Date: November 19, 2019
Author(s): Rob Williams
AMD has augmented its Radeon Navi lineup with its first workstation card: the Radeon Pro W5700. This 8GB card shares traits with the RX 5700, but with optimizations in place, it can sometimes exceed expectations. We’re pitting AMD’s latest against NVIDIA’s Quadro RTX 4000, and a handful of others.
Well, well, well… what do we have here? After months of wondering whether or not the Radeon Pro series would be treated to a Navi GPU, the 8GB W5700 hit our doorstep, begging to be benchmarked. As the name might suggest, the W5700 is based on the same GPU powering the RX 5700. Notably, we didn’t mention “XT” there, which leaves room for AMD to release a follow-up (perhaps with 16GB of memory?)
It might seem obvious that Navi was going to wind up as a Radeon Pro card, but messaging around the Navi launch could have made it seem like Vega was going to be the only way forward. As far as we understand it, Vega is going to be focused more on enterprise work, whereas Navi will take over the reigns for RPro.
We believe that in some cases, Vega is going to outperform Navi, but that’s mostly in workloads that the Radeon Pro series doesn’t necessarily target, like compute. In our launch news post, we mentioned that AMD’s internal testing showed the W5700 to beat out the WX 8200 in a handful of workloads, despite that card’s use of both Vega and HBM2 memory.
We continue to have the worst luck in the world, as a couple of months ago, our own WX 8200 decided to stop working, so we won’t have that comparison here. We will however have the even more relevant comparison: Quadro RTX 4000. In addition to those, we’ll also compare against the Radeon Pro WX 7100, WX 5100, RX 5700 XT, Quadro P2000, and GeForce RTX 2060 SUPER. We hoped to include more to paint a fuller picture, but CPUs have been dominating our benchmarking time.
Over its gaming counterpart, the Radeon Pro W5700 delivers optimized drivers for workloads that are important to designers. In some cases, this improves performance, while in all cases, the goal is to deliver unparalleled stability. In the case of RPro, 10-bit OpenGL support is offered for all applications, which contrasts to regular Radeon, which holds it back. We hope that AMD ends up enabling full 10-bit support on all of its GPUs at some point, as NVIDIA has done with its own, so that we don’t have to think about whether someone has it or not.
The W5700 requires a 6- and 8-pin power connector, and for video, there are five mini-DisplayPorts, and also USB-C. We would have loved to have seen regular DisplayPorts here, since adapters / dongles are a chore – but that’s only really for us, who swap components all of the time. AMD kindly includes two mini-DP to DP adapters, and one DVI to mini-DP adapter in the box.
|AMD’s Radeon Pro Workstation GPU Lineup|
|Cores||Base MHz||Peak FP32||Memory||Bandwidth||TDP||Price|
|W5700||2304||1183||8.89 TFLOPS||8 GB 1||484 GB/s||205W||$799|
|WX 9100||4096||1200||12.3 TFLOPS||16 GB 8||484 GB/s||230W||$1399|
|WX 8200||3584||1200||10.8 TFLOPS||8 GB 8||512 GB/s||230W||$999|
|WX 7100||2304||1188||5.73 TFLOPS||8 GB 3||224 GB/s||130W||$549|
|WX 5100||1792||713||3.89 TFLOPS||8 GB 3||160 GB/s||75W||$359|
|WX 4100||1024||1125||2.46 TFLOPS||4 GB 3||96 GB/s||50W||$259|
|WX 3100||512||925||1.25 TFLOPS||4 GB 3||96 GB/s||50W||$169|
|WX 2100||512||925||1.25 TFLOPS||2 GB 3||56 GB/s||50W||$129|
|Notes||1 GDDR6; 2 GDDR5X; 3 GDDR5; 4 HBM2|
5 GDDR6 (ECC); 6 GDDR5X (ECC); 7 GDDR5 (ECC); 8 HBM2 (ECC)
WX 2100~7100 = Polaris; WX 8200~9100 = Vega, 5700 = Navi
Another thing worth pointing out about this card is that it’s built around PCIe 4.0, providing much higher bandwidth over 3.0. In the real-world, and for the vast majority of ProViz use-cases, PCIe 4.0 is not going to offer a single noticeable benefit. Where that will change is with the use of 8K uncompressed video, which can actually take advantage of all of the gained bandwidth to ensure smoother playback. PCIe 4.0 will benefit storage more than anything, at least right now.
There’s a lot more to Radeon Pro than just the hardware. Over the past couple of years, AMD’s had strong momentum with its Radeon ProRender engine. The company had a lot to talk about during SIGGRAPH this past summer, and now, it’s doling out even more good stuff. A handful of plugins have released new versions, including 3ds Max, which has been without an update for a while. There’s also new support for Houdini, SolidWorks Visualize, Z Emotion, as well as InstaLOD. It’s great to see some good progress here.
On the following pages, the results of our workstation GPU test gauntlet will be seen. The tests chosen cover a wide range of scenarios, from rendering to compute, and includes the use of both synthetic benchmarks and tests with real-world applications from the likes of Adobe and Autodesk.
A review like this would ordinarily include more performance results than it does, but we were roadblocked with some tests we wanted to run. We’ve been without working Radeon ProRender projects since the spring, and for whatever reason, we can’t get SiSoftware’s Sandra to work with Navi. Some have had success, but for some reason, we keep getting GPU call fails. The other missing application is MAGIX Vegas Pro, as it doesn’t support Radeon Navi right now at all. We’re told the next build should include support for it, and we hope that’s the case, as the first GPUs have been available since July.
Here’s the specs of our test rig:
|Techgage Workstation Test System|
|Processor||Intel Core i9-9980XE (18-core; 3.0GHz)|
|Motherboard||ASUS ROG STRIX X299-E GAMING|
|Memory||Corsair Dominator Platinum RGB (CMT64GX4M4Z3600C16)|
4x16GB; DDR4-3600 16-18-18
|Graphics||AMD Radeon Pro W5700 (8GB, Enterprise 19.Q4)|
AMD Radeon Pro WX 7100 (8GB, Enterprise 19.Q4)
AMD Radeon Pro WX 5100 (8GB, Enterprise 19.Q4)
AMD Radeon RX 5700 XT (8GB, Adrenaline 19.11.2)
NVIDIA Quadro RTX 4000 (8GB, Quadro 441.12)
NVIDIA Quadro P2000 (5GB, Quadro 441.12)
NVIDIA GeForce RTX 2060S (8GB, GeForce 441.12)
|Storage||Kingston KC1000 960GB M.2 SSD|
|Power Supply||Corsair 80 Plus Gold AX1200|
|Chassis||Corsair Carbide 600C Inverted Full-Tower|
|Cooling||NZXT Kraken X62 AIO Liquid Cooler|
|Et cetera||Windows 10 Pro build 18362 (1903)|
Our benchmark results are categorized and spread across the following four pages. On page 2, we’re taking a look at renderers with the help of Blender and LuxMark. Page 3 is home to our encoding tests, which are handled by Adobe’s Premiere Pro, BRAW Speed Test, and an application we just happen to throw into the same pile: Agisoft’s photogrammetry tool, Metashape. Page 4 handles viewport performance across a range of popular suites, and finally, page 5 tackles some simple gaming.
And with all of that covered, let’s get on with it:
It’s not wise to take the performance from one workload and expect that it’ll scale the same way in another. In some cases, even different projects inside of the same application can scale quite a bit differently, something we’re seeing proof of with the results above. In the BMW test, which has become a standard Blender benchmark, NVIDIA’s GPUs prove to be faster than AMD’s Navi-based models, but the roles reverse in the more complex Classroom scene, albeit to a smaller degree.
The results above take care of Cycles, but what about the new-to-2.80 Eevee?
Our Eevee results seem to agree with the scaling of the Cycles / BMW test, in that NVIDIA’s comparative cards edge out AMD’s. What’s incredibly clear is that the older lower-end cards leave a lot to be desired. If you’re running a graphics card over two, and especially three years old, you are going to be enjoying a serious performance uptick when you finally upgrade (and you might just be overdue).
In the Eevee render, the Radeon Pro W5700 finishes behind the RX 5700 XT, which is to be expected, given its extra cores. At the same time, it’s hard to figure out why the RX 5700 XT didn’t win the BMW and Classroom tests, or likewise why the W5700 won. It’s certainly not the first time we’ve seen odd scaling like this, and it won’t be the last.
To keep things organized better, Blender viewport performance can be found on the dedicated page, which happens to be page four.
We’re continuing to see some great variation in the results with LuxMark. NVIDIA’s Quadro RTX 4000 has made it to the top, while the W5700 and RX 5700 XT swap places depending on the test. We’re glad to see that the instance of the same switch-up seen in the Blender test between these two GPUs has carried over to another test – so it’s not just a fluke. How the W5700 ever manages to come ahead with fewer cores, we’re not sure, but it bodes well for a potential 5700 XT-equivalent card in the future.
Unfortunately, this wraps up our rendering tests for the W5700. We’ve been without working Radeon ProRender tests for many months, so we’re still forced to continue ignoring that one for now. If you happen to be a ProRender user who wouldn’t donating a working ProRender project for us to use in our benchmarks (with full credit given), please leave a comment. Likewise, if you’re a Radeon user who works with a renderer we don’t cover, we’d love to hear about it.
The projects used for our Premiere Pro testing are not the most complicated, but they do reflect real videos that have made it to our YouTube channel. Despite both being fairly similar in design, with similar effects used, but we can clearly see that they behave pretty differently overall. In the 1080p test, NVIDIA claims the top two spots, with the W5700 actually falling behind the last-gen NVIDIA Quadro P2000.
We’re not sure why AMD’s fallen behind in the 1080p test, but the 4K one reflects a much better impression overall, tying the lead with the Quadro RTX 4000. Part of these issues can simply be explained by the erratic behavior of Adobe products at times.
Since the last workstation GPU review, we’ve updated some of our tests, including Premiere Pro. The charts below reflect codec performance for AVC, RED, and ProRes, each of which are encoded to AVC and HEVC. We plan to bring similar tests into DaVinci Resolve soon and begin publishing regular results for that. But alas, let’s first continue with the topic at hand:
First and foremost, the RX 5700 XT doesn’t appear in these charts because the computer would blue screen each and every attempt – even after a hardware and driver reinstall. We’re not sure what the deal is, but we’ve stopped being surprised by these anomalies. We’ll just revisit before the next ProViz article.
Looking at these results, we’re left wondering why we waited so long to add them. They have far more interesting scaling than the project encodes do, and they shake things up quite a bit depending on the test and target codec. With AVC as a source, the Radeon Pro W5700 and Quadro RTX 4000 share the same performance when encoding to AVC, with the HEVC nod leaning in NVIDIA’s favor.
We can draw a couple of simple conclusions from the other results. NVIDIA’s HEVC encode performance exceeds AMD’s in every case, but AVC performance is similar… in all but ProRes. That encoder clearly likes the Turing GPUs quite a bit.
With Blackmagic’s recently released RAW Speed Test benchmark, we’re seeing completely different levels of performance between AMD and NVIDIA hardware, which is a bit odd since AMD definitely optimizes for Resolve workflows. As mentioned above, we’ll be adding Resolve to our suite at some point soon, so we’ll see if our real-world results match up with the above.
Metashape can use the GPU during multiple stages of building the project, but the depths maps generation is when it’s used to best effect, and where it will scale the best. The RTX 4000 again takes the lead here, but that’s not the most interesting thing about these results. Despite its fewer cores, the W5700 once again outperformed the RX 5700 XT.
We begin our look at viewport performance with SolidWorks, an application that makes it obvious that gaming GPUs are second-rate. In real use, gaming GPUs are going to fare well enough, as long as the GPU you’re talking about is at least mid-range. Performance in SolidWorks isn’t crippled for gaming GPUs like it is in some others, but for the heaviest of projects, a pro card like the W5700 should definitely be emphasized.
It’s worth noting that SolidWorks also offers an advanced shading mode, RealView, which is exclusive to workstation graphics cards. We don’t have persistent access to a SolidWorks license, or else we would have included a look at its performance here. You can see RealView comparisons from our RTX 4000 review.
AMD gave NVIDIA a hard fight in SolidWorks with its W5700, and that scuffle has carried onto CATIA. Both the W5700 and RTX 4000 wound up with matched performance at 1080p, but at 4K, NVIDIA inched ahead. Since there’s not too much to talk about here, it might be worth highlighting the fact that (while gaming GPUs are not suggested for critical workloads,) AMD’s gaming GPUs perform a lot better than NVIDIA’s. The RTX 4000 absolutely trounces the RTX 2060 SUPER, but the W5700 only modestly beats RX 5700 XT in both SolidWorks and CATIA.
As we like to say, it pays to know your workload, and this page proves why there’s sense to that. AMD’s W5700 topped out the CATIA and SolidWorks results overall, but the RTX 4000 has struck back in Creo.
How fun are these graphs? The vendors of all of the software on this page (aside from Blender) would recommend using a workstation graphics card over a gaming one, but Siemens has made it clear that gaming GPUs are pure scum. In terms of these scores, NVIDIA’s workstation GPUs perform about 20x better than its gaming GPUs, while AMD’s is a much more modest 5x – which, to be clear, is still awful.
Overall, both the RTX 4000 and W5700 perform similarly, with AMD getting the edge in both resolutions.
3ds Max and Maya are often grouped together, but both applications perform a bit differently. NVIDIA’s top two GPUs here share the top in the 3ds Max test (which doesn’t have a 4K version in SPECviewperf), while AMD’s Radeon Pro W5700 managed to overtake the technically quicker RX 5700 XT.
In Maya, AMD’s W5700 keeps glued to the top, delivering a really impressive showing overall. At 4K, even the RX 5700 XT managed to beat out both of NVIDIA’s top cards tested here.
Blender 2.80’s LookDev mode is meant to allow users to get quick feedback on how their scene is coming along, without having the need to wait on an actual render. The computation is high in this mode, and that means the frame rates in the viewport will tank. You won’t design with this mode on, but you will definitely want to pan around while in it, so the more fluid it is, the better.
As these result show, it takes a mighty GPU to deliver quality performance at higher resolutions. 4K isn’t entirely realistic on even the top-end GPUs here. We don’t need gaming-level performance, but 30 FPS is a lot better than 20. You can check this screenshot to see how much of the UI is filled with viewport for these tests.
AMD’s RX 5700 XT has proven to be a really powerful GPU in this test, with the RTX 4000 and W5700 not falling far behind it. Perhaps the most clear point of all is that GPUs like the P2000 and WX 7100 are going to offer pretty painful performance overall. These results will become a lot more interesting as we get more GPUs tested for an upcoming general ProViz performance look.
For the most part, these 3DMark results scale to our expectations. AMD’s RX 5700 XT has done well to lead the DX11 Fire Strike charts, falling victim only to the RTX 2060 in the DX12 Time Spy test. The RTX 4000 doesn’t fare as well as the RTX 2060 SUPER, falling behind it each time.
We’re sure seeing some swinging performance from different GPUs here. In some tests, AMD takes the reigns, while in others, NVIDIA will be the one sitting on top. Overall though, the RX 5700 XT and RTX 2060 perform a lot closer to one another than they did in the 3DMark results. In the ongoing RTX 4000 vs. W5700 battle, both swap their positions between tests, proving equals in the end.
We’re wrapping our results up by witnessing more places being traded. Both the RTX 2060 SUPER and 5700 XT swap positions in each test, to lead the pack, while the RTX 4000 and W5700 swap places beneath those. For pure gaming performance, these workstation cards are held back a little bit.
As with most of our workstation graphics card reviews, it’s hard to sum up AMD’s Radeon Pro W5700 without breaking down what it’s good at. We have many results spread across the previous pages that can highlight just how much performance can change from one vendor to the next, or even from workstation to gaming series.
In our rendering tests, the W5700 performed great for its price point as a workstation GPU. For heavier workloads, it goes without saying that you’re going to greatly value a higher-end option than the W5700 (or competing RTX 4000). By saying that, we’re talking workloads that see renders span many hours, not just minutes. Those familiar with the need for top-end GPUs are no doubt familiar with overnight renders.
Where rendering is king, and budgets are tight, gaming cards are almost always a better choice, unless you’re dealing with some serious software (eg: SolidWorks, CATIA, Siemens, et cetera.) You also have the option of using a lower-end Radeon Pro as the primary GPU, and dedicate a secondary gaming GPU (also Radeon) to rendering duties. Both GPUs could even be used at the same time. Your choice of hardware will largely revolve around your specific requirements.
AMD’s Navi architecture has proven to be a great one for Blender Cycles use, with the W5700 actually managing to surpass the (technically quicker) RX 5700 XT in the more complex Classroom scene. We’d be remiss to not mention the fact that NVIDIA has accelerated Cycles rendering quite a bit with its RTX technologies, something we took a look at a couple of months ago. This tech is currently in beta, but it is impressive, and makes us wish AMD had its own ray tracing solution ready-to-go.
While rendering is great with the W5700, the video encoding performance leaves a little bit to be desired – but again, what impacts you depends on your workflow. AMD had strong performance in our AVC and RED transcodes, but it fell behind NVIDIA with ProRes. That detriment carried over to the BRAW Speed Test, which showed NVIDIA with double the performance.
We’re of the mind that video encoding performance will improve over time, not just as AMD iterates on its drivers, but as software developers better support the Navi architecture. Despite the fact that Navi launched on July 7, MAGIX Vegas Pro 17 still doesn’t support it (or else it would have been included here). We’re told that the next build should add support, and we’re hoping that one will land before we finish testing for our next general ProViz performance look, since AMD tends to be strong there.
The viewport tests are where the real fun happens, because it’s basically a free-for-all to see which GPU will come out on top. While gaming GPUs will work fine in many of these applications, performance isn’t everything. Workstation graphics drivers are specifically vetted for stability in all of their supported applications. For a regular end-user sitting at home doing stuff as a hobby, a gaming GPU can easily fit the bill (depending on workload), but for more critical work, workstation GPUs are suggested.
We have to admit that the W5700 surprised us with its super-strong performance in SPECviewperf. In SolidWorks, Siemens NX, and Maya, it led the pack, and traded blows based on resolution in CATIA. While super-high performance will be enjoyed by those with high-refresh monitors, it’s nice to know if there’s anything holding back performance, it’s not going to be the GPU.
Overall, AMD’s Radeon Pro W5700 is a great workstation graphics card for the money. Ultimately, though, what turns out to be the best choice for you will depend entirely on your workflows. It’s an unrelenting fact that it pays to know your workload, and as our results have highlighted once again, they’re all far from being alike.
Copyright © 2005-2019 Techgage Networks Inc. - All Rights Reserved.