WWW.CGCHANNEL.COM
Protected: Review: HP Z4 Rack G5 rackmount workstation
html PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN" "http://www.w3.org/TR/REC-html40/loose.dtd"
HP’s Z4 Rack G5 is a graphics workstation designed to be housed in a server rack and accessed remotely, for studios or freelance artists who want to work more flexibly. Jason Lewis assesses how it stacks up against a conventional desktop system in a range of DCC benchmark tests.
Welcome to my first full review of 2025. It’s one I have been working on for quite some time: the HP Z4 Rack G5 workstation – a rackmount system in a data center 1U rack configuration.Reviewing a rackmount system is a first for me, and it took me some time to figure out how to tackle it. So before we get into the specs of the HP system on test, what exactly is a rackmount workstation, and why might you want to use one, particularly if you don’t work for a large studio?
Jump to another part of this review
Why choose a rackmount system?
System specifications
Testing procedure
Benchmark results
Power consumption
Price
Verdict
What is a rackmount system?
Rackmount systems are built to mount in tall vertical cages within data centers. Stacking rackmount units in this way makes it possible to fit more individual systems into the space available, while having all of the units together in a separate facility and accessing them remotely makes it easier to provide dedicated cooling and an uninterruptible power supply.
Rackmount units have standard dimensions, set out in the EIA-310 specification, with a width of 19 inches and a height measured in multiples of 1.75 inches or Rack Units (U). A typical full-size cabinet used in data centers stands over six feet tall, and can accommodate 42U of equipment.
The HP Z4 G5 is in a 1U configuration, so it fits into a rack space 1.75 inches high, but rackmount system can also be configured in 2U or 4U layouts and, less commonly, 3U layouts.
So does a rack unit actually need to be installed in a rack? As I had no first-hand experience with data center systems, this was a question I had to ask myself, but the answer is ‘no’.
While there is little reason to buy a rackmount system if you don’t intend to mount it in a rack, the cabinet itself is really just shelving. The units themselves will work lying on the ground or leaning up against a wall, so long as you don’t block any of the air vents, cooling being more critical than with a conventional desktop system, particularly for 1U units.
Why opt for a 1U configuration, rather than 2U or 4U? It comes down to power density. Rack cages are, for the most part, a fixed height, and floor space is at a premium in data centers, so the idea is to cram in as many systems as you can, taking into account your processing needs and power delivery capabilities.
If you aren’t using an external cooling units or network switches, a 42U cabinet can hold 42 x 1U rack units, which provides more processing power than 21 x 2U units or 10 x 4U units.
Larger units can accommodate more processors than could be adequately cooled in a 1U configuration, with higher core counts, but unless you need very powerful workstations, more smaller units is usually better.
Why choose a rackmount system?
The systems found in data centers are usually used as servers, or reserved for compute-only tasks, like rendering or AI training. So why would you use a rackmount system as a workstation?
One reason is security: according to HP engineers I talked to, a growing number of their clients want to put their workstations in the data center, where they are harder for unauthorised users to access, and have their artists remote into them securely.
Another is flexibility: with the workstation in the data center, the machine that an artist remotes in from can be a lower-power system, or even a laptop, saving space in the office, and promoting hybrid working, with artists able to work from the studio on some days, and remotely on others.
This also applies to freelancers, for whom setting up a rackmount system in the garage could help to keep their workspace tidy, and let them remote into their workstation while traveling.
Of course, for this to work, you need to be able to access the workstation remotely. HP offers its own software solution for this, HP Anyware, but for this review, I turned to a remote access solution that I was more familiar with, Parsec.
After setting up Parsec in Host mode on the HP Z4 Rack system, and Client mode on a laptop on my local network, I was able to work with no perceptible lag and very minimal compression artifacts in the H.264/H.265 video stream sent by the host system.
It’s also worth noting that Parsec works well over the internet, and the required bandwidth is actually quite low, so you don’t need the fastest connection possible. In my experience, latency is the more important factor, and a ping of 80 ms or less is sufficient for a seamless experience.
System specifications
Next, let’s look at the specs of the particular HP Z4 Rack G5 system on test.
At its heart lies a 16-core Intel Xeon W5-2465X workstation processor running at a 3.1 GHz base frequency and a 4.7 GHz boost frequency, and supporting up to 64 PCIe 5.0 lanes.
It sits on a custom HP Socket LGA-4677 mainboard with 128 GB of DDR5 4800 memory in a 4 x 32 GB layout. For storage, it has two 1TB Samsung 990 Pro PCIe 4.0 NVMe drives.
Lastly, the system sports a 48GB NVIDIA RTX 6000 Ada Generation GPU: it originally shipped with a previous-generation Ampere-based RTX A6000 due to a shortage of Ada Generation hardware at the time, but since the Ampere GPU architecture is getting long in the tooth, I decided that the RTX 6000 would be more representative of a typical current build.
Testing procedure
Since I don’t have another system that falls into a similar performance category to the Z4 Rack G5, as a point of comparison for this review, I also ran the benchmark tests on the trusty AMD Ryzen Threadripper 3990X system I have used for my recent GPU group tests.
Its CPU is now two generations old, but it has similar base and boost clock speeds to the Z4 Rack G5, and four times the number of cores. The Threadripper system uses DDR4 memory, not the newer DDR5 in the Z4 Rack G5, and only supports PCIe 4.0 – although since the Z4 Rack G5’s storage drives and GPU are both PCIe 4.0 devices, that may make little difference.
You can see a side-by-side comparison of the key specs of the two systems above.
For testing, I used the following applications:
Viewport performance
3ds Max 2025, Blender 4.3, Maya 2025, Solidworks 2023Rendering
Arnold 7.3 for Maya 2025, Blender 4.3 (Cycles renderer), Corona 12 for 3ds Max 2025, V-Ray 7 for 3ds Max 2025Real-time rendering
Chaos Vantage, Unreal Engine 4.27 and 5.5Other benchmarks
Houdini 20.4, Metashape Pro 1.8.5, Premiere Pro 2024, World Machine Build 4031Synthetic benchmarks
3DMark (Speed Way and Port Royal benchmarks), Cinebench 2024, OctaneBench 2020.1, V-Ray 6 benchmarkIn the viewport benchmarks, the frame rate scores represent the figures attained when manipulating the 3D assets shown, averaged over three to five testing sessions.
All testing was done on a 32-inch 4K display running at its native resolution of 3,840 x 2,160px.
Benchmark results
Viewport performance
The viewport benchmarks reflect interactive performance in three popular DCC applications – 3ds Max, Blender and Maya – and one CAD application, SolidWorks.
The Z4 Rack G5 beats the Threadripper 3990X system in all four tests. It’s interesting that the GPU isn’t the key factor here: the GeForce RTX 4090 in the Threadripper system is technically faster than the RTX 6000 Ada Generation in the Z4 Rack G5, with higher clock speeds and more memory bandwidth, and none of the test scenes exceed its 24GB frame buffer.
Instead, the CPU determines the result. The performance of DCC applications typically tops out at four to eight CPU cores, so the Threadripper system’s extra cores are of no benefit, and although both systems have similar clock speeds, the Z4 Rack G5’s Xeon CPU is much newer and has much better performance per core.
Rendering
The first set of rendering tests cover offline renderers: Arnold and Cycles, commonly used for VFX work; Corona, used for architectural visualization; and V-Ray, which is used for both.
Here, the results are perfectly divided. When rendering on the CPU, the Threadripper 3990X system wins out, thanks to its massive advantage in CPU core count; on the GPU, the Z4 Rack G5 wins out.
Real-time rendering
The second set of rendering tests cover real-time renderers: Unreal Engine, used for virtual production as well as games; and Vantage, Chaos’s renderer for exploring large V-Ray scenes.
The Unreal Engine tests yield some very interesting results. In theory, they should follow a similar pattern to the viewport benchmarks, but the older Threadripper system wins in two of the tests, while the Z4 Rack G5 wins in one.
As a games card, the GeForce RTX 4090 in the Threadripper system may be a better fit for the Unreal Editor, but if that were the only factor, the Z4 Rack G5, with its RTX 6000 Ada Generation GPU, would not have won the Valley of the Ancients test, let alone by such a large margin. For all of its advanced features, since version 5.0, Unreal Engine has exhibited some inconsistent and unpredictable behavior when it comes to performance across different hardware.
With the Chaos Vantage Scene, the Z4 Rack takes a commanding lead over the Threadripper system. Again, the CPU is the deciding factor: the test scene fits inside the 24GB frame buffer on the Threadripper system’s GeForce RTX 4090, so GPU memory is not an issue.
Other benchmarks
The next benchmarks covers a range of tasks: fluid simulation in Houdini, photogrammetry in Metashape, video encoding in Premiere Pro, and terrain generation in World Machine.
The miscellaneous benchmarks give a range of results, some expected, and some unexpected.
The fact that the Threadripper system wins in the Houdini fluid simulation is no surprise: Houdini’s FLIP solver is fully multi-threaded, so despite the age difference, you would expect its 64-core CPU to beat the 16-core CPU in the Z4 Rack G5. But with World Machine, also multi-threaded, the Z4 Rack G5 takes a decent lead over the Threadripper system.
The Premiere Pro and Metashape tests give more predictable results. Both apps are multi-threaded, but can use far fewer CPU cores than rendering and simulations can, so the newer, faster Xeon CPU in the Z4 Rack G5 wins out. Although both apps do make use of the GPU, GPU usage caps out at 20-30% during the tests, so it isn’t the major factor.
Synthetic benchmarks
Finally, we have an assortment of synthetic benchmarks. They don’t accurately predict how a GPU will perform in production, but they’re a decent measure of its performance relative to other GPUs, and the scores can be compared to those available online for other cards.
Again, the synthetic benchmarks give mixed results, with the Z4 Rack G5 winning out in the Cinebench GPU and single-core CPU tests, and the V-Ray GPU benchmark, while the Threadripper system wins the Cinebench multi-core CPU test, the V-Ray CPU benchmark, the OctaneBench test, and the 3Dmark Speed Way and Port Royal tests.
Power consumption
Since the Z4 Rack G5 is primarily a data center system, power consumption – particularly current drawn – is less of a concern than it would be for a home or small office system, as data centers are designed with power supplies adequate for their operating needs.
My figures are for the entire the power or current drawn by the entire system, measured with a Kill-A-Watt P3 meter at the wall outlet while performing a variety of DCC tasks.
As you can see, the Z4 Rack G5 uses substantially less power than the Threadripper system, as is to be expected of a machine with a quarter of the CPU cores and a slightly less power-hungry GPU. But all in all, these are pretty decent numbers.
Acoustic performance
Rackmount systems have a reputation for being noisy. This is especially true for 1U systems, since they require many small, high-speed fans to keep the internal components cool. 2U, 3U and 4U systems have more space for air flow, but generally speaking, still create more fan noise than their desktop counterparts. This is rarely an issue in data centers, which are very noisy places, to the extent that staff often wear hearing protection, but more so in an office space.
The HP Z4 Rack G5 was no exception, but surprisingly, it was only during periods of high CPU workload that the fans would spin up enough to make the system unpleasantly loud. During light-to-medium CPU workloads, it was much quieter that I had expected, and even when the GPU was under heavy load, noise really wasn’t much of a problem.
Price
Before we get to the overall verdict, some thoughts on cost.
For medium-to-large businesses operating compute clusters, hardware budgets can run to hundreds of thousands, or even millions, of dollars, which makes the cost of a rackmount system relatively trivial. But since the Z4 Rack G5 is specifically marketed as a graphics workstation, I am going to look at it in the context of similarly powerful desktop workstations.
As equipped, with the NVIDIA RTX 6000 Ada Generation GPU that I swapped in, the Z4 Rack G5 system on test comes in at just a tick under $20,000.
This is pretty reasonable for a data center product, but compared to a similarly powerful desktop system, it is expensive. For comparison, I priced out an equivalent current desktop workstation with a 16-core AMD Threadripper Pro 7955WX CPU, 128GB of RAM, a NVIDIA GeForce RTX 5090, 2 x 4TB PCIe 5.0 M.2 SSDs and a 1300 W Seasonic PSU, and that system came in at around $6,000.
In other words, for around a third of the cost of the Z4 Rack G5, you can get a desktop system with a faster GPU. If you really need the RTX 6000 Ada Generation GPU, with its 48 GB of VRAM, that still only raises the cost to $12,000.
The cost of the Z4 Rack G5, and that of the support contracts that usually come with data center systems, is well within the norm for these types of products. However, if budget is a concern, going with a desktop system will give you slightly better performance for less money.
Verdict
So what conclusions can we draw from these tests?
First off, the HP Z4 Rack G5 Workstation is a very capable system. If you ignore the generational gap between it and our Threadripper comparison system, it beats out a workstation with four times the number of CPU cores.
Secondly, despite being a data center product, designed to be used via a remote login workflow, it feels remarkably like using a local system. Using it with Parsec was nearly seamless.
The build out will cost more than a traditional desktop-based solution, but if you need your workstations to be located within a data center for logistical or security reasons, the HP Z4 Rack G5 will serve you quite well.
Thanks for taking the time to stop by and check out this review. If you have questions, you can contact me here.
Links
Read more about the Z4 Rack G5 workstation on the HP website
About the reviewer
Jason Lewis is Senior Hard Surface Artist at Lightspeed LA. You can see more of his work in his ArtStation gallery. Contact Jason at jason [at] cgchannel [dot] com.
Acknowledgements
I would like to give special thanks to the following people for their assistance in bringing you this review:
Have your say on this story by following CG Channel on Facebook, Instagram and X (formerly Twitter). As well as being able to comment on stories, followers of our social media accounts can see videos we don’t post on the site itself, including making-ofs for the latest VFX movies, animations, games cinematics and motion graphics projects.
English
Arabic
French
Spanish
Portuguese
Deutsch
Turkish
Dutch
Italiano
Russian
Romaian
Portuguese (Brazil)
Greek
