Date: November 23, 2021
Author(s): Rob Williams
AMD has so far released two Radeon Pro models built around its RDNA2 architecture – W6600 and W6800 – and compared to the previous-gen, there’s a lot of improvement. In addition to faster performance in general, we see additions like hardware RT, as well as support for resizable BAR and variable rate shading. Let’s see how both cards fare in our gauntlet of tests.
It doesn’t seem like there’s much room for a dull moment at AMD nowadays, as the company continues to ride the wave of its resurgence in both the CPU and GPU markets. When the company released its first RDNA GPUs in 2019, there was a lot to like, both with performance and features. With RDNA2, which graces the Radeon Pros we’ll be taking a look at here, the company managed to impress us more than we expected.
With RDNA2, AMD delivers performance to allow it to be competitive at the top-end of the market, and adds support for valuable features like resizable BAR, variable rate shading, and hardware ray tracing – all of which can be great for both working and gaming.
To date, AMD has released two RDNA2-based Radeon Pros: the 8GB W6600 at $649, and the 32GB W6800 at $2,249. If this is somehow the first time you’ve heard of the Radeon Pro W6800, then yes – we really did say 32GB.
|AMD’s Radeon Pro Workstation GPU Lineup|
|Cores||Base MHz||Peak FP32||Memory||Bandwidth||TDP||Price|
|W6800||3840||2555||17.83 TFLOPS||32 GB 5||512 GB/s||250W||$2249|
|W6600||1792||2331||10.4 TFLOPS||8 GB 1||224 GB/s||100W||$649|
|W5700||2304||1630||8.89 TFLOPS||8 GB 1||484 GB/s||205W||$799|
|W5500||1408||1744||5.35 TFLOPS||8 GB 1||224 GB/s||125W||$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)
W6000 = RDNA2
W5000 = RDNA
WX 8200~9100 = Vega
WX 2100~7100 = Polaris
If its price and performance level isn’t enough to explain itself, the Radeon Pro W6800 is best-suited for those who don’t want to be held back by anything – memory limits, or even memory errors. If you want ECC (Error Correction Code) memory, the W6800 will be your go-to on the Radeon side. Neither of the two Radeon Pros from the previous RDNA generation offered ECC, so it’s good to see an option return.
For those with more modest needs, especially those working mostly with CAD, or just want a stable workstation, there’s the low-power single-slot 8GB Radeon Pro W6600.
At the back of the W6600 are four full-sized DisplayPort connectors, allowing for four monitors at 4K or 5K, or one monitor at 8K. Opting for even greater monitor support with the W6800, AMD’s included six mini-DisplayPort connectors, supporting six 4K or 5K monitors, or two 8K.
With the release of PCIe 4.0, AMD saw an opportunity to enable resizable BAR support on its RDNA2 cards, letting certain workloads to send data more efficiently between GPU and CPU to enable a performance uplift. Not too long after AMD introduced its reBAR support, or Smart Access Memory in its terms, NVIDIA followed suit, and so will Intel with its upcoming Arc-series GPUs.
AMD has so far released seven different models in its gaming-focused Radeon lineup, so as it stands today, there will be obvious gaps that could be filled in AMD’s current-gen Radeon Pro lineup. Whether or not the company will augment the series with additional models is unknown. We only saw two models in the previous-gen Radeon Pro lineup, as well, so AMD may believe the bases are well covered here.
It’s important to note that the feature set of the Radeon Pro series goes beyond regular Radeon, as you would expect with a different market focus. Most notable is strong support for using a remote workstation, but also noteworthy is the completely overhauled Radeon Pro Software suite that allows you to access every aspect of your GPU quickly and intuitively:
Considering the lack of a call-out in the earlier specs table, it probably won’t surprise you that neither of the new Radeon Pro models deliver strong double-precision (FP64) performance. The focus of the Radeon Pro series is more on creation and design than scientific or financial computation, so there are better-suited models out there if those workloads are important to you. In early 2020, AMD delivered such a solution with its Radeon Pro VII. Its high-end Instinct compute cards likewise have uncapped FP64.
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.
To paint a good overall picture of current performance, we’re using a blend of gaming and workstation GPUs. For some users, the best performance for the buck is the most important factor when scouting out a new GPU, while for others, increased platform stability and optimization through drivers is more integral.
Our system as tested:
|Techgage Workstation Test System|
|Processor||AMD Ryzen 9 5950X (3.4GHz Base, 4.9GHz Turbo, 16C/32T)|
|Motherboard||ASRock X570 TAICHI (EFI: P4.30 04/14/2021)|
|Memory||Corsair Vengeance RGB Pro
8GB x 4 (CMW32GX4M4C3200C16)
Operates at DDR4-3200 16-18-18 (1.35V)
|Graphics||AMD Radeon Pro & Radeon:
AMD Radeon Pro W6800 (32GB; $2,249)
AMD Radeon Pro W6600 (8GB; $649)
AMD Radeon Pro W5700 (8GB; $849)
AMD Radeon Pro W5500 (8GB; $429)
AMD Radeon RX 6900 XT (16GB; $999)
AMD Radeon RX 6800 XT (16GB; $649)
AMD Radeon RX 6800 (16GB; $579)
AMD Radeon RX 6700 XT (12GB; $479)
AMD Radeon RX 6600 XT (8GB; $379)
AMD Radeon RX 6600 (8GB; $329)
NVIDIA Quadro & GeForce:
NVIDIA Quadro RTX 6000 (24GB, $3,999)
NVIDIA Quadro RTX 4000 (8GB, $899)
NVIDIA TITAN RTX (24GB, $2,499)
NVIDIA RTX 3090 (24GB, $1,499)
NVIDIA RTX 3080 Ti (12GB, $1,199)
NVIDIA RTX 3080 (10GB, $699)
NVIDIA RTX 3070 Ti (8GB, $599)
NVIDIA RTX 3070 (8GB, $499)
NVIDIA RTX 3060 Ti (8GB, $399)
NVIDIA RTX 3060 (12GB, $329)
|Storage||AMD: Samsung 480GB SATA SSD
NVIDIA: Samsung 480GB SATA SSD
Export SSD: Aorus 2TB PCIe 4.0 GP-ASM2NE6200TTTD
|Power Supply||Corsair RM850x (850W)|
|Chassis||Fractal Design Define C Mid-tower|
|Cooling||Corsair iCUE H115i RGB PLATINUM (240mm)|
|Et cetera||Windows 10 Pro build 19043 (21H1)
AMD Drivers: Adrenalin 21.9.2 / Enterprise 21.Q3
NVIDIA Drivers: Studio 472.12 / Quadro 472.12
Our benchmark results are categorized and spread across the following pages. On page 2, we’re taking a look at renderers with the help of Blender, Radeon ProRender, and LuxMark. Page 3 is home to our encoding tests, which are handled by Adobe’s Premiere Pro, VEGAS Pro, BRAW Speed Test, and an application we throw into the same pile: Agisoft’s photogrammetry tool, Metashape. Page 4 handles viewport performance across a range of popular suites. Page 5 tackles synthetic system tests with the help of SiSoftware’s Sandra, while page 6 covers the same synthetic angle for gaming. The final page will include some power tests along with our final thoughts.
And with all of that covered, let’s get on with it:
Much to AMD’s chagrin, we’re sure, NVIDIA has a bit of a stranglehold over Blender performance right now. That’s something that applies whether we’re talking rendering with Cycles or EEVEE, or editing inside of the viewport. With the purely apples-to-apples Cycles results above, it’s clear that NVIDIA’s Ampere architecture really suits this workload, because even the lowbie GeForce RTX 3060 Ti manages to beat out not just AMD’s top-end GPUs, but even NVIDIA’s own last-gen top-end GPUs (eg: TITAN RTX).
What’s notable about these Cycles results is that NVIDIA’s GPUs weren’t tested with the OptiX API, which would have resulted in even greater performance deltas. Unfortunately for AMD, the EEVEE render engine gives us similar scaling:
On the Radeon Pro side specifically, we can see that AMD’s workstation cards perform right alongside the respective gaming counterparts – eg: the W6800 delivers the same performance as the RX 6800. You’ll notice RX 6800 is not AMD’s top-end GPU, so there’s still room for the company to expand with more high-performance Radeon Pro models in the future.
As we recently covered in our news, Blender’s developers are planning to eventually migrate some of the software’s functions to use the Vulkan API. When that happens, we’re sure to see the results shaken up a bit, as each GPU vendor has had different levels of focus for optimizing for the API up to this point.
Rendering is just one aspect of any 3D design application; viewport is another, and we’ll be taking a look at that in more detail later.
Whereas NVIDIA proved a bit too strong for AMD with Cycles and EEVEE, its own Radeon ProRender engine fares a lot better. While NVIDIA still manages to edge out Radeon a bit, it’s not to the level we saw before, with AMD’s top-end GPUs actually appearing near the top of the chart.
Ever since the launch of AMD’s Vega architecture, the company has performed extremely well in LuxMark, and not too much has changed with the latest generation RDNA2 – although once again, NVIDIA manages to keep hold of a lead at the top-end.
It’s worth pointing out that it’s just the gaming cards that manage to peg the top of the chart; when we look explicitly at the workstation GPUs, AMD’s perform well against NVIDIA’s last-gen Turing GPUs. Given NVIDIA’s A series GPUs are built on Ampere like the RTX gaming cards here, we’d expect that no matter how you spin it, AMD is likely to fall slightly behind NVIDIA here.
In advance of benchmarking for this article, we wanted to revisit our Premiere Pro tests, and see if they could be improved. One thing we realized with different encodes was that testing 1080p encoding is really difficult nowadays because today’s GPU hardware doesn’t consider it much of a challenge. What we saw were results that were a bit too sporadic, so we chopped out the 1080p YouTube test we used to include.
Similarly, we used to output to 1080p for our 4K tests, and 4K for our 8K tests – both bad moves. As it happens, if you output to a lower resolution, it’s not as grueling as it would be to encode to the same resolution. Ultimately, the decoder would perform similarly between GPUs in the same series, and then the encoder would output to a low resolution that again can lead to sporadic results.
For this article, we fixed all of that, and made sure our projects outputted to the same resolution as the source – which is probably how most people would be using their high-res footage. While we chopped out the 1080p YouTube test, we added a 4K project that was designed with many common effects in place, all of which happen to use the GPU to pretty good effect.
Let’s check out the results:
Our 4K YouTube test is pretty standard fare, but represents a typical project someone would upload to their channel. Overall, NVIDIA places ahead in that test, and by a pretty significant margin. Roles reverse a bit with second project, however, with AMD really doing well to crunch through the GPU-intensive effects.
What’s really quite interesting with the heavy effects result is that the W6800 places alongside the RX 6900 XT, even though the latter is faster from an overall performance viewpoint. It’s as if the extra memory on the W6800 helped us out in this 4K test, something that bodes well for even higher resolutions.
We’re not done – let’s check out some codec encodes:
The results above paint another interesting picture: NVIDIA dominates AVC encoding, while AMD dominates HEVC – the latter of which has a slightly greater delta than the former. This is quite a good look for Radeon. And guess what? The interesting results are still not done:
It’s immediately clear that if you want to work with 8K footage, then AMD’s last-gen RDNA1-based GPUs are not for you. The encodes on those models couldn’t even begin the process. Considering that other 8GB GPUs were able to pass the tests, we’d have to imagine this is a limitation of the encoder chip.
On the topic of 8K, we have yet another interesting result, with the RED test. We used to see good scaling with this test, but lo and behold, as soon as we decided to output the encode to 8K as well, every GPU began to perform the same. This is something we can’t quite explain, but going forward, we’ll re-introduce the 8K > 4K encode for both RED and ProRes.
While AMD’s last-gen Radeon Pros struggle with 8K, the new-gen loves it, with every single RDNA2-based Radeon proving faster than NVIDIA’s Ampere and Turing in our ProRes test. In another nod to AMD, Radeon proved quickest in all of our HEVC encode tests.
Right off the bat, our VEGAS results show slightly different scaling than Premiere Pro did. The above tests represent a project with absolutely no effects; it’s straight footage that’s transcoded. In the case of both AVC and HEVC, AMD proves itself the champion – but especially so with the latter. The performance deltas there are pretty huge.
Let’s test out some effects:
We’ve been testing VEGAS for quite a while now, and overall, AMD has always seemed to have an advantage – and nothing has changed up to this point. We’ve yet to dig into the new VEGAS Pro 19, but will be doing so soon. We don’t expect any major shake-ups there, from what we’ve seen so far.
While not the focus of this review, it’s worth highlighting some of the odd results here. The TITAN RTX falls to the bottom in both of these tests, nonsensically, and the last-gen Quadros don’t fare much better. It used to be that adding VEGAS Pro as an application inside of NVIDIA’s Control Panel would “fix” performance issues like these, but that no longer seems to be the case. It goes without saying that we couldn’t suggest going with Quadro or TITAN for VEGAS Pro use.
Overall, it seems that NVIDIA has an inherent advantage with BRAW, as it manages to beat out AMD in both of these compression tests – the latter of which (3:1) highlights rather stark differences. Considering the W6800 and RX 6800 flip-flopped between tests, it doesn’t appear the 32GB frame buffer of the Radeon Pro makes much of a difference here, but naturally, it would if your project could actually use that much memory.
An entire photogrammetry process will consist of different tasks throughout, each of which will use either the CPU or GPU, or both together. In Metashape, the Align Photos function uses both processors, but the overall time the process takes makes the results non-interesting from a scaling standpoint. Build Depth Maps is different, as the above chart highlights.
AMD lags a bit behind NVIDIA in this particular test, when comparing one competitive card to another. Fortunately, Metashape, and every other photogrammetry application heavily uses CPUs, so AMD isn’t exactly losing out on much here, since its Zen processors are great for these tasks.
Some CAD suites perform a lot better on workstation graphics cards than their gaming counterparts, but SolidWorks isn’t one of them. That said, when using such a high-end suite like this, we feel compelled to suggest going with a workstation card even if performance lags a bit behind those gaming cards. The reality is, companies like Dassault focus their entire support efforts on workstation GPUs, and likewise, AMD and NVIDIA have special tuning in their respective drivers for things beyond the viewport.
If on a budget, you’re not going to be out-of-luck with a gaming card, but it’s worth highlighting the fact that AMD’s and NVIDIA’s workstation GPUs are required for SolidWork’s RealView mode, so you will need to investigate how important of a feature that would be in your workflow.
All of that said, results like these highlight the fact that AMD can fill many gaps in its Radeon Pro lineup. On the gaming side of things, there are currently seven RDNA2-based cards, while there are only two in the Radeon Pro series. Hopefully we’ll see more models added before long.
In any event, considering that the Quadro RTX 6000 itself places ahead of AMD’s Radeon Pro W6800 here, we can only imagine that NVIDIA’s latest-gen cards would pull even further ahead. It’s hard to beat the 32GB frame buffer of the W6800, however.
As we covered above with SolidWorks, we tend to suggest workstation GPUs for any Dassault product, and that includes CATIA. You can get by with a gaming card, but on the support front, you need to realize all parties involved are focusing entirely on workstation GPUs for their tuning and optimal support. While workstation GPUs cost more than their gaming counterparts, it’s a fraction of what it costs for a license for one of these tools.
What’s impressive to us here is just how much further ahead both the TITAN RTX and Quadro RTX 6000 are from the other GPUs, which is another thing that hints that current-gen Ampere workstation GPUs could appear even higher up. On the Radeon Pro front, the W6800 delivers a strong gain over the gaming-oriented RX 6800 that sports similar hardware (minus the mega amount of memory).
NVIDIA’s been a bit of a force to be reckoned with so far in these viewport results, and that carries on through to PTC’s Creo. Both last-gen Quadros in our possession soar to the top, sitting alongside the TITAN RTX, a card that effectively places itself in between the gaming and workstation series in terms of optimizations.
Yet again, we see the current-gen Radeon Pros vastly surpass the performance of their gaming equivalents, highlighting what those extra driver optimizations can manage.
If you’re trying to decide between a gaming or workstation GPU for Siemens NX, the decision has effectively been made for you. Gaming GPUs may fare pretty well in solutions from Dassault, but in SNX’s case, the gaming cards are almost actively worked against. AMD’s newest Radeon Pros perform great here, with the W6600 matching the Quadro RTX 4000 overall. We’re not sure the same could be said if it was the current-gen A4000 we had in here.
Compared to most of the viewport tests here, 3ds Max doesn’t care if you use it with a gaming or workstation GPU, with Autodesk even going as far as to list gaming SKUs in its support pages. That said, both AMD and NVIDIA will add extra focus on applications like 3ds Max (and Maya) in its drivers, and if you want ECC, you’re not going to get it on a gaming GPU (or even the W6600, unfortunately.)
Overall, AMD’s new Radeon Pros perform similarly to their gaming equivalents. It goes without saying, though, that most of these GPUs are going to deliver apt performance here; the biggest differences will be seen with rendering performance, and not to mention the amount of memory on tap.
With Maya, you could mirror many of the sentiments that was said about about 3ds Max. Gaming cards are treated like normal silicon citizens here, with most of the GPUs tested here proving more than apt for the job at hand. The outlier really seems to be the W5500 from the last-gen Radeon Pro series – it sits quite a bit further back in the pack than we would have expected.
Naturally, viewport performance is just one aspect that’s important to consider with any GPU. While an 8GB model might slay SPECviewperf’s test, it doesn’t necessarily mean that the limited memory won’t act as a roadblock. That’s one reason it will be hard to beat a GPU like the 32GB Radeon Pro W6800, or even NVIDIA’s GPUs that sport 24GB of memory. It’s nice to have enough VRAM that you know you won’t be held back.
Blender is another design suite that doesn’t favor workstation GPUs over gaming, but as we’ve seen in the past, NVIDIA tends to have a bit of an edge here. Nothing about that changes in these viewport results, with the GeForces and Quadros eking slightly ahead of the Radeon competition. That said, most of these GPUs can hit 60 FPS at 4K with this simpler Controller project, so your most important decision will likely hinge on the amount of memory available.
The Medical viewport test, conducted with the help of SPECviewperf, highlights some interesting results between resolutions here. NVIDIA’s Quadro RTX 6000 pegs to the top of the 1080p test, but comes back down to parity with the second-place GPU at 4K. We’re really not sure what exactly propels that Quadro to the top at 1080p, or why it doesn’t exhibit the same lead at 4K.
Nonetheless, one thing that really stands out here is that AMD’s current-gen Radeons naturally love this workloads. It’s impressive to see just how close the mid-range RX 6600 XT gaming GPU manages to keep up to NVIDIA’s last-gen TITAN RTX, or even current-gen RTX 3070 Ti. For both AMD and NVIDIA, workstation GPUs exhibit big boosts here when compared to the gaming equivalents.
To wrap up our viewport tests, we take a look at another serious test: energy. As we saw with the Medical test above, both AMD’s and NVIDIA’s workstation GPUs are clearly optimized for this energy workload, with AMD’s (current) top-end Radeon Pro W6800 pushing to the top in both resolutions, sitting just behind the Quadro RTX 6000.
Interestingly, most of these specific GPU computing tests scale similarly, with the exception being with the memory bandwidth test. In the pure performance standpoint, NVIDIA’s gaming-focused GeForce RTX 3090 keeps itself pegged to the top of every single chart here.
Compared to some of the tests we conduct, Sandra is about as neutral as it gets, utilizing any hardware’s capabilities to show what’s possible with proper optimization. We’d love to see AMD’s Radeon Pro W6800 place higher in these charts, but as mentioned earlier, the W6800 isn’t AMD’s highest-performing SKU – so hopefully we’ll be seeing a W6900 in the future, to help spice things up more. That said, it wouldn’t be hard to picture where it’d place, given the performance we see from the RX 6900 XT.
The primary focus of this Radeon Pro review is to tackle creative benchmarking, but since it’s been a while, we wanted to run each of the tested GPUs through our usual gamut of synthetic gaming benchmarks. We won’t offer any commentary here, since these results wouldn’t influence our final opinions in a meaningful way. Ultimately, neither workstation nor gaming cards have inherent advantages here, so the Radeon Pro W6800 behaves much like the RX 6800 – as we’d hope.
To take a look at our tested GPUs from a power perspective, we’re using UL’s 3DMark Fire Strike 4K stress test, and our PC plugged into a Kill-A-Watt to monitor full system power draw. The PC is left to sit idle for about five minutes after a fresh boot, at which point the Fire Strike test is kicked-off. The load value is recorded after about five minutes, at the same point in the looped test.
We normally like to include load values for both gaming and rendering workloads, but unfortunately, we discovered too much variability with our chosen Blender render test, so we’re going to figure out how to negate that going forward, in time for the next round of power testing.
While the Radeon Pro W6800 has twice as much memory as the RX 6800, that didn’t make any difference in the power consumption of the card in our gaming test. Overall, AMD’s current-gen RDNA2 GPUs prove really efficient, which is great to see, as some Radeon architectures in recent years haven’t been so lucky (eg: RX Vega). As for the W6600, it proves to be one of the most modest power sippers of this bunch.
As with all workstation GPU reviews, it’s impossible to sum up cards like the new Radeon Pro W6600 and W6800 based on a simple metric. Unlike with gaming, where one competitive card will usually scale similarly to another throughout most of the game tests overall, some of our results in this article highlight once again that not all workloads are built alike.
Overall, there’s a lot to like about AMD’s new Radeon Pros, and it’s why we hope to see some additional models help pad the lineup – especially something like a W6700, which could sit nicely between the W6600 and W6800 in both performance and pricing, ideally with 16GB of memory.
In some cases, AMD’s new cards impressed us quite a bit. It outperformed NVIDIA in a number of video encode tests – not just some of our Adobe Premiere Pro tests, but the entirety of VEGAS Pro. Considering how renowned NVIDIA’s encoder has become, it’s interesting to see AMD beat it out in a number of HEVC tests, but not only, as our VEGAS Pro tests revolve exclusively around AVC output. NVIDIA doesn’t lag behind in all cases, however, so as always, it pays to know your workload, and what’s important to you. One thing we do know: if you work with 8K and want Radeon, you want one of these new cards.
With rendering, AMD’s new cards deliver the level of performance we’d expect from each card, but unfortunately for the company, if a renderer supports NVIDIA’s OptiX API, then it’s probably going to fall quite a bit behind in the results. AMD’s RDNA2 cards can also do accelerated ray tracing, but so far, only Radeon ProRender supports it, and even from our testing there, NVIDIA still manages to come ahead. Ultimately, what could prove alluring for rendering on a card like the Radeon Pro W6800 would be the enormous 32GB frame buffer.
As for viewport performance, AMD’s latest cards hit all the marks they need to. NVIDIA may still place ahead in certain cases, but the reality is, with solutions like SolidWorks and CATIA, you don’t really need a high-end card to have a good experience. That helps make a card like the W6600 attractive, given its price-point, but larger projects could benefit from the W6800’s larger frame buffer – and not to mention improved rendering performance.
While hardware RT isn’t available in many solutions for AMD yet in creative applications, it is available for use in many games. Whereas NVIDIA’s Turing architecture remained exclusive for a while in the hardware RT game, RDNA2’s launch brought competition back. With the addition of that, as well as resizable BAR and variable rate shading, the current-gen Radeon Pros are definitely a step-up over the previous generation.
Copyright © 2005-2021 Techgage Networks Inc. - All Rights Reserved.