Which workstation-class GPU is best for your DCC work? Jason Lewis reviews the current Quadro RTX cards from Nvidia and three of the Radeon Pro GPUs from AMD using a series of real benchmarks.
Back in February, I checked Nvidia's GeForce RTX 2080 Ti and compared it to several other high-end consumer GPUs from the company: both the current cards and the previous generation cards that are still widely used in production.
This time, I'll be testing the cards against the same benchmarks as I did in February, but looking at many of the professional workstation GPUs currently on the market, including Nvidia's entire Quadro RTX lineup – the Quadro RTX 8000, 6000, 5000, and 4000 – and three of the AMD Radeon Pro cards: the current generation Radeon Pro W5700 and W5500 and the previous generation Radeon Pro WX8200.
Jump to another part of this review
Technology focus: GPU architectures, memory types and APIs
Technology focus: consumer vs. Workstation cards
Technical data and prices
Technology focus: GPU architectures, memory types and APIs
Before we discuss the cards to test, let's go over some of the technical terms that will appear throughout this review. If you are already familiar with them, you can move on.
Nvidia's current Turing GPU architecture offers three types of processor cores: CUDA cores for general GPU computing; Machine learning tensor cores; and RT cores, new to the Turing-based RTX cards and designed for ray tracing calculations.
Now that the Turing architecture is nearly two years old, we're seeing widespread support for hardware-accelerated ray tracing in CG software, with offline renderers like Arnold, KeyShot, V-Ray, and Blender's Cycles Engine over OptiX from Nvidia Access the RT cores API and game engines like Unity and Unreal Engine do this via DXR, DirectX 12's ray tracing API.
It's important to note that comparing core counts is a reasonable way to gauge the likely relative performance of different Nvidia cards, but not to compare Nvidia cards to AMD's that are of a different architecture, and instead To group processing cores into "arithmetic units".
The two companies also differ in the APIs they support for GPU computing: while both AMD and Nvidia support the open standard OpenCL, Nvidia focuses on its own CUDA and OptiX APIs. OpenCL has the advantage that it is hardware independent, but CUDA and OptiX are currently more strongly supported in DCC applications. OpenCL 3.0, the specs released in April, may change that, but there is still a long way to go.
Technology focus: consumer vs. Workstation cards
One of the main differences between the workstation cards we tested and their cheaper consumer counterparts is the amount of on-board memory they have. Professional GPUs typically have much more memory than consumer variants based on similar parts. Although they all use the TU102 processor, the Quadro RTX 8000 and RTX 6000 have 48GB and 24GB of graphics memory, respectively, while the consumer Quadro RTX 2080 Ti has 11GB.
That's not always the case – the Quadro RTX 4000 has the same 8GB of storage as its consumer counterpart, the GeForce RTX 2070 – but since workstation cards are geared towards memory-intensive tasks like GPU rendering, the added capacity is meaningful helps justify the additional costs.
Another important difference between workstation GPUs and their consumer counterparts is their drivers. Workstation card drivers are optimized for graphics applications, including Adobe's Creative Cloud tools and popular DCC and CAD packages such as 3ds Max, Maya, AutoCAD, and SolidWorks. These optimizations provide greater stability and, in some cases, higher performance when compared to their desktop counterparts.
Nvidia and AMD also offer productivity tools for use with their professional GPUs. In addition to dedicated desktop and display management tools, both Nvidia's Quadro Experience and AMD's Radeon Pro Software Suite offer 4K screen recording, and Radeon Pro software also offers remote desktop access.
In addition, manufacturers offer more extensive customer support for professional cards: in many cases, workstation cards are the only ones officially certified for use with DCC or CAD software. Professional cards are essentially aimed at users who need not only speed, but also stability and support. A consumer card may offer reasonable performance. In times of crisis when you need both reliability and performance, a professional GPU is more secure.
Technical data and prices
Prices were correct at the time of writing.
As with the previous review, my test system was an APEXX T3 workstation from BOXX Technologies:
Central processor: AMD Threadripper 2990WX (32 cores, 3.0 GHz base clock rate, 4.2 GHz boost clock)
Motherboard: ASRock X399 Taichi
R.A.M.: 64 GB 2,666 MHz Samsung DDR4 ECC RAM
camp: 512 GB Samsung 970 Pro NVMe SSD
power adapter: 1000 W seasonal gold
CPU cooler: Cooler Master 240mm AIO liquid cooler
operating system: Windows 10 Pro for workstations
I used the following applications for testing:
Viewport and editing performance
3ds Max 2020, Blender 2.82, Maya 2020, Modo 13.0v1, SolidWorks 2020, Substance Painter 2019, Unreal Engine 4.24
Blender 2.82 (with Cycles), Cinema 4D R21 (with Radeon ProRender), KeyShot 9, Maverick Studio Build 410, OctaneRender 2019 (with OctaneBench), Redshift 3.0.22 for 3ds Max, SolidWorks 2020 Visualize, V-Ray Next (V- Ray 4.3 for 3ds Max (with V-Ray GPU), Radeon ProRender 2.7 for 3ds Max
Metashape 1.5.1, Premiere Pro CC 2019, Substance Alchemist 2019
In the viewport and editing benchmarks, the frame rate values represent the numbers achieved while editing the displayed 3D elements, averaged over five test sessions to eliminate inconsistencies. The CPU was deactivated in all rendering benchmarks so that only the GPU was used for the calculation. The tests were performed on a single 32-inch 4K display with a native resolution of 3840 x 2160 pixels at 60 Hz.
Viewport and editing performance
Viewport benchmarks include many of the most important DCC applications – both universal 3D software like 3ds Max, Blender, and Maya, and specialty tools like Substance Painter – as well as the SolidWorks CAD package and the Unreal Engine game engine.
When comparing the Nvidia GPUs, all tests gave the same results: The Quadro RTX 8000 took first place, the RTX 6000 second, the RTX 5000 third and the RTX 4000 fourth.
The pattern was also consistent with the AMD GPUs: the Radeon Pro W5700 almost always took first place, with the WX 8200 usually taking second and the W5500 third, although their places were temporarily reversed.
When looking at the cards from both manufacturers together, the AMD Radeon Pro W5500 and WX 8200 usually came behind the Nvidia cards. The AMD Radeon Pro W5700 ran neck to neck with the Nvidia Quadro RTX 4000, but in some cases moved closer to the RTX 5000 and was on the heels of the RTX 6000 and RTX 8000 with 3ds Max.
Next we look at the GPU rendering: via OpenCL for the AMD cards and via CUDA and OptiX for the Nvidia cards. In applications where I was able to turn off OptiX support, I added a second set of numbers for rendering with CUDA only to show the effect of the RT cores in the RTX cards on performance.
When comparing the Nvidia GPUs with one another, the pattern corresponds to the benchmarks of the view window and corresponds to the positions of the cards in the Quadro RTX product range: the RTX 8000 comes first, the RTX 6000 second, the RTX 5000 third and the RTX 4000 fourth.
Since most of the benchmark applications are just CUDA or OptiX, the options for testing the AMD GPUs were much more limited, and the results varied from test to test: for AMD cards, the Radeon Pro W5700 was always the best W5500 and WX 8200 swapped places.
In the benchmarks where I was able to use both Nvidia and AMD GPUs, the AMD cards didn't do well. In all cases, they fell significantly behind the Nvidia GPUs.
GPU rendering: high memory scenes
I also ran some tests on very large scenes to see the impact of GPU memory capacity on performance. Each was run twice for each GPU. The first time I tested it, I only used the GPU, so it had to do both rendering and desktop viewing tasks. The second time, I added a Quadro RTX 5000 to the system as a dedicated graphics card so that all of the RAM of the GPU under test can be used for rendering. None of these benchmark scenes are set up for OpenCL renderers, so only Nvidia cards are used in these tests.
If you run a single GPU in your system, the available memory is critical: Neither the Quadro RTX 4000 nor the RTX 5000 can render scenes and control the desktop display at the same time. Only the Quadro RTX 6000 and RTX 8000 with their 24 GB and 48 GB graphics RAM were up to the task.
If a dedicated display GPU is added, things are very different: all GPUs rendered two of the scenes, and all but the 8GB Quadro RTX 4000 rendered the visualization scene of the V-Ray architecture.
The next benchmarks serve to test the use of the GPU for more specialized tasks. The Metashape photogrammetry application uses the GPU to support the CPU in image processing, point cloud and mesh generation. This time, I ran the builds with the High settings rather than the Medium settings that I used in the February test. Substance Alchemist uses the GPU for material processing and to run certain filters and effects. Adobe Premiere Pro uses the GPU to assist the CPU with video encoding.
The results here are very mixed. With Substance Alchemist, the Nvida GPUs take a dominant lead over the AMD GPUs in the material production test. Since Alchemist's AI-based De-Lighting filter requires the tensor cores for machine learning in Nvidia cards, none of the AMD GPUs were able to complete the De-light test.
In contrast, Metashape seems to prefer AMD GPUs as they massively outperform Nvidia cards. In the Premiere Pro test, the AMD Radeon Pro WX 8200, which came last in several other benchmarks, took first place, followed by the Nvidia GPUs and the two other AMD cards.
Finally, we have the synthetic benchmark 3DMark. Synthetic benchmarks are a cornerstone of many online reviews, but they do not predict how a GPU will perform with real accuracy in production because they are often based on best-case usage scenarios. However, they can be a useful point of comparison with GPUs that are not included in a group test, as results for a wide variety of cards are available online.
The 3DMark results are as expected, with the Nvidia Quadro RTX 8000 coming first, followed by the RTX 6000, and then the RTX 5000. The Radeon Pro W5700 is the only AMD GPU that is among the Nvidia cards, followed by the Quadro RTX 4000, then the Radeon Pro WX 8200 and the Radeon Pro W5500.
Benchmark results: Quadro RTX 8000 versus Quadro RTX 6000
If you look at the benchmarks as a whole, it's strange that the Quadro RTX 8000 consistently outperforms the RTX 6000 as both cards use identical processors and have identical boost clock speeds. The only real difference between them is that the RTX 8000 has 48GB of RAM and the RTX 6000 has 24GB. So if a test scene is using less than 24GB of RAM, in theory they should perform the same. I suspect that either the RTX 8000 is increasing its clock speed a little higher than advertised or that there are additional optimizations in the AMD drivers or in the DCC applications themselves.
Before I come to judgment, I would like to address a few other issues. For anyone wondering how these cards would work in multi-GPU setups, I looked at this with a pair of Quadro RTX 5000 GPUs in November 2019. See the original group test for full details, but to sum it up: Most DCC applications support multi-GPU setups, and for GPU computing, performance scaling is usually good to excellent. However, using an SLI setup offers little to no benefit in viewport performance.
One interesting aspect of this generation of Quadro GPUs is that they all use traditional fan coolers with one fan, while their GeForce counterparts use a two-axis cooling concept. This is also the case when comparing AMD's Radeon Pro GPUs to consumer Radeon cards, although the difference is less noticeable as many of AMD's board partners use their own cooler designs.
This may seem counter-intuitive as it is believed that axial designs with multiple fans will have better cooling performance. However, professional cards are more likely to be used in multi-GPU configurations, and axial designs are not ideal for this because they release hot air inside a workstation case and potentially cause the internal temperature to rise significantly. Fan coolers let hot air out of the back of the card and act like exhaust fans for the workstation.
In most of these benchmarks, the Nvidia GPUs outperform the AMD cards. Additionally, the AMD GPUs just couldn't take a lot of rendering tests because the applications use proprietary Nvidia APIs.
I don't think this is a bad AMD hardware design, but rather a software support issue. AMD only supports the OpenCL open standard for GPU computing, while Nvidia has gone to great lengths to develop CUDA and, more recently, OptiX. According to several software developers I spoke to, both are now faster and more robust APIs and therefore have become more widely used in DCC applications. I don't see any change anytime soon unless OpenCL 3.0 offers a significant leap forward.
AMD has tried to alleviate the lack of support for OpenCL in the GPU rendering market by providing its own renderon engine, Radeon ProRender, which is built into several DCC applications and available as a free plug-in to others. While I'm quite fond of ProRender, I'm less than optimistic that there is real market share. Applications like Arnold and V-Ray are so well established that it is very difficult for new renderers to hit the market, and when they do, it takes time. In my opinion, in the future, AMD would best develop its own API to compete with CUDA and OptiX and then try to bring software developers on board.
While Nvidia GPUs generally offer better GPU processing power, one thing AMD GPUs do a little better is compute multitasking. The Radeon Pro GPUs keep desktop viewing tasks as a high priority even when the GPU is under full load to ensure system reactivity.
For example, when I was rendering a scene in 3ds Max using a V-Ray GPU but was working in a separate Maya scene, the viewport stuttered significantly on both the Quadro RTX 4000 and RTX 5000 and there were delays when switching between applications. The Quadro RTX 6000 and RTX 8000 were not affected. Using EVGA's Precision X1 utility, I measured total memory usage during testing on both the RTX 6000 and RTX 8000 at around 19GB: higher than the RTX 4000 or RTX 5000, which suggests that memory capacity may be the problem . In contrast, all three AMD GPUs performed well when switching from the V-Ray GPU to the Radeon ProRender, even though they only had 8GB of built-in RAM. The system remained fast and responsive, only odd hiccups here and there.
Finally, let's talk about the price. I'm trying not to overemphasize this with professional GPUs: artists who are used to consumer hardware often fall off their chairs when they see the prices of workstation cards, but in production, the benefits of professional cards warrant it often the cost. Let's look at the actual prices. As you can see from the graphic below, the introductory prices for the Nvidia cards have mainly increased slightly between the previous and current Quadros generation. Introductory prices for AMD's Radeon Pro cards stayed the same or decreased slightly.
The takeaway message
While both manufacturers offer good options for anyone in need of a workstation GPU, based on these benchmarks, I'd recommend the Nvidia cards over the AMD cards we tested, unless price is your primary consideration. While the AMD GPUs take the lead over their Nvidia counterparts in some benchmarks, the Quadros generally outperformed the Radeon Pro cards.
Among the DCC applications, CUDA and OptiX currently have a lot more support than OpenCL, and while this is good, Radeon ProRender isn't going to steal people from CUDA-based renderers like Arnold, Redshift, and V-Ray in the near future. With the built-in hardware for machine learning and ray tracing, the Quadro RTX cards are more versatile in my opinion.
One thing I've heard many times is that the Radeon Pro GPUs are performing well in the CAD market. I'm a DCC artist, not a CAD guy, so I haven't tested any design applications other than SolidWorks, and even there I know just enough to do some basic benchmarking with them. If you are looking for a professional grade GPU to run CAD applications, you may need to dig deeper yourself.
But once you've decided on a Quadro RTX card, which one should you buy? That depends on how complex the scenes you are working with and what budget you have. All four Quadro RTX GPUs tested here do the same tasks, but the high-end models do them faster and their larger storage capacity allows more complexity in your projects.
However, for serious current production work, I'd say the Quadro RTX 5000, with its 16GB of RAM, should be the minimum pick. The RTX 6000 with 24 GB of RAM is just the thing in most cases, while the RTX 8000 with 48 GB of RAM can be reserved for the GPU to render extremely complex scenes with many high-resolution textures or with resolutions of 8 KB and more .
Lastly, I just wanted to say thank you for sharing your time with us. My goal is to answer hardware questions specifically related to digital art and content creation, and I hope this review was helpful and informative.
For more information on the Radeon Pro GPUs, visit the AMD website
For more information on the Quadro RTX GPUs, visit the Nvidia website
About the reviewer
Jason Lewis is a Senior Environment Artist at Obsidian Entertainment and a regular reviewer for Greenscreen. You can see more of his work in his ArtStation gallery. Contact him at jason (at) cgchannel (dot) com
I would like to express my special thanks to the following people for their assistance in creating this review:
Shannon McPhee from Nvidia
Kasia Johnston from Nvidia
Emily Pallack from Edelman
Cole Hagedorn from Edelman
Bruno Oliveira doing the blend swap
Stephen G Wells
Tags: 2020, 3DMark, 3ds Max, Adobe, AMD, Autodesk, Benchmark, best professional GPU, best Quadro RTX GPU, best Radeon Pro GPU, best workstation GPU, Blender, CAD, Chaos Group, Cinema 4D, clock rate, Computing power, processing units, CUDA, CUDA cores, cycles, DCC, FP16, FP32, Game Engine, GCN, GDDR6, GPU, GPU computing, GPU rendering, graphics card, group test, HMB2, Jason Lewis, KeyShot, starting price, Maverick Render , Maverick Studio, Maxon, memory, memory bandwidth, Metashape, Navi, NVIDIA, NVLink, OctaneRender, OpenCL, OptiX, Otoy, photogrammetry, Polaris, power consumption, Premiere Pro, price, processor, production test, professional GPU, Quadro RTX, Quadro RTX 4000 , Quadro RTX 5000, Quadro RTX 6000, Quadro RTX 8000, Radeon Pro, Radeon Pro W5500, Radeon Pro W5700, Radeon Pro WX 8200, Radeon ProRender, RAM, real-time, redshift, rendering, review, RNDA, round-up, RT- Cores, SLI, Solidworks, specifications, stream processors, Substance Alchemist, synthetic benchmark, TDP, tensor cores, test, Tur V-Ray, video encoding, viewport performance, visualization, workstation GPU