Maya 2017 Quadro RTX 4000 vs. Radeon Pro WX 7100: Testing AMD’s Claims

A couple of weeks ago, AMD posted Maya 2017 performance claims to its social media accounts, with us catching it on Twitter. Both AMD and NVIDIA regularly post similar things, but this one stuck out to us more than usual, because it didn’t actually make sense to us. Here’s the performance chart:

This claims that the Radeon Pro WX 7100 is 30% faster than NVIDIA’s Quadro RTX 4000 in SPECapc Maya 2017. Unlike SPECviewperf, which uses application traces to benchmark different viewports, SPECapc uses real applications, essentially meaning that the results it spits out are a bit more comprehensive than SPECviewperf. It also means you need to have access to the exact version of the software that SPECapc supports, and fortunately, we do.

In the past, we’ve typically seen NVIDIA perform stronger in Autodesk software – not just Maya, but even AutoCAD. So to see a claim that the last-gen WX 7100 was 30% faster than the current-gen RTX 4000 had to be one we tested ourselves. Here’s a question: if there’s a 30% advantage on WX 7100 over RTX 4000, then surely the WX 8200 would absolutely demolish it? Both the RTX 4000 and WX 8200 cost about the same.

Unfortunately, our WX 8200 has essentially died on us, so we can’t test that claim (we have a replacement in the form of the WX 9100, but it’s not relevant to this particular look). Since it could be a little important: this is not all of the SPECapc Maya testing we’re doing. We’re currently testing a dozen or so cards with it, and will produce an article later this week. For now, we wanted to single out these two GPUs, since it’s those that AMD used in its marketing.

SPECapc Maya 2017 project sample

For its testing, AMD used the appropriate Update 4 version of Maya 2017, and so we followed suit (that is the same version officially supported by SPECapc Maya; Update 5 is newer, so we didn’t touch it to avoid issues.) AMD tested the WX 7100 with the 19.Q3 driver, and NVIDIA’s RTX 4000 with the 431.02 driver. We matched everything again here. The only critical difference between test beds is that AMD’s used a 4-core Xeon CPU, whereas ours used an 18-core Core i9 CPU. AMD used 16GB of memory; we used 32GB.

Here’s our performance after three runs each (a reboot was done in between each run):

In AMD’s chart above, the value it’s reporting on is GPGPU Composite, which in our chart shows NVIDIA in the lead for. We reached out to SPEC to inquire what GPGPU Composite even represented, because we thought it was straight rendering, but had to make sure. We were told that this test is essentially a filler for future enhancement, and is generally unimportant at this point in time due to its inability to scale properly. It’s the Interactive and Animation scores that are much more important. AMD edges out NVIDIA in animation, while the roles reverse in the interactive result.

We reached out to AMD to inquire about this performance discrepancy, and were told that its tests were done on a (4-core) Xeon, had 16GB of memory, and used the appropriate version of Windows, Maya and SPECapc. We of course exceeded all of these requirements, but still don’t see anything close to what AMD is reporting.