As Russia continues its relentless assaults on Ukraine despite in defiance of continuing efforts to work towards a peace deal, multiple western security agencies have issued a new advisory warning of a Moscow-backed  campaign of cyber intrusions targeting logistics and technology organisations in the west. The campaign, run through Unit 26165 of the Main Directorate of the General Staff of the Armed Forces of the Russian Federation (GRU), better known as Fancy Bear, includes credential guessing, spear-phishing attacks, exploitation Microsoft Exchange and Roundcube vulnerabilities, and flaws in public-facing infrastructure including VPNs. This pattern of activity likely dates back to the early days of the war in February 2022 – at which point Fancy Bear was more heavily involved in cyber operations for purposes of espionage. However, as Russia failed to achieve its military objectives as quickly as it had wanted, the group expanded its targeting to include entities involved in the delivery of support and aid to Ukraine’s defence. Over the past three years its victims have included organisations involved in air traffic control, airports, defence, IT services, maritime and port systems sectors across various Nato countries. The advanced persistent threat (APT) actor is also understood to be targeting internet-connected cameras at Ukraine’s border crossings and around its military bases. These intrusions mostly took place in Ukraine but have also been observed in neighbouring states including Hungary, Poland, Romania and Slovakia. The GCHQ-run National Cyber Security Centre (NCSC) urged UK organisations to familiarise themselves with Unit 26165’s tactics and take action to safeguard themselves. “This malicious campaign by Russia’s military intelligence service presents a serious risk to targeted organisations, including those involved in the delivery of assistance to Ukraine,” said Paul Chichester, NCSC Director of Operations. “The UK and partners are committed to raising awareness of the tactics being deployed. We strongly encourage organisations to familiarise themselves with the threat and mitigation advice included in the advisory to help defend their networks.” The NCSC’s latest warning comes a couple of weeks after the cyber body’s CEO, Richard Horne, talked of a “direct connection” between Russian cyber attacks and physical threats to the UK at its annual conference. Horne told an audience at the CyberUK event that Russia was focusing on acts of sabotage, often involving criminal proxies. He said these threats, which are thought to have included arson attacks, are now manifesting on the streets of the UK, “putting lives, critical services and national security” at risk. Rafe Pilling, director of threat intelligence at the Sophos (formerly Secureworks) Counter Threat Unit (CTU) – which tracks Fancy Bear as Iron Twilight – said that the group's targeting of spear-phishing and vulnerability exploitation to gain access to target mailboxes had been a staple tactic for some time. “The focus of their operations pivots as the intelligence collection of the Russian military change and since 2022 Ukraine has been a significant focus of their attention. The targeting of Nato  and Ukranian defense and logistics companies involved in the support of the Ukrainian war effort makes a lot of sense in that context,” Pilling told Computer Weekly.   “The targeting of IP cameras for intelligence collection purposes is interesting and is a tactic generally associated with state-sponsored adversaries like Iron Twilight where they anticipate a physical effects aspect to their operations. As an intelligence provider to the Russian military this access would assist in the understanding of what goods were being transported, when, in what volumes and support kinetic targeting.   “We've seen other APT actors make use of compromised CCTV feeds to monitor the effects of cyber-physical attacks, for example the 2022 attacks against steel mills in Iran where video from the CCTV feed was used to time the execution of the attack in an attempt to avoid harm to people at the site and confirm the damage being caused,” he added. The NCSC said Britain’s support for Ukraine remained “steadfast”. Having already committed £13bn in military aid, the UK this week announced 100 new sanctions on Russia targeting entities and organisations involved in its energy, financial and military systems. This comes in the wake of the largest drone attack on Ukraine staged so far during the three-year war, which Russian dictator Vladimir Putin launched mere hours before a scheduled call with US president Donald Trump. The full advisory – which can be read here – sets out Fancy Bear’s tactics, techniques and procedures (TTPs) in its latest campaign in accordance with the Mitre ATT&CK framework, and also details a number of the common vulnerabilities and exposures (CVEs) being used to attain initial access. Besides the UK and US, the advisory is cosigned by cyber and national security agencies from Australia, Canada, Czechia, Denmark, Estonia, France, Germany, the Netherlands and Poland. Read more about Russian state cyber campaigns Russia is using phishing attacks to compromise encrypted Signal Messenger services used by targets in the Ukraine. Experts warn that other encrypted app users are at risk. The Russian cyber spy operation known as Star Blizzard changed tactics after a takedown operation by Microsoft and the US authorities, turning to widely used messaging platform WhatsApp to try to ensnare its targets. Computer Weekly talks to GCHQ’s National Cyber Security Centre operations director Paul Chichester and former NCSC chief executive Ciaran Martin on Russia, China and Salt Typhoon.

Metal additive manufacturing (AM) is expected to reach a market value of $13 billion by 2035, nearly tripling in size over the next decade, according to a new report from IDTechEx. The report, Metal Additive Manufacturing 2025–2035: Technologies, Players, and Market Outlook, outlines how metal AM is shifting from prototyping and tooling to end-use production in sectors such as aerospace, automotive, and general manufacturing. IDTechEx, an independent research firm focused on emerging technologies since 1999, identifies three key trends currently shaping the metal AM market: the dominance of laser powder bed fusion (LPBF), regional acceleration in China, and the economic uncertainty driven by tariff policies. Laser Powder Bed Fusion Retains Market Leadership Due to Versatility and Maturity Despite rising interest in alternative technologies like metal binder jetting (MBJ) and metal-polymer filament extrusion (MPFE), LPBF continues to dominate metal AM in both revenue and adoption. LPBF, which uses lasers to fuse metal powder into parts layer by layer, is the most commercially mature metal 3D printing method. It is backed by extensive operational experience from both equipment manufacturers and industrial users. MBJ and MPFE were expected to challenge LPBF due to their simplified workflows and potential cost advantages. However, IDTechEx reports that end-users are still grappling with technical and economic limitations in these newer methods. For example, MBJ is limited by sintering constraints, making it suitable only for specific geometries and part sizes, while MPFE is often confined to prototyping, jigs, and fixtures due to material and strength limitations. LPBF’s strength lies in its ability to scale across a wide range of applications. Depending on machine configuration—such as build volume or number of lasers—LPBF systems are used for producing small injection molds and large aerospace components alike. This range of use cases reinforces LPBF’s position as the leading technology in the metal AM sector. Breakdown of the 2024 global metal AM install base by process. Image via IDTechEx. Domestic Growth in China Redefines Global Market Share Metal AM in China is expanding rapidly, supported by a domestic ecosystem that prefers cost-competitive local manufacturers. Companies such as Xi’an Bright Laser Technologies (BLT), HBD, and EPlus3D have significantly increased their market presence over the past decade, supplying Chinese firms across sectors including aerospace and automotive. These manufacturers have grown without heavily relying on exports. While international attention to Chinese metal AM remained low until recent years, domestic demand has allowed firms like BLT to scale consistently. Double-digit annual growth rates have enabled several of these companies to open offices in Europe and North America, establishing new distribution partnerships and service networks. Price remains a key selling point. Metal AM systems require high upfront capital, and the affordability of Chinese-built machines is a primary factor for international customers evaluating suppliers. However, IDTechEx notes that escalating trade tensions and the introduction of new tariffs could complicate overseas expansion plans for Chinese firms. Even so, the scale of China’s internal market is expected to sustain ongoing development in the region. Tariff Pressures Generate Mixed Signals for Investment in Additive Manufacturing Tariff policies designed to promote domestic manufacturing in the United States are producing conflicting effects for the metal AM industry. On one side, tariffs have prompted renewed interest in reshoring and local production. This shift could favor metal AM technologies that enable decentralized manufacturing and rapid lead times. However, uncertainty surrounding tariffs has already triggered a pullback in capital spending across several industries. Companies are delaying investments in new manufacturing facilities, and many are scaling back research and development budgets. These developments affect the willingness of firms to adopt emerging production technologies like metal AM, particularly for applications that require substantial upfront validation and certification. IDTechEx reports that in the second half of 2024, economic uncertainty contributed to a slowdown in metal AM growth. The report suggests that this trend could continue depending on how trade and industrial policy evolve in 2025. Forecast from IDTechEx showing projected growth of the metal additive manufacturing hardware and materials market from 2025 to 2035, with a compound annual growth rate (CAGR) of 10.4%. Image via IDTechEx. Market Projections Segmented by Technology and Material Class The report provides detailed forecasts for metal AM hardware, materials, and installed base growth through 2035. Projections are broken down into ten metal printing processes and nine categories of metal materials, offering granular insight into market dynamics. Forecast data is supplemented with comparative benchmarks for each technology and application case studies. Material demand is also expected to rise as metal AM transitions to production-scale use. The report highlights increasing requirements for powder uniformity, process stability, and repeatable part quality—factors critical for certification in aerospace and other high-regulation sectors. Profiles of leading companies are based on direct interviews and market performance tracking. These include manufacturers of LPBF systems, providers of MBJ technologies, and developers of metal feedstocks. The analysis also outlines competitive strategies and regional positioning. Access to Full Research and Industry Insights The full Metal Additive Manufacturing 2025–2035 report is available at www.IDTechEx.com/MetalAM, where sample pages and additional data are also offered. IDTechEx’s broader 3D printing research portfolio includes coverage of polymers, ceramics, electronics, and large-scale applications such as construction and biomedical devices. To view the full library of additive manufacturing market studies, visit www.IDTechEx.com/Research/3D. Cover of Metal Additive Manufacturing 2025–2035: Technologies, Players, and Market Outlook. Image via IDTechEx. Ready to discover who won the 20243D Printing Industry Awards? Subscribe to the 3D Printing Industry newsletter to stay updated with the latest news and insights. Take the 3DPI Reader Survey — shape the future of AM reporting in under 5 minutes. Featured image shows the cover of Metal Additive Manufacturing 2025–2035. Image via IDTechEx. Anyer Tenorio Lara Anyer Tenorio Lara is an emerging tech journalist passionate about uncovering the latest advances in technology and innovation. With a sharp eye for detail and a talent for storytelling, Anyer has quickly made a name for himself in the tech community. Anyer's articles aim to make complex subjects accessible and engaging for a broad audience. In addition to his writing, Anyer enjoys participating in industry events and discussions, eager to learn and share knowledge in the dynamic world of technology.

Quantization is one of the key techniques for reducing the memory footprint of large language models (LLMs). It works by converting the data type of model parameters from higher-precision formats such as 32-bit floating point (FP32) or 16-bit floating point (FP16/BF16) to lower-precision integer formats, typically INT8 or INT4. For example, quantizing a model to 4-bit means each parameter uses only 0.5 bytes, compared to 4 bytes in FP32. Post-training quantization methods like GPTQ and AWQ can dramatically reduce the size of large models. A model like Llama 3 with 70 billion parameters can occupy around 140 GB in FP16, but this can be reduced to approximately 40 GB using 4-bit quantization, while still maintaining strong performance on downstream tasks. However, despite this substantial reduction, such models still exceed the memory capacity of most consumer-grade GPUs, which typically offer 24 GB to 32 GB of VRAM. To make these models truly accessible, quantization to even lower bitwidths, such as 2-bit, is required. While recent advances in low-bit quantization are promising, achieving stable and accurate 2-bit quantization remains a significant challenge. In this article, we review a technique called EoRA that helps compensate for quantization-induced errors. EoRA is a training-free method, meaning it can be applied quickly and efficiently to any model, even the largest ones. We’ll check how EoRA works and demonstrate how it can significantly improve the performance of 2-bit quantized models, bringing them close to the accuracy of their full-precision counterparts while being up to 5.5x smaller. We’ll analyze experimental results obtained using large models such as Qwen3-32B and Qwen2.5-72B, both quantized to 2-bit using state-of-the-art quantization techniques, to illustrate the effectiveness of EoRA. Diving into the Eigenspace in Search of an Adapter Post-training quantization or, more generally, compression aims to reduce model size or inference cost by minimizing the output difference between the original weights Wl​ and compressed weights Ŵl  using only a small calibration dataset. Most quantization methods are framed layer-wise, but the choice of compression formats is rigid and limits flexibility across diverse deployment needs. To bypass format constraints and improve accuracy, previous work, such as QLoRA [1] and HQQ+ [2], directly fine-tuned a Lora adapter on top of the frozen quantized models. It is also possible to reframe compression as a compensation problem: given a compressed model, introduce low-rank residual paths that specifically correct compression errors. A straightforward method uses SVD to decompose the compression error: \[\Delta W_l = W_l – \hat{W}_l\] into \[U_l \Sigma_l V_l^T\] forming low-rank approximations via two matrices: \[B_l = U_l \Sigma_l \] \[A_l = V_l^T\] where Al and Bl are the standard tensors of a LoRA adapter. However, plain SVD has two limitations: it does not minimize the original layerwise compression loss directly, and it allocates capacity uniformly across all error components, ignoring the varying importance of different parts of the model. To address this, NVIDIA proposes EoRA [3]. EoRA: Training-free Compensation for Compressed LLM with Eigenspace Low-Rank Approximation EoRA first projects the compression error into the eigenspace defined by the input activation covariance: \[\tilde{X} \tilde{X}^T\] where X̃ is the average activation over the calibration set. Then, by performing eigendecomposition, we get: \[\tilde{X} \tilde{X}^T = Q \Lambda Q^T\] The compression error ΔW is projected as: \[\Delta W’ = \Delta W Q’\] where Q′=QΛ. Then SVD is applied on ΔW′ to produce a low-rank approximation, and the result is projected back to the original space, adjusting the low-rank factors accordingly. This eigenspace projection changes the optimization objective: it weights the importance of different error components according to their contribution to the layerwise output (via eigenvalues), making the approximation more efficient. It can be computed quickly without any training, requires only calibration activations, and does not introduce extra inference latency. Moreover, the derivation shows that this approach leads to a direct minimization of the layerwise compression loss, not just the raw weight error. Analytically, truncating a singular value in the projected space corresponds to minimizing the true compression error under reasonable assumptions about the calibration activations. In their paper, NVIDIA presents a wide range of strong results showing that EoRA can significantly boost the accuracy of quantized models. However, their experiments focus mostly on older Quantization methods like GPTQ and are limited to mid-sized LLMs, up to 13B parameters, at 3-bit and 4-bit precisions. This leaves an open question: can EoRA still be effective for much larger models, using more modern quantization techniques, and even pushing down to 2-bit precision? Let’s find out. Calibrating an EoRA Adapter Suppose we have quantized models that show significantly degraded performance compared to their full-precision counterparts on certain tasks. Our goal is to reduce this performance gap using EoRA. For the experiments, I used Qwen2.5-72B Instruct and Qwen3-32B, both quantized to 2-bit using AutoRound (Apache 2.0 license), a state-of-the-art quantization algorithm developed by Intel. AutoRound leverages SignSGD optimization to fine-tune quantization, and is particularly effective for low-bit settings. All the models I made are available here (Apache 2.0 license): Quantized Qwen3 Quantized Qwen2.5 The 2-bit models were quantized with a group size of 32, except for which used a group size of 128. A larger group size reduces model size by storing less quantization metadata, but it introduces greater quantization error. I evaluated the models on IFEval, a benchmark that measures instruction-following capabilities. Results showed a noticeable drop in performance for the quantized versions. Image by the author To compensate for this degradation, I applied an EoRA adapter using the implementation provided in the GPTQModel library (licensed under Apache 2.0). The integration is straightforward. If you’re curious about how it’s implemented in PyTorch, the codebase is compact, clean, and easy to follow: GPTQModel’s EoRA implementation: eora.py EoRA requires a calibration dataset. Ideally, this dataset should reflect the model’s intended use case. However, since we don’t have a specific target task in this context and aim to preserve the model’s general capabilities, I used 1,024 randomly sampled examples from the C4 dataset (licensed under ODC-BY). Another key parameter is the LoRA rank, which greatly influences the effectiveness of the EoRA adapter. Its optimal value depends on the model architecture, the target task, and the calibration data. A higher rank may yield better performance but risks overfitting to the calibration set. It also increases the size of the adapter, counterproductive when the overall goal of quantization is to reduce memory usage. Conversely, a lower rank keeps the adapter lightweight but might not capture enough information to effectively compensate for quantization errors. In my experiments, I tested LoRA ranks of 32, 64, and 256. Below is the code used to create the EoRA adapter with GPTQModel: from gptqmodel import GPTQModel from gptqmodel.adapter.adapter import Lora from datasets import load_dataset calibration_dataset = load_dataset( "allenai/c4", data_files="en/c4-train.00001-of-01024.json.gz", split="train", download_mode="force_redownload" ).select(range(1024))["text"] eora_adapter_path = "Qwen3-32B-autoround-2bit-gptq-r256" model_path = "kaitchup/Qwen3-32B-autoround-2bit-gptq" eora = Lora( path=eora_adapter_path, rank=256, ) GPTQModel.adapter.generate( adapter=eora, model_id_or_path="Qwen/Qwen3-32B", quantized_model_id_or_path=model_path, calibration_dataset=calibration_dataset, calibration_dataset_concat_size=0, auto_gc=False) Using an NVIDIA A100 GPU on RunPod (referral link), it took approximately 4 hours to generate the EoRA adapter for the model Qwen3-32B-autoround-2bit-gptq. All EoRA adapters created for these models are publicly available (Apache 2.0 license): EoRA Adapters for Qwen2.5 and Qwen3 Evaluating EoRA Adapters for 2-bit LLMs Let’s evaluate the effect of the EoRA adapters. Do they improve the accuracy of the 2-bit models? Image by the author It works! The improvements are particularly notable for Qwen3-14B and Qwen3-32B. For instance, applying EoRA to Qwen3-32B, quantized to 2-bit with a group size of 128, resulted in an accuracy gain of nearly 7.5 points. Increasing the LoRA rank, from 32 to 64, also led to improvements, highlighting the impact of rank on performance. EoRA is also effective on larger models like Qwen2.5-72B, though the gains are more modest. Lower-rank adapters showed little to no benefit on this model; it wasn’t until I increased the rank to 256 that significant improvements began to appear. Memory Consumption of EoRA Using the EoRA adapter during inference results in the following increase in memory consumption: Image by the author The overhead is generally negligible. For instance for 2-bit Qwen3-14B, the adapters only add 257 MB and 514 MB to the total model size, with ranks of 32 and 64. With larger ranks, using an EoRA adapter becomes questionable as the total memory consumption may surpass the memory consumption of the same model quantized at a higher precision. For instance, 2-bit Qwen2.5 72B with an EoRA adapter of rank 256 is larger than 3-bit Qwen2.5 72B. Note: This estimate includes only the memory consumed by the adapter’s parameters. For completeness, we could also account for the memory used by adapter activations during inference. However, these are extremely small relative to other tensors (such as the model’s attention and MLP layers) and can safely be considered negligible. Conclusion EoRA works. We’ve confirmed that it’s a simple yet effective method for compensating quantization errors, even at 2-bit precision. It’s intuitive, training-free, and delivers meaningful performance gains. That said, there are a few trade-offs to consider: Rank search: Finding the optimal LoRA rank requires experimentation. It’s difficult to predict in advance whether a rank of 32 will be sufficient or whether a higher rank, like 256, will cause overfitting. The optimal value depends on the model, calibration data, and target task. Increased memory consumption: The goal of quantization is to reduce memory usage, often for highly constrained environments. While EoRA adapters are relatively lightweight at lower ranks, they do slightly increase memory consumption, particularly at higher ranks, reducing the overall efficiency of 2-bit quantization. Looking ahead, NVIDIA’s paper also demonstrates that EoRA adapters make excellent starting points for QLoRA fine-tuning. In other words, if you plan to fine-tune a 2-bit model using QLoRA, initializing from an EoRA-adapted model can lead to better results with less training effort. I’ve written about fine-tuning adapters for GPTQ model last year, in my newsletter: QLoRA with AutoRound: Cheaper and Better LLM Fine-tuning on Your GPU The main difference is that instead of initializing the adapter from scratch, we would load the EoRA adapter. This adapter will be fine-tuned. References [1] Dettmers et al, QLoRA: Efficient Finetuning of Quantized LLMs (2023), arXiv [2] Badri and Shaji, Towards 1-bit Machine Learning Models (2024), Mobius Labs’ Blog [3] Liu et al., EoRA: Training-free Compensation for Compressed LLM with Eigenspace Low-Rank Approximation (2024), arXiv The post Boost 2-Bit LLM Accuracy with EoRA appeared first on Towards Data Science.

Wizards of the Coast has been gradually rolling out what it's got in store for Magic: The Gathering and its big Final Fantasy crossover this summer. Over the weekend, the team debuted a significant chunk of cards, from both the main set and the Commander decks, and fans are enjoying some of the fun and even surprising cuts from their favorite Final Fantasy games.The Universes Beyond collaboration with Final Fantasy is getting both the usual Commander treatment and a booster-driven series, so there are a lot of cards to go over. The main set tinkers with the entirety of Final Fantasy's mainline, numbered games, running from Final Fantasy 1 to Final Fantasy 16. The Commander Decks, meanwhile, focus on four specific games: Final Fantasy 6, 7, 10, and 14, with their respective Commanders of Terra, Cloud, Tidus, and Y'shtola.Out June 13Starter KitOut June 13Play Booster Box (30 Packs)Out June 13BundleOut June 13Collector’s Edition Commander Deck Bundle$599.96 at AmazonOut June 13Commander Deck Bundle$279.96 at AmazonOut June 13Commander Deck 1 - Revival TranceOut June 13Commander Deck 2 - Limit BreakOut June 13Commander Deck 3 - Counter BlitzOut June 13Commander Deck 4 - Scions & SpellcraftOut June 13Gift BundleOut June 13Collector Booster BoxSome of the cards shown at last Saturday's Magic: The Gathering panel at PAX East captured the journey of individual characters throughout their games. Final Fantasy IV's Cecil, for example, has an entire mechanic around transforming from a Dark Knight into a Paladin. The summons of Final Fantasy utilize a "Saga Creature" set-up, to act as powerful but temporary forces on the field, melding Magic and Final Fantasy concepts in the process.Then you have cards like Sephiroth, Fabled SOLDIER, with artwork from Square's Tetsuya Nomura. The transformation along the card illustrates Sephiroth's various forms throughout the FF7 universe, from a young fighter into, eventually, Safer Sephiroth on the card Sephiroth, One-Winged Angel.And what would a Final Fantasy be without some Cids? The Final Fantasy set will include a Cid from each of the numbered Final Fantasy games, though Wizards is opting for the older Cid from Final Fantasy XV over his daughter, Cindy.After Saturday's panel, though, Wizards of the Coast released the decklist for the four aforementioned Commander decks, and fans started poring over the art and additions to see how the team adapted those four Final Fantasy games into Magic. Pretty well, it turns out.Reprints, or new versions of existing Magic cards, get some clever crossovers with Final Fantasy lore. Skullclamp, for instance, draws on the Jenova project in Final Fantasy 7 for its art.Together, Forever also takes an emotional stab at the Final Fantasy 10 fans out there. It shouldn't be surprising that Final Fantasy 10 has some emotional wellsprings to draw from, with its central protagonists' journey through a ruined land, on a somber pilgrimage. Several decks' cards reference some notable character moments, by the way, so you might want to go into these decklists aware that there will be some story spoilers for each respective game.But to lighten things up, one of Final Fantasy 10's more iconic, and frequently memed, moments also made the cut with Inspiring Call.In fact, Wizards found more than a few ways to incorporate Final Fantasy's less-serious moments into cards. Clever Concealment puts a spin on the Wall Market scene with Cloud, Aerith, and Tifa in dresses, heading to Don Corneo's, and even has a mechanic that seems fitting for the occassion. I'm a big fan of the Secret Rendezvous card, which has four different versions of the Golden Saucer date scene from Final Fantasy 7 with each potential date: Aerith, Tifa, Yuffie, and Barret.Even still, fans have been discovering the tiny details that show how much homage is being paid to the source material. The Final Fantasy 14 set, in particular, has quite a few references to the critically acclaimed MMORPG. The Hildibrand Manderville card mimics the infamous gentleman's habit of disappearing and returning, making for quite an interesting White/Black card in the process.One user on the FFXIV subreddit noted that the Archaeomancer's Map card uses a real possible location found in a Gaganaskin Treasure Map, for instance.In turn, Final Fantasy 14 players have been having fun with the cards. There are some jokes about wondering who the Thancred card represents, referencing the "This is Thancred" meme. And there's the alternative flavor text a user added to the Rite of Replication card, commemorating everyone's favorite speech from A Realm Reborn:Even the Tome of Legends card depicts a Wondrous Tails book, with all the little details you see in-game when a player picks up their Tails and, of course, several missed four-in-a-row connections.There's some clear attention to detail put into these cards, and it's certainly got me intrigued to see the full set. While the Commander games got a lot of love in these cards, a few reveals from Saturday's panel like Commune With Beavers have me eager to see how some of the other Final Fantasy games got adapated into the full Magic set.Magic: The Gathering's Final Fantasy set is out on June 13, 2025, with pre-release weekend one week prior.Eric is a freelance writer for IGN.