ExpressVPN's hype distract you from the fact that it keeps its front-page promise of "just working." Outside of solid security, the two best things ExpressVPN offers are fast speeds and a simple interface. Our tests showed only a 7% average drop in download speed and a 2% loss of upload speed, worldwide. And while the lack of extra features may frustrate experienced users, it makes for a true set-and-forget VPN on any platform. This isn't to say ExpressVPN is without flaws — it's nearly bereft of customization options and it's notably more expensive than its competition — but it beats most VPNs in a head-to-head matchup. For this review, we followed our rigorous 10-step VPN testing process, exploring ExpressVPN's security, privacy, speed, interfaces and more. Whether you read straight through or skip to the sections that are most important for you, you should come away with all the information you need to decide whether to subscribe. Editors' note: We're in the process of rebooting all of our VPN reviews from scratch. Once we do a fresh pass on the top services, we'll be updating each review with a rating and additional comparative information. Table of contents Findings at a glance Installing, configuring and using ExpressVPN ExpressVPN speed test: Very fast averages ExpressVPN security test: Checking for leaks How much does ExpressVPN cost? ExpressVPN side apps and bundles Close-reading ExpressVPN's privacy policy Can ExpressVPN change your virtual location? Investigating ExpressVPN's server network Extra features of ExpressVPN ExpressVPN customer support options ExpressVPN background check: From founding to Kape Technologies Final verdict Findings at a glance Category Notes Installation and UI All interfaces are clean and minimalist, with no glitches and not enough depth to get lost in Windows and Mac clients are similar in both setup and general user experience Android and iOS are likewise almost identical, but Android has a nice-looking dark mode Speed Retains a worldwide average of 93% of starting download speeds Upload speeds average 98% of starting speeds Latency rises with distance, but global average stayed under 300 ms in tests Security OpenVPN, IKEv2 and Lightway VPN protocols all use secure ciphers Packet-sniffing test showed working encryption We detected no IP leaks Blocks IPv6 and WebRTC by default to prevent leaks Pricing Base price: $12.95 per month or $99.95 per year Lowest prepaid rate: $4.99 per month Can save money by paying for 28 months in advance, but only once per account 30-day money-back guarantee Bundles ExpressVPN Keys password manager and ID alerts included on all plans Dedicated IP addresses come at an extra price ID theft insurance, data removal and credit scanning available to new one-year and two-year subscribers for free 1GB eSIM deal included through holiday.com Privacy policy No storage of connection logs or device logs permitted The only risky exceptions are personal account data (which doesn't leave the ExpressVPN website) and marketing data (which the policy says should be anonymized) An independent audit found that ExpressVPN's RAM-only server infrastructure makes it impossible to keep logs Virtual location change Successfully unblocked five international Netflix libraries, succeeding on 14 out of 15 attempts Server network 164 server locations in 105 countries 38% of servers are virtual, though most virtual locations are accessed through physical servers within 1,000 miles A large number of locations in South America, Africa and central Asia Features Simple but effective kill switch Can block ads, trackers, adult sites and/or malware sites but blocklists can't be customized Split tunneling is convenient but unavailable on iOS and modern Macs Aircove is the best VPN router, albeit expensive Customer support Setup and troubleshooting guides are organized and useful, with lots of screenshots and videos Live chat starts with a bot but you can get to a person within a couple minutes Email tickets are only accessible from the mobile apps or after live chat has failed Background check Founded in 2009; based in the British Virgin Islands Has never been caught selling or mishandling user data Turkish police seized servers in 2017 but couldn't find any logs of user activity Owned by Kape Technologies, which also owns CyberGhost and Private Internet Access A previous CIO formerly worked on surveillance in the United Arab Emirates; no evidence of shady behavior during his time at ExpressVPN Windows Version 12 leaked some DNS requests when Split Tunneling was active Installing, configuring and using ExpressVPN This section focuses on how it feels to use ExpressVPN on each of the major platforms where it's available. The first step for any setup process is to make an account on expressvpn.com and buy a subscription. Windows Once subscribed, download the Windows VPN from either expressvpn.com or the Microsoft Store, then open the .exe file. Click "Yes" to let it make changes, wait for the install, then let your computer reboot. Including the reboot, the whole process takes 5-10 minutes, most of it idle. To finish, you'll need your activation code, which you can find by going to expressvpn.com and clicking "Setup" in the top-right corner. You can install ExpressVPN's Windows app from the Microsoft store, but we found the website more convenient. Sam Chapman for Engadget Extreme simplicity is the watchword for all ExpressVPN's designs. The Windows client's launch panel consists of three buttons and less than ten words. You can change your location or let the app pick a location for you — the "Smart Location" is the server with the best combination of being nearby and unburdened. Everything else is crammed into the hamburger menu at the top left. Here, in seven tabs, you'll find the Network Lock kill switch, the four types of content blockers, the split tunneling menu and the option to change your VPN protocol. You can also add shortcuts to various websites, useful if you regularly use your VPN for the same online destinations. To sum up, there's almost nothing here to get in the way: no delays, no snags, no nested menus to get lost in. It may be the world's most ignorable VPN client. That's not a bad thing at all. Mac ExpressVPN's app for macOS is almost identical in design to its Windows app. The process for downloading and setting it up is nearly the same too. As on Windows, it can be downloaded from the App Store or sideloaded directly from the expressvpn.com download center. Only a few features are missing and a couple others have been added. Split tunneling is gone (unless you're still on a macOS lower than 11), and you won't see the Lightway Turbo setting. ExpressVPN recommends some servers, but it's easy to search the whole list. Sam Chapman for Engadget Mac users do gain access to the IKEv2 protocol, along with the option to turn off automatic IPv6 blocking — Windows users have to leave it blocked at all times. Almost every website is still accessible via IPv4, but it's useful if you do need to access a specific IPv6 address while the VPN is active. Android Android users can download ExpressVPN through the Google Play Store. Open the app, sign in and you're ready to go. The Android app has a very nice dark-colored design, only slightly marred by an unnecessary information box about how long you've used the VPN this week. ExpressVPN's Android app puts a little more information on the screen than it needs to, but still runs well. Sam Chapman for Engadget There's a large button for connecting. Clicking on the server name takes you to a list of locations. On this list, you can either search or scroll and can choose individual locations within a country that has more than one. We connected to as many far-flung server locations as we could, but not a single one took longer than a few seconds. The options menu is organized sensibly, with no option located more than two clicks deep. You will see a couple of options here that aren't available on desktop, the best of which is the ability to automatically connect to your last-used ExpressVPN server whenever your phone connects to a non-trusted wifi network. There are also a few general security tools: an IP address checker, DNS and WebRTC leak testers and a password generator. These are also available on the website, but here, they're built into the app. With the exception of the latter, we'd recommend using third-party testing tools instead — even a VPN with integrity has an incentive to make its own app look like it's working. iPhone and iPad You can only install ExpressVPN's iOS app through the app store. During setup, you may need to enter your password to allow your phone to use VPN configurations. Otherwise, there are no major differences from the Android process. ExpressVPN looks good on iPhone and iPad. Sam Chapman for Engadget The interface is not quite as pleasing as the dark-mode Android app, but it makes up for that by cutting out some of the clutter. The tabs and features are similar, though split tunneling and shortcuts are absent. Also, both mobile apps make customer support a lot more accessible than their desktop counterparts — plus, mobile is the only way to send email support tickets. Browser extension ExpressVPN also includes browser extensions for Firefox and Chrome. These let you connect, disconnect and change server locations without leaving your browser window. It's nice, but not essential unless you have a very specific web browser flow you like. ExpressVPN speed test: Very fast averages Connecting to a VPN almost always decreases your speed, but the best VPNs mitigate the drop as much as possible. We used Ookla's speed testing app to see how much of your internet speed ExpressVPN preserves. For this test, we emphasized the locations ExpressVPN uses for most of its virtual servers, including the Netherlands, Brazil, Germany and Singapore. Some terms before we start: Latency, measured in milliseconds (ms), is the time it takes one data packet to travel between your device and a web server through the VPN. Latency increases with distance. It's most important for real-time tasks like video chatting and online gaming. Download speed, measured in megabits per second (Mbps), is the amount of information that can download onto your device at one time — such as when loading a web page or streaming a video. Upload speed, also measured in Mbps, is the amount of information your device can send to the web at once. It's most important for torrenting, since the amount of data you can seed determines how fast you can download in exchange. The table below shows our results. We conducted this on Windows, using the automatic protocol setting with the Lightway Turbo feature active — a recent ExpressVPN addition that keeps speed more consistent by processing connections in parallel. Server location Latency (ms) Increase factor Download speed (Mbps) Percentage dropoff Upload speed (Mbps) Percentage dropoff Portland, Oregon, USA (unprotected) 18 -- 58.77 -- 5.70 -- Seattle, Washington, USA (best server) 26 1.4x 54.86 6.7% 5.52 3.2% New York, NY, USA 156 8.7x 57.25 2.6% 5.57 2.3% Amsterdam, Netherlands 306 17x 53.83 8.4% 5.58 2.1% São Paulo, Brazil 371 20.6x 53.82 8.4% 5.65 0.9% Frankfurt, Germany 404 22.4x 55.71 5.2% 5.67 0.5% Singapore, Singapore 381 21.2x 52.76 10.2% 5.64 1.0% Average 274 15.2x 54.71 6.9% 5.61 1.6% These are extremely good results. ExpressVPN is a winner on both download and upload speed. No matter where we went in the world, we never lost more than about 7% of our download speeds, and upload lost an astoundingly low average of 2%. This suggests that ExpressVPN deftly distributes its user load between servers to eliminate bottlenecks. This Ookla speedtest shows you can still get fast internet while connected to ExpressVPN -- our unprotected speeds are around 58 Mbps. Sam Chapman for Engadget The latency numbers look worse, but the rise in the table is less sharp than we projected. Ping length depends far more on distance than download speed does, so we expect it to shoot up on servers more than 1,000 miles from our location. Keeping the average below 300 ms, as ExpressVPN does here, is a strong showing. ExpressVPN security test: Checking for leaks A VPN's core mission is to hide your IP address and make you untraceable online. Our task in this section is to figure out if ExpressVPN can carry out this mission every time you connect. While we can't be 100% certain, the tests we'll run through below have led us to believe that ExpressVPN is currently leak-proof. Available VPN protocols A VPN protocol is like a common language that a VPN server can use to mediate between your devices and the web servers you visit. If a VPN uses outdated or insecure protocols, or relies on unique protocols with no visible specs or source code, that's a bad sign. Not all protocols are available on all apps, but Mac has the full range. Sam Chapman for Engadget ExpressVPN gives you a selection of three protocols: IKEv2, OpenVPN and Lightway. The first two are solid choices that support the latest encryption algorithms. OpenVPN has been fully open-source for years and is the best choice if privacy is your goal. While IKEv2 started life as a closed project by Microsoft and Cisco, ExpressVPN uses an open-source reverse-engineering, which is both better for privacy and quite fast. Lightway is the odd one out, a protocol you'll only find on ExpressVPN, though its source code is available on Github. It's similar to WireGuard, in that both reach for faster speeds and lower processing demands by keeping their codebases slim. However, Lightway was recently rewritten in Rust to better protect the keys stored in its memory. Ultimately, you can't go wrong with any of ExpressVPN's protocol options. 99% of the time, your best choice will be to set the controls to Automatic and let the VPN decide which runs best. Testing for leaks ExpressVPN is one of the best services, but it's not leak-proof (as you can read in the Background Check below). Luckily, checking for DNS leaks is a simple matter of checking your IP address before and after connecting to a VPN server. If the new address matches the VPN server, you're good; if not, your VPN is leaking. First, we checked the Windows app with split tunneling active to ensure the flaw really had been patched. We tested several servers and didn't find any leaks, which suggests the patch worked, though leaks were rare even before ExpressVPN fixed the vulnerability. We checked our IP while connected to the virtual India location, which is run from a physical server in Singapore. Don't worry -- it still looks like India to streaming services. Sam Chapman for Engadget In fact, we didn't find any leaks on any ExpressVPN server we tested on any platform. Though questions remain about iOS, as you'll see later in this section, that's a problem on Apple's end that even the best VPNs can do very little about for now. The most common cause of VPN leaks is the use of public DNS servers to connect users to websites, which can mistakenly send browsing activity outside the VPN's encrypted tunnel. ExpressVPN avoids the risks of the public system by installing its own DNS resolvers on every server. This is the key factor behind its clean bill of health in our leak testing. Two other common flaws can lead to VPN leaks: WebRTC traffic and IPv6. The former is a communication protocol used in live streaming and the latter is a new IP standard designed to expand domain availability. Both are nice, but currently optional, so ExpressVPN automatically blocks both to ensure there's no opportunity for leaks to arise. One note about VPN security on iOS: it's a known and continuing problem that iOS VPNs do not prevent many online apps from communicating with Apple directly, outside the VPN tunnel. This risks leaking sensitive data, even with Lockdown Mode active in iOS 16. A blog post by Proton VPN shares a workaround: connect to a VPN server, then turn Airplane Mode on and off again to end all connections that were active before you connected to the VPN. Testing encryption We finished up our battery of security tests by checking out ExpressVPN's encryption directly. Using WireShark, a free packet sniffer, we inspected what it looks like when ExpressVPN transmits data from one of its servers to the internet. The screenshot below shows a data stream encrypted with Lightway UDP. After connecting to ExpressVPN, HTTP packets were rendered unreadable while in transit. Sam Chapman for Engadget That lack of any identifiable information, or even readable information, means encryption is working as intended. We repeated the test several times, always getting the same result. This left us satisfied that ExpressVPN's core features are working as intended. How much does ExpressVPN cost? ExpressVPN subscriptions cost $12.95 per month. Long-term subscriptions can bring the monthly cost down, but the great deals they offer tend to only last for the first billing period. A 12-month subscription costs $99.95 and includes three months for free with your first payment, costing a total of $6.67 per month. The bonus disappears for all subsequent years, raising the monthly cost to $8.33. You can also sign up for 28 months at a cost of $139.72, but this is also once-only — ExpressVPN can only be renewed at the $99.95 per year level. There are two ways to test ExpressVPN for free before making a financial commitment. Users on iOS and Android can download the ExpressVPN app without entering any payment details and use it free for seven days. On any platform, there's a 30-day money-back guarantee, which ExpressVPN has historically honored with no questions asked. You will have to pay before you can use it, though. In our opinion, ExpressVPN's service is solid enough that it's worth paying extra. Perhaps not this much extra, but that depends on what you get out of it. We recommend using the 30-day refund period and seeing how well ExpressVPN works for you. If it's a VPN you can enjoy using, that runs fast and unblocks everything you need, that's worth a server's weight in gold. ExpressVPN side apps and bundles ExpressVPN includes some special features that work mostly or wholly separate from its VPN apps. Some of these come free with a subscription, while others add an extra cost. Every subscription includes the ExpressVPN keys password manager. This is available under its own tab on the Android and iOS apps. On desktop, you'll need to download a separate extension from your browser's store, then sign in using your account activation code. It's available on all Chromium browsers, but not Firefox. Starting in 2025, new subscribers get an eSIM plan through holiday.com, a separate service linked to ExpressVPN. The baseline 1GB holiday eSIM plans last for 5 days and can apply to countries, regions, or the entire world (though it's not clear whether the package deal applies to the regional and global plans). Longer-term plans include larger eSIM plans. You can add a dedicated IP address to your ExpressVPN subscription for an additional cost per month. A dedicated IP lets you use the same IP address every time you connect to ExpressVPN. You can add the address to whitelists on restricted networks, and you're assured to never be blocked because of someone else's bad activity on a shared IP. Unlike many of its competitors, ExpressVPN doesn't currently offer antivirus or online storage services, but there is a comprehensive bundle of ID protection tools called Identity Defender. We haven't reviewed any of these products in detail, but here's a list for reference: ID Alerts will inform you if any of your sensitive information is leaked or misused online. It's free with all plans, but you'll have to enter your personal information on your ExpressVPN account page or a mobile app. ID Theft Insurance grants up to $1 million in identity theft reimbursement and comes free with new ExpressVPN one-year or two-year subscriptions. It's not yet available to those who subscribed before it launched in October 2024. Data Removal scans for your information in data brokerages and automatically requests that it be deleted. It's also free with one-year and two-year plans. Credit Scanner is only available for United States users. It monitors your activity on the three credit bureaus so you can quickly spot any suspicious transactions. The Identity Defender features are currently only available to new ExpressVPN customers in the US. Close-reading ExpressVPN's privacy policy Although we worry that the consolidation of VPN brands under the umbrella of Kape Technologies (ExpressVPN's parent company) will make the industry less competitive, we don't believe it's influencing ExpressVPN to take advantage of its users' privacy. To confirm, and get a full sense of what sort of privacy ExpressVPN promises its users, we set out to read ExpressVPN's privacy policy in detail. It's long, but thankfully aimed at casual users instead of lawyers. You can see it for yourself here. In the introduction, ExpressVPN states that it does not keep either activity logs (such as a user's browsing history while connected to the VPN) or connection logs (such as the duration of a user's session and their IP address, which can be used to extrapolate browsing activity). It then specifies the seven types of data it's legally allowed to collect: Data used to sign up for an account, such as names, emails and payment methods. VPN usage data which is aggregated and can't be traced to any individual. Credentials stored in the ExpressVPN Keys password manager. Diagnostic data such as crash reports, which are only shared upon user request. IP addresses authorized for MediaStreamer, which is only for streaming devices that don't otherwise support VPN apps. Marketing data collected directly from the app — a "limited amount" that's kept anonymous. Data voluntarily submitted for identity theft protection apps. Of those seven exceptions, the only ones that count as red flags are account data and marketing data. Both categories are highly personal and could be damaging if mishandled. Fortunately, complying with subpoenas is not one of the allowed uses listed for either data category, nor does the policy let ExpressVPN sell the data to other private parties. The only really annoying thing here is that if you ask ExpressVPN to delete your personal data, you won't be able to use your account from then on. You aren't even eligible for a refund in this case, unless you're within 30 days of your initial subscription. As for marketing data, ExpressVPN collects device fingerprints and location data when you sign up for an account on its website. The privacy policy also claims this is anonymized, as its "systems are engineered to decouple such data from personally identifiable information." Audits corroborate this, as we'll see in the next section. So, while it would be better if ExpressVPN didn't collect any personal data at all, its practices don't appear to pose a risk to anything you do while using the VPN — just the ExpressVPN website. Privacy audits VPN providers often get third-party accounting firms to audit their privacy policies. The idea is that a well-known firm won't mortgage its reputation to lie on behalf of a VPN, so their results can be trusted. For the last several years, ExpressVPN has had KPMG look over its privacy policy and relevant infrastructure (see "TrustedServer" below). KPMG's most recent report, completed in December 2023 and released in May 2024, found that ExpressVPN had enough internal controls in place that users could trust its privacy policy. The report is freely available to read. This is a very good sign, though we're looking out for a more up-to-date audit soon. TrustedServer "TrustedServer" is a marketing term ExpressVPN uses for its RAM-only server infrastructure. RAM-only servers have no hard drives for long-term storage and return to a standard disk image with every reboot. This makes it theoretically impossible to store user activity logs on them, even if ExpressVPN wanted to do that. The KPMG audit, linked above, reports that TrustedServer works as advertised. Between its many clean privacy audits and the Turkish server incident in 2017, we're prepared to say ExpressVPN is a private VPN, in spite of its aggravating exception for marketing. Can ExpressVPN change your virtual location? Next, we tested whether ExpressVPN can actually convince websites that you're somewhere other than your real location. Our security tests have already proven it can hide your IP address, but it takes more than leak-proofing to fool streaming sites these days — Netflix and the others have gotten very good at combing through metadata to sniff out proxy users. The process for testing this is a lot like how we handled the DNS leak tests: try several different servers and see if we get caught. We checked five sample locations outside the U.S. to see if we a) got into Netflix and b) saw different titles in the library. The results are below. Server Location Unblocked Netflix? Library changed? Canada Y Y United Kingdom Y (second try; Docklands failed) Y Slovakia Y Y India Y Y (different from UK library) Australia Y Y In fifteen tests, ExpressVPN slipped up only once. Docklands, the UK server it chose as the fastest, wasn't able to access Netflix. We switched to a server labeled simply "London" and unblocked it without issue. ExpressVPN can change your virtual location so you can explore the wonderful world of K-drama. Sam Chapman for Engadget All the other locations got us access to an alternate Netflix library on the first try. We even checked whether the India server, which is physically located in the UK, showed us different videos than the UK servers. It did, which makes us even more confident that ExpressVPN's virtual locations are airtight. Investigating ExpressVPN's server network ExpressVPN users can connect to a total of 164 server locations in 105 countries and territories. These locations are reasonably well distributed across the globe, but as with all VPNs, there's a bias toward the northern hemisphere. There are 24 locations in the U.S. alone and a further 66 in Europe. That isn't to say users in the Global South get nothing. ExpressVPN has IP addresses from nine nations in South America (Argentina, Brazil, Bolivia, Chile, Colombia, Ecuador, Peru, Uruguay and Venezuela) and six in Africa (Algeria, Egypt, Ghana, Kenya, Morocco and South Africa). The network even includes Kazakhstan, Uzbekistan and Mongolia, impressive since central Asia may be the region most often shafted by VPNs. However, many of these servers have virtual locations different from their real ones. For those of you choosing a server based on performance instead of a particular IP address, ExpressVPN's website has a helpful list of which servers are virtual. The bad news is that it's a big chunk of the list. A total of 63 ExpressVPN locations are virtual, or 38% of its entire network. To reduce the sting, ExpressVPN takes care to locate virtual servers as close to their real locations as possible. Its virtual locations in Indonesia and India are physically based in Singapore. This isn't always practical, leading to some awkwardness like operating a Ghana IP address out of Germany. But it helps ExpressVPN perform better in the southern hemisphere. Extra features of ExpressVPN Compared to direct competitors like NordVPN and Surfshark, ExpressVPN doesn't have many special features. It's aimed squarely at the casual market and will probably disappoint power users. Having said that, what they do include works well. In this section, we'll run through ExpressVPN's four substantial features outside its VPN servers themselves. Network Lock kill switch "Network Lock" is the name ExpressVPN gives to its kill switch (though it's called "Network Protection" on mobile). A VPN kill switch is a safety feature that keeps you from broadcasting outside the VPN tunnel. If it ever detects that you aren't connected to a legitimate ExpressVPN server, it cuts off your internet access. You won't be able to get back online until you either reconnect to the VPN or disable Network Lock. ExpressVPN's kill switch is called Network Lock on desktop, and Network Protection on mobile (Android pictured) Sam Chapman for Engadget This is important for everyone, not just users who need to hide sensitive traffic. The recently discovered TunnelVision bug theoretically allows hackers to set up fake public wi-fi networks through which they redirect you to equally fake VPN servers, which then harvest your personal information. It's unlikely, but not impossible, and a kill switch is the best way to prevent it — the switch always triggers unless you're connected to a real server in the VPN's network. Like most of ExpressVPN's features, all you can do with Network Lock is turn it on and off. You can also toggle whether you'll still be able to access local devices while the kill switch is blocking your internet — this is allowed by default. Threat manager, ad blocker and parental controls ExpressVPN groups three tools under the heading of "advanced protection" — Threat Manager, an ad blocker and parental controls. Threat Manager consists of two checkboxes: one that blocks your browser from communicating with activity tracking software and one that blocks a list of websites known to be used for malware. Check any of these boxes to use the pre-set blocklists whenever you're connected to ExpressVPN. Sam Chapman for Engadget You can't customize the lists, so you're limited to what ExpressVPN considers worthy of blocking. They share their sources on the website. While the lists are extensive and open-source, they rely on after-the-fact reporting and can't detect and block unknown threats like a proper antivirus. The adblock and parental control options work the same way: check a box to block everything on the list, uncheck it to allow everything through. In tests, the ad blocker was nearly 100% effective against banner ads, but failed to block any video ads on YouTube or Netflix. The parental control option blocks a list of porn sites. It's an easy option for concerned parents, but only works while ExpressVPN is connected. As such, it's meant to be used in conjunction with device-level parental controls that prevent the child from turning off or uninstalling the VPN client. Split tunneling Sometimes, you'll find it helpful to have your device getting online through two different IP addresses at once — one for your home services and one for a location you're trying to spoof. That's where split tunneling is helpful: it runs some apps through the VPN while leaving others unprotected. This can also improve your speeds, since the VPN needs to encrypt less in total. You can configure split tunneling through either a blocklist or an allowlist. Sam Chapman for Engadget ExpressVPN includes split tunneling on Windows, Android and Mac (though only on versions 10 and below). You can only split by app, not by website, but it's still pretty useful. For example, you can have BitTorrent handling a heavy download in the background while you use your browser for innocuous activities that don't need protecting. ExpressVPN Aircove router By now, it should be clear that we find ExpressVPN to be a highly reliable but often unexceptional VPN service. However, there's one area in which it's a clear industry leader: VPN routers. ExpressVPN Aircove is, to our knowledge, the only router with a built-in commercial VPN that comes with its own dashboard interface. Usually, installing a VPN on your router requires tinkering with the router control panel, which turns off all but the most experienced users — not to mention making it a massive pain to switch to a new server location. Aircove's dashboard, by contrast, will be instantly familiar to anyone who already knows how to use an ExpressVPN client. It even allows different devices in your home to connect to different locations through the router VPN. Aircove's biggest drawback is its price. Currently retailing at $189 (not including an ExpressVPN subscription), it's around three times more expensive than an aftermarket router fitted with free VPN firmware. Some of you might still find the convenience worth the one-time payment. ExpressVPN customer support options ExpressVPN's written help pages are some of the best on the market. Its live chat is more of a mixed bag, and complex questions may cause delays. However, it is at least staffed with human agents who aim to reply accurately, rather than resolve your ticket as quickly as possible. You can directly access both live chat and email from ExpressVPN's mobile apps (on desktop, you'll have to go to the website). Sam Chapman for Engadget We approached ExpressVPN's support features with a simple question: "If I requested that ExpressVPN delete all my personal data, would I be able to get a refund for my unused subscription time?" (Remember from the Privacy Policy section that submitting a full deletion request also cancels your ExpressVPN account.) Our first stop was expressvpn.com/support, the written support center and FAQ page. It's divided into setup guides, troubleshooting, account management and information on each of ExpressVPN's products. The setup guides are excellent, including screenshots and clearly written steps; each one includes a video guide for those who learn better that way. Troubleshooting is just as good — no videos, but the same standards of clarity and usefulness prevail. The section starts with general problems, then delves into specific issues you might face on each operating system. Each article clearly derives from a real customer need. The live support experience To get answers on our refund question, we visited the account management FAQs. This section stated that the refund policy only applies within 30 days of purchase. Pretty clear-cut, but we still wanted an answer on our special case, so we contacted live chat by clicking the button at the bottom-right of every FAQ page. Live chat is in the bottom-right corner of every page of expressvpn.com. Sam Chapman for Engadget Live chat starts with an AI assistant, which is not too hard to get past — just ask it a question it can't answer, then click "Transfer to an Agent." We got online with (what claimed to be) a human in less than a minute. Answering the question took longer and involved an uncomfortable 10-minute silence, but we did get a clear verdict from a real person: refunds are within 30 days only, no matter what. If the live chat agent can't answer your question, you'll be redirected to open an email support ticket. Annoyingly, there's no way to go directly to email support through the website or desktop apps, though mobile users have the option to skip directly there. ExpressVPN background check: From founding to Kape Technologies ExpressVPN launched in 2009, which makes it one of the oldest consumer VPNs in continual operation. In more than 15 years of operation, it's never been caught violating its own privacy policy, though its record isn't free of more minor blemishes. Headquarters in the British Virgin Islands Founders Dan Pomerantz and Peter Burchhardt registered the company in the British Virgin Islands from the start to take advantage of that territory's favorable legal environment for online privacy. The BVIs have no law requiring businesses to retain data on their users, and the process for extraditing data is famously difficult, requiring a direct order from the highest court. In 2021, the BVI implemented the Data Protection Act (DPA) [PDF link], which prevents companies based in the territory from accessing data on their users anywhere in the world. It's a great privacy law in theory, modeled on best-in-class legislation in the EU. However, we couldn't find any evidence that its supervising authority — the Office of the Information Commissioner — has a leader or staff. In other words, while ExpressVPN is not legally required to log any data on its users, there's technically nobody stopping them from doing so. Whether you trust the jurisdiction depends on whether you trust the company itself. Let's see what the other evidence says. Security and privacy incidents Two significant incidents stand out from ExpressVPN's 16-year history. In 2017, when Andrei Karlov, Russia's ambassador to Turkey, was shot to death at an art show. Turkish police suspected someone had used ExpressVPN to mask their identity while they deleted information from social media accounts belonging to the alleged assassin. To investigate, they confiscated an ExpressVPN server to comb for evidence. They didn't find anything. A police seizure is the best possible test of a VPN's approach to privacy. The provider can't prepare beforehand, fake anything, or collude with investigators. The Turkey incident is still one of the best reasons to recommend ExpressVPN, though eight years is a long time for policy to change. The second incident began in March 2024, when a researcher at CNET informed ExpressVPN that its version 12 for Windows occasionally leaked DNS requests when users enabled the split tunneling feature. While these users remained connected to an ExpressVPN server, their browsing activity was often going directly to their ISP, unmasked. The bug only impacted a few users, and to their credit, ExpressVPN sprang into action as soon as they learned about it. The team had it patched by April, as confirmed by the researcher who initially discovered the vulnerability. But while their quick and effective response deserves praise, it's still a mark against them that a journalist noticed the bug before they did. Kape Technologies ownership and management questions In 2021, an Israeli-owned, UK-based firm called Kape Technologies purchased a controlling interest in ExpressVPN. In addition to ExpressVPN, privately held Kape owns CyberGhost, Private Internet Access, and Zenmate (before it merged into CyberGhost). As shown on its website, it also owns Webselenese, publisher of VPN review websites WizCase and vpnMentor, which poses an apparent conflict of interest. When reached for comment, a representative for ExpressVPN said that "ExpressVPN does not directly engage with, nor seek to influence, the content on any Webselenese site," and pointed us to disclosure statements on the websites in question — here's one example. Even so, it's a good reminder not to take VPN reviews at face value without knowing who's behind them (Engadget is owned by Yahoo, which does not own any VPNs). Diving deeper into the background of Kape's ownership will lead you to owner Teddy Sagi. Go back far enough, and you'll see he did prison time in Israel and was mentioned in the Pandora Papers, among other things. More recently, headlines about the billionaire have focused more his businesses in the online gambling and fintech arenas, as well as his real estate ventures. An ExpressVPN representative told us that "Kape's brands continue to operate independently," and our investigation bore that out — we couldn't find any proof that Kape or Sagi have directly attempted to influence ExpressVPN's software or daily operations. Closer to the immediate day-to-day operations of ExpressVPN was the company's employment of Daniel Gericke as CTO from 2019 through 2023. During that time, the US Justice Department announced it had fined Gericke and two others for their previous employment on a surveillance operation called Project Raven, which the United Arab Emirates (UAE) used to spy on its own citizens. The revelation prompted a public response from ExpressVPN defending its decision to hire Gericke, arguing that "[t]he best goalkeepers are the ones trained by the best strikers." ExpressVPN's representative confirmed that the company still stands by that linked statement. Gericke parted ways with ExpressVPN in October 2023, per his LinkedIn profile. While we don't know what we don't know, we can say that ExpressVPN has not notably changed its public-facing security and privacy policies during the time it's been connected to Kape, Sagi, or Gericke. In the end, how much ExpressVPN's history matters to you is a personal choice. If you object to any current or past actions by Kape Technologies or Teddy Sagi, there are other premium VPN options you might prefer. If you need more information to make up your mind, we recommend reading through CNET's 2022 deep dive on ExpressVPN's corporate history. Final verdict ExpressVPN is the VPN we most often recommend to beginners. It takes zero training to use, and consistently gets past filters on streaming sites. It also runs in the background with virtually no impact. If anything is worth the high price of admission, it's the excellent speeds distributed evenly across the worldwide server network. However, for certain specific cases, ExpressVPN may not be the best choice. There's no way to set up your own server locations, like NordVPN offers, and no double VPN connections, like you can build for yourself on Surfshark. Its corporate background is more suspect than the entities backing Proton VPN, and unlike Mullvad, ExpressVPN doesn't work in China — it's so well-known that the government targets its servers specifically. We suggest going with ExpressVPN for general online privacy, for spoofing locations in your home country while traveling, or if you regularly need to unblock sites in other countries. That encompasses 19 of every 20 users, which is fine by us, as ExpressVPN is a great service. It's just more of a reliable old screwdriver than a multi-tool. This article originally appeared on Engadget at https://www.engadget.com/cybersecurity/vpn/expressvpn-review-2025-fast-speeds-and-a-low-learning-curve-160052884.html?src=rss

In June, we hosted a webinar featuring experts from Arm, the Unity Accelerate Solutions team, and SYBO Games, the creator of Subway Surfers. The resulting roundtable focused on profiling tips and strategies for mobile games, the business implications of poor performance, and how SYBO shipped a hit mobile game with 3 billion downloads to date.Let’s dive into some of the follow-up questions we didn’t have time to cover during the webinar. You can also watch the full recording.We hear a lot about the Unity Profiler in relation to CPU profiling, but not as much about the Profile Analyzer (available as a Unity package). Are there any plans to improve it or integrate it into the core Profiler toolset?There are no immediate plans to integrate the Profile Analyzer into the core Editor, but this might change as our profiling tools evolve.Does Unity have any plans to add an option for the GPU Usage Profiler module to appear in percentages like it does in milliseconds?That’s a great idea, and while we can’t say yes or no at the time of this blog post, it’s a request that’s been shared with our R&D teams for possible future consideration.Do you have plans for tackling “Application Not Responding” (ANR) errors that are reported by the Google Play store and don’t contain any stack trace?Although we don’t have specific plans for tracking ANR without stack trace at the moment, we will consider it for the future roadmap.How can I share my feedback to help influence the future development of Unity’s profiling tools?You can keep track of upcoming features and share feedback via our product board and forums. We are also conducting a survey to learn more about our customers’ experience with the profiling tools. If you’ve used profiling tools before (either daily or just once) or are working on a project that requires optimization, we would love to get your input. The survey is designed to take no more than 5–10 minutes to complete.By participating, you’ll also have the chance to opt into a follow-up interview to share more feedback directly with the development team, including the opportunity to discuss potential prototypes of new features.Is there a good rule for determining what counts as a viable low-end device to target?A rule of thumb we hear from many Unity game developers is to target devices that are five years old at the time of your game’s release, as this helps to ensure the largest user base. But we also see teams reducing their release-date scope to devices that are only three years old if they’re aiming for higher graphical quality. A visually complex 3D application, for example, will have higher device requirements than a simple 2D application. This approach allows for a higher “min spec,” but reduces the size of the initial install base. It’s essentially a business decision: Will it cost more to develop for and support old devices than what your game will earn running on them?Sometimes the technical requirements of your game will dictate your minimum target specifications. So if your game uses up large amounts of texture memory even after optimization, but you absolutely cannot reduce quality or resolution, that probably rules out running on phones with insufficient memory. If your rendering solution requires compute shaders, that likely rules out devices with drivers that can’t support OpenGL ES 3.1, Metal, or Vulkan.It’s a good idea to look at market data for your priority target audience. For instance, mobile device specs can vary a lot between countries and regions. Remember to define some target “budgets” so that benchmarking goals for what’s acceptable are set prior to choosing low-end devices for testing.For live service games that will run for years, you’ll need to monitor their compatibility continuously and adapt over time based on both your actual user base and current devices on the market.Is it enough to test performance exclusively on low-end devices to ensure that the game will also run smoothly on high-end ones?It might be, if you have a uniform workload on all devices. However, you still need to consider variations across hardware from different vendors and/or driver versions.It’s common for graphically rich games to have tiers of graphical fidelity – the higher the visual tier, the more resources required on capable devices. This tier selection might be automatic, but increasingly, users themselves can control the choice via a graphical settings menu. For this style of development, you’ll need to test at least one “min spec” target device per feature/workload tier that your game supports.If your game detects the capabilities of the device it’s running on and adapts the graphics output as needed, it could perform differently on higher end devices. So be sure to test on a range of devices with the different quality levels you’ve programmed the title for.Note: In this section, we’ve specified whether the expert answering is from Arm or Unity.Do you have advice for detecting the power range of a device to support automatic quality settings, particularly for mobile?Arm: We typically see developers doing coarse capability binning based on CPU and GPU models, as well as the GPU shader core count. This is never perfect, but it’s “about right.” A lot of studios collect live analytics from deployed devices, so they can supplement the automated binning with device-specific opt-in/opt-out to work around point issues where the capability binning isn’t accurate enough.As related to the previous question, for graphically rich content, we see a trend in mobile toward settings menus where users can choose to turn effects on or off, thereby allowing them to make performance choices that suit their preferences.Unity: Device memory and screen resolution are also important factors for choosing quality settings. Regarding textures, developers should be aware that Render Textures used by effects or post-processing can become a problem on devices with high resolution screens, but without a lot of memory to match.Given the breadth of configurations available (CPU, GPU, SOC, memory, mobile, desktop, console, etc.), can you suggest a way to categorize devices to reduce the number of tiers you need to optimize for?Arm: The number of tiers your team optimizes for is really a game design and business decision, and should be based on how important pushing visual quality is to the value proposition of the game. For some genres it might not matter at all, but for others, users will have high expectations for the visual fidelity.Does the texture memory limit differ among models and brands of Android devices that have the same amount of total system memory?Arm: To a first-order approximation, we would expect the total amount of texture memory to be similar across vendors and hardware generations. There will be minor differences caused by memory layout and alignment restrictions, so it won’t be exactly the same.Is it CPU or GPU usage that contributes the most to overheating on mobile devices?Arm: It’s entirely content dependent. The CPU, GPU, or the DRAM can individually overheat a high-end device if pushed hard enough, even if you ignore the other two completely. The exact balance will vary based on the workload you are running.What tips can you give for profiling on devices that have thermal throttling? What margin would you target to avoid thermal throttling (i.e., targeting 20 ms instead of 33 ms)?Arm: Optimizing for frame time can be misleading on Android because devices will constantly adjust frequency to optimize energy usage, making frame time an incomplete measure by itself. Preferably, monitor CPU and GPU cycles per frame, as well as GPU memory bandwidth per frame, to get some value that is independent of frequency. The cycle target you need will depend on each device’s chip design, so you’ll need to experiment.Any optimization helps when it comes to managing power consumption, even if it doesn’t directly improve frame rate. For example, reducing CPU cycles will reduce thermal load even if the CPU isn’t the critical path for your game.Beyond that, optimizing memory bandwidth is one of the biggest savings you can make. Accessing DRAM is orders of magnitude more expensive than accessing local data on-chip, so watch your triangle budget and keep data types in memory as small as possible.Unity: To limit the impact of CPU clock frequency on the performance metrics, we recommend trying to run at a consistent temperature. There are a couple of approaches for doing this:Run warm: Run the device for a while so that it reaches a stable warm state before profiling.Run cool: Leave the device to cool for a while before profiling. This strategy can eliminate confusion and inconsistency in profiling sessions by taking captures that are unlikely to be thermally throttled. However, such captures will always represent the best case performance a user will see rather than what they might actually see after long play sessions. This strategy can also delay the time between profiling runs due to the need to wait for the cooling period first.With some hardware, you can fix the clock frequency for more stable performance metrics. However, this is not representative of most devices your users will be using, and will not report accurate real-world performance. Basically, it’s a handy technique if you are using a continuous integration setup to check for performance changes in your codebase over time.Any thoughts on Vulkan vs OpenGL ES 3 on Android? Vulkan is generally slower performance-wise. At the same time, many devices lack support for various features on ES3.Arm: Recent drivers and engine builds have vastly improved the quality of the Vulkan implementations available; so for an equivalent workload, there shouldn’t be a performance gap between OpenGL ES and Vulkan (if there is, please let us know). The switch to Vulkan is picking up speed and we expect to see more people choosing Vulkan by default over the next year or two. If you have counterexamples of areas where Vulkan isn’t performing well, please get in touch with us. We’d love to hear from you.What tools can we use to monitor memory bandwidth (RAM <-> VRAM)?Arm: The Streamline Profiler in Arm Mobile Studio can measure bandwidth between Mali GPUs and the external DRAM (or system cache).Should you split graphical assets by device tiers or device resolution?Arm: You can get the best result by retuning assets, but it’s expensive to do. Start by reducing resolution and frame rate, or disabling some optional post-processing effects.What is the best way to record performance metric statistics from our development build?Arm: You can use the Performance Advisor tool in Arm Mobile Studio to automatically capture and export performance metrics from the Mali GPUs, although this comes with a caveat: The generation of JSON reports requires a Professional Edition license.Unity: The Unity Profiler can be used to view common rendering metrics, such as vertex and triangle counts in the Rendering module. Plus you can include custom packages, such as System Metrics Mali, in your project to add low-level Mali GPU metrics to the Unity Profiler.What are your recommendations for profiling shader code?You need a GPU Profiler to do this. The one you choose depends on your target platform. For example, on iOS devices, Xcode’s GPU Profiler includes the Shader Profiler, which breaks down shader performance on a line-by-line basis.Arm Mobile Studio supports Mali Offline Compiler, a static analysis tool for shader code and compute kernels. This tool provides some overall performance estimates and recommendations for the Arm Mali GPU family.When profiling, the general rule is to test your game or app on the target device(s). With the industry moving toward more types of chipsets (Apple M1, Arm, x86 by Intel, AMD, etc.), how can developers profile and pinpoint issues on the many different hardware configurations in a reasonable amount of time?The proliferation of chipsets is primarily a concern on desktop platforms. There are a limited number of hardware architectures to test for console games. On mobile, there’s Apple’s A Series for iOS devices and a range of Arm and Qualcomm architectures for Android – but selecting a manageable list of representative mobile devices is pretty straightforward.On desktop it’s trickier because there’s a wide range of available chipsets and architectures, and buying Macs and PCs for testing can be expensive. Our best advice is to do what you can. No studio has infinite time and money for testing. We generally wouldn’t expect any huge surprises when comparing performance between an Intel x86 CPU and a similarly specced AMD processor, for instance. As long as the game performs comfortably on your minimum spec machine, you should be reasonably confident about other machines. It’s also worth considering using analytics, such as Unity Analytics, to record frame rates, system specs, and player options’ settings to identify hotspots or problematic configurations.We’re seeing more studios move to using at least some level of automated testing for regular on-device profiling, with summary stats published where the whole team can keep an eye on performance across the range of target devices. With well-designed test scenes, this can usually be made into a mechanical process that’s suited for automation, so you don’t need an experienced technical artist or QA tester running builds through the process manually.Do you ever see performance issues on high-end devices that don’t occur on the low-end ones?It’s uncommon, but we have seen it. Often the issue lies in how the project is configured, such as with the use of fancy shaders and high-res textures on high-end devices, which can put extra pressure on the GPU or memory. Sometimes a high-end mobile device or console will use a high-res phone screen or 4K TV output as a selling point but not necessarily have enough GPU power or memory to live up to that promise without further optimization.If you make use of the current versions of the C# Job System, verify whether there’s a job scheduling overhead that scales with the number of worker threads, which in turn, scales with the number of CPU cores. This can result in code that runs more slowly on a 64+ core Threadripper™ than on a modest 4-core or 8-core CPU. This issue will be addressed in future versions of Unity, but in the meantime, try limiting the number of job worker threads by setting JobsUtility.JobWorkerCount.What are some pointers for setting a good frame budget?Most of the time when we talk about frame budgets, we’re talking about the overall time budget for the frame. You calculate 1000/target frames per second (fps) to get your frame budget: 33.33 ms for 30 fps, 16.66 ms for 60 fps, 8.33 ms for 120 Hz, etc. Reduce that number by around 35% if you’re on mobile to give the chips a chance to cool down between each frame. Dividing the budget up to get specific sub-budgets for different features and/or systems is probably overkill except for projects with very specific, predictable systems, or those that make heavy use of Time Slicing.Generally, profiling is the process of finding the biggest bottlenecks – and therefore, the biggest potential performance gains. So rather than saying, “Physics is taking 1.2 ms when the budget only allows for 1 ms,” you might look at a frame and say, “Rendering is taking 6 ms, making it the biggest main thread CPU cost in the frame. How can we reduce that?”It seems like profiling early and often is still not common knowledge. What are your thoughts on why this might be the case?Building, releasing, promoting, and managing a game is difficult work on multiple fronts. So there will always be numerous priorities vying for a developer’s attention, and profiling can fall by the wayside. They know it’s something they should do, but perhaps they’re unfamiliar with the tools and don’t feel like they have time to learn. Or, they don’t know how to fit profiling into their workflows because they’re pushed toward completing features rather than performance optimization.Just as with bugs and technical debt, performance issues are cheaper and less risky to address early on, rather than later in a project’s development cycle. Our focus is on helping to demystify profiling tools and techniques for those developers who are unfamiliar with them. That’s what the profiling e-book and its related blog post and webinar aim to support.Is there a way to exclude certain methods from instrumentation or include only specific methods when using Deep Profiling in the Unity Profiler? When using a lot of async/await tasks, we create large stack traces, but how can we avoid slowing down both the client and the Profiler when Deep Profiling?You can enable Allocation call stacks to see the full call stacks that lead to managed allocations (shown as magenta in the Unity CPU Profiler Timeline view). Additionally, you can – and should! – manually instrument long-running methods and processes by sprinkling ProfilerMarkers throughout your code. There’s currently no way to automatically enable Deep Profiling or disable profiling entirely in specific parts of your application. But manually adding ProfilerMarkers and enabling Allocation call stacks when required can help you dig down into problem areas without having to resort to Deep Profiling.As of Unity 2022.2, you can also use our IgnoredByDeepProfilerAttribute to prevent the Unity Profiler from capturing method calls. Just add the IgnoredByDeepProfiler attribute to classes, structures, and methods.Where can I find more information on Deep Profiling in Unity?Deep Profiling is covered in our Profiler documentation. Then there’s the most in-depth, single resource for profiling information, the Ultimate Guide to profiling Unity games e-book, which links to relevant documentation and other resources throughout.Is it correct that Deep Profiling is only useful for the Allocations Profiler and that it skews results so much that it’s not useful for finding hitches in the game?Deep Profiling can be used to find the specific causes of managed allocations, although Allocation call stacks can do the same thing with less overhead, overall. At the same time, Deep Profiling can be helpful for quickly investigating why one specific ProfilerMarker seems to be taking so long, as it’s more convenient to enable than to add numerous ProfilerMarkers to your scripts and rebuild your game. But yes, it does skew performance quite heavily and so shouldn’t be enabled for general profiling.Is VSync worth setting to every VBlank? My mobile game runs at a very low fps when it’s disabled.Mobile devices force VSync to be enabled at a driver/hardware level, so disabling it in Unity’s Quality settings shouldn’t make any difference on those platforms. We haven’t heard of a case where disabling VSync negatively affects performance. Try taking a profile capture with VSync enabled, along with another capture of the same scene but with VSync disabled. Then compare the captures using Profile Analyzer to try to understand why the performance is so different.How can you determine if the main thread is waiting for the GPU and not the other way around?This is covered in the Ultimate Guide to profiling Unity games. You can also get more information in the blog post, Detecting performance bottlenecks with Unity Frame Timing Manager.Generally speaking, the telltale sign is that the main thread waits for the Render thread while the Render thread waits for the GPU. The specific marker names will differ depending on your target platform and graphics API, but you should look out for markers with names such as “PresentFrame” or “WaitForPresent.”Is there a solid process for finding memory leaks in profiling?Use the Memory Profiler to compare memory snapshots and check for leaks. For example, you can take a snapshot in your main menu, enter your game and then quit, go back to the main menu, and take a second snapshot. Comparing these two will tell you whether any objects/allocations from the game are still hanging around in memory.Does it make sense to optimize and rewrite part of the code for the DOTS system, for mobile devices including VR/AR? Do you use this system in your projects?A number of game projects now make use of parts of the Data-Oriented Technology Stack (DOTS). Native Containers, the C# Job System, Mathematics, and the Burst compilerare all fully supported packages that you can use right away to write optimal, parallelized, high-performance C# (HPC#) code to improve your project’s CPU performance.A smaller number of projects are also using Entities and associated packages, such as the Hybrid Renderer, Unity Physics, and NetCode. However, at this time, the packages listed are experimental, and using them involves accepting a degree of technical risk. This risk derives from an API that is still evolving, missing or incomplete features, as well as the engineering learning curve required to understand Data-Oriented Design (DOD) to get the most out of Unity’s Entity Component System (ECS). Unity engineer Steve McGreal wrote a guide on DOTS best practices, which includes some DOD fundamentals and tips for improving ECS performance.How do you go about setting limits on SetPass calls or shader complexity? Can you even set limits beforehand?Rendering is a complex process and there is no practical way to set a hard limit on the maximum number of SetPass calls or a metric for shader complexity. Even on a fixed hardware platform, such as a single console, the limits will depend on what kind of scene you want to render, and what other work is happening on the CPU and GPU during a frame.That’s why the rule on when to profile is “early and often.” Teams tend to create a “vertical slice” demo early on during production – usually a short burst of gameplay developed to the level of visual fidelity intended for the final game. This is your first opportunity to profile rendering and figure out what optimizations and limits might be needed. The profiling process should be repeated every time a new area or other major piece of visual content is added.Here are additional resources for learning about performance optimization:BlogsOptimize your mobile game performance: Expert tips on graphics and assetsOptimize your mobile game performance: Expert tips on physics, UI, and audio settingsOptimize your mobile game performance: Expert tips on profiling, memory, and code architecture from Unity’s top engineersExpert tips on optimizing your game graphics for consolesProfiling in Unity 2021 LTS: What, when, and howHow-to pagesProfiling and debugging toolsHow to profile memory in UnityBest practices for profiling game performanceE-booksOptimize your console and PC game performanceOptimize your mobile game performanceUltimate guide to profiling Unity gamesLearn tutorialsProfiling CPU performance in Android builds with Android StudioProfiling applications – Made with UnityEven more advanced technical content is coming soon – but in the meantime, please feel free to suggest topics for us to cover on the forum and check out the full roundtable webinar recording.

AI could soon surpass Bitcoin mining in energy consumption, according to a new analysis that concludes artificial intelligence could use close to half of all the electricity consumed by data centers globally by the end of 2025.The estimates come from Alex de Vries-Gao, a PhD candidate at Vrije Universiteit Amsterdam Institute for Environmental Studies who has tracked cryptocurrencies’ electricity consumption and environmental impact in previous research and on his website Digiconomist. He published his latest commentary on AI’s growing electricity demand last week in the journal Joule. AI already accounts for up to a fifth of the electricity that data centers use, according to de Vries-Gao. It’s a tricky number to pin down without big tech companies sharing data specifically on how much energy their AI models consume. De Vries-Gao had to make projections based on the supply chain for specialized computer chips used for AI. He and other researchers trying to understand AI’s energy consumption have found, however, that its appetite is growing despite efficiency gains — and at a fast enough clip to warrant more scrutiny.“Oh boy, here we go.”With alternative cryptocurrencies to Bitcoin — namely Ethereum — moving to less energy-intensive technologies, de Vries-Gao says he figured he was about to hang up his hat. And then “ChatGPT happened,” he tells The Verge. “I was like, Oh boy, here we go. This is another usually energy-intensive technology, especially in extremely competitive markets.” There are a couple key parallels he sees. First is a mindset of “bigger is better.” “We see these big tech [companies] constantly boosting the size of their models, trying to have the very best model out there, but in the meanwhile, of course, also boosting the resource demands of those models,” he says. That chase has led to a boom in new data centers for AI, particularly in the US, where there are more data centers than in any other country. Energy companies plan to build out new gas-fired power plants and nuclear reactors to meet growing electricity demand from AI. Sudden spikes in electricity demand can stress power grids and derail efforts to switch to cleaner sources of energy, problems similarly posed by new crypto mines that are essentially like data centers used to validate blockchain transactions. The other parallel de Vries-Gao sees with his previous work on crypto mining is how hard it can be to suss out how much energy these technologies are actually using and their environmental impact. To be sure, many major tech companies developing AI tools have set climate goals and include their greenhouse gas emissions in annual sustainability reports. That’s how we know that both Google’s and Microsoft’s carbon footprints have grown in recent years as they focus on AI. But companies usually don’t break down the data to show what’s attributable to AI specifically.To figure this out, de Vries-Gao used what he calls a “triangulation” technique. He turned to publicly available device details, analyst estimates, and companies’ earnings calls to estimate hardware production for AI and how much energy that hardware will likely use. Taiwan Semiconductor Manufacturing Company (TSMC), which fabricates AI chips for other companies including Nvidia and AMD, saw its production capacity for packaged chips used for AI more than double between 2023 and 2024. After calculating how much specialized AI equipment can be produced, de Vries-Gao compared that to information about how much electricity these devices consume. Last year, they likely burned through as much electricity as de Vries-Gao’s home country of the Netherlands, he found. He expects that number to grow closer to a country as large as the UK by the end of 2025, with power demand for AI reaching 23GW. Last week, a separate report from consulting firm ICF forecast a 25 percent rise in electricity demand in the US by the end of the decade thanks in large part to AI, traditional data centers, and Bitcoin mining. It’s still really hard to make blanket predictions about AI’s energy consumption and the resulting environmental impact — a point laid out clearly in a deeply reported article published in MIT Technology Review last week with support from the Tarbell Center for AI Journalism. A person using AI tools to promote a fundraiser might create nearly twice as much carbon pollution if their queries were answered by data centers in West Virginia than in California, as an example. Energy intensity and emissions depend on a range of factors including the types of queries made, the size of the models answering those queries, and the share of renewables and fossil fuels on the local power grid feeding the data center. It’s a mystery that could be solved if tech companies were more transparentIt’s a mystery that could be solved if tech companies were more transparent about AI in their sustainability reporting. “The crazy amount of steps that you have to go through to be able to put any number at all on this, I think this is really absurd,” de Vries-Gao says. “It shouldn’t be this ridiculously hard. But sadly, it is.”Looking further into the future, there’s even more uncertainty when it comes to whether energy efficiency gains will eventually flatten out electricity demand. DeepSeek made a splash earlier this year when it said that its AI model could use a fraction of the electricity that Meta’s Llama 3.1 model does — raising questions about whether tech companies really need to be such energy hogs in order to make advances in AI. The question is whether they’ll prioritize building more efficient models and abandon the “bigger is better” approach of simply throwing more data and computing power at their AI ambitions. When Ethereum transitioned to a far more energy efficient strategy for validating transactions than Bitcoin mining, its electricity consumption suddenly dropped by 99.988 percent. Environmental advocates have pressured other blockchain networks to follow suit. But others — namely Bitcoin miners — are reluctant to abandon investments they’ve already made in existing hardware (nor give up other ideological arguments for sticking with old habits). There’s also the risk of Jevons paradox with AI, that more efficient models will still gobble up increasing amounts of electricity because people just start to use the technology more. Either way, it’ll be hard to manage the issue without measuring it first. See More: