Why you can trust Tom's Hardware Our expert reviewers spend hours testing and comparing products and services so you can choose the best for you. Find out more about how we test.
GPUs are also used with professional applications, AI training and inferencing, and more. Alongside our professional tests, we've added Stable Diffusion benchmarks on the various GPUs. We're using the Automatic1111 Stable Diffusion project for the Nvidia cards, with TensorRT extensions enabled. Intel has an OpenVINO fork, and AMD has a DirectML fork, both of which are recommended by the respective companies for their GPUs.
AMD RX 7900 GRE AI Performance
The RX 7900 GRE specs are worth revisiting here, in that it has 8% less memory bandwidth than the 7800 XT but 23% more compute. Most of our gaming benchmarks ended up with the 7900 GRE about 10% faster on average, but in Stable Diffusion at least, it seems memory bandwidth plays a bigger role.
The RX 7900 GRE comes in just slightly behind the 7800 XT at 512x512 image generation, and 4% ahead with 768x768 images. Higher resolutions with Stable Diffusion are often more compute intensive, so that makes sense, but it also shows that there are at least some instances where the normally faster 7900 GRE will fall behind the cheaper card. Notice also that the 7900 XT is 34% faster at 512x512 and 31% faster at 768x768, and it has 39% more bandwidth.
Looking at Nvidia's competing GPUs, their AI prowess shows up in a big way, with the RTX 4070 providing 86% faster 512x512 image generation, and 76% faster 768x768 generation. Even the significantly less expensive RTX 4060 Ti 16GB beats the 7900 GRE by 32% and 16%, respectively, at those same resolutions.
Not only are Nvidia's GPUs substantially faster at AI, but there are far more projects that target Nvidia hardware — not to mention things like the new Chat With RTX app that Nvidia now provides. As with ray tracing games, if you're really interested in AI, we strongly recommend picking an Nvidia GPU right now. The company has a substantial lead over its competitors in this arena, and it could be many years before the gap narrows.
AMD RX 7900 GRE Professional Workloads
We use SPECviewperf 2020 v3 to assess professional workstation graphics performance on the various GPUs. Somewhat akin to the rasterization results — but for entirely different reasons — AMD's RX 7900 GRE does extremely well overall compared to the RTX 4070. That's because Nvidia only offers full performance tuning in its drivers for professional GPUs like the RTX 2000 Ada Generation, while AMD has decent levels of driver tuning even on its consumer cards.
Overall, the RX 7900 GRE delivers more than double the performance in SPECviewperf 2020 — it's 127% faster in our geomean. But the geomean here isn't quite as useful, as a lot of professionals may use only one or two specific applications. Even so, out of the eight individual tests, the 7900 GRE beats the RTX 4070 in every single case — it's 8% faster in 3dsmax-07, 21% faster in maya-06, 35% faster in solidworks-07, and 41% faster in creo-03. And then we get to the apps where it absolutely destroys the Nvidia card: 120% faster in catia-06, 126% faster in energy-03, 252% faster in medical-03, and 1540% faster — yes, over 16X faster! — in snx-04.
Nvidia might say that it's GeForce cards simply aren't intended to run many of these workloads, at least not at higher levels of performance, but that's by design. The underlying hardware in the GeForce and professional cards is the exact same; it's just that Nvidia has specific optimizations in the professional drivers that it doesn't enable for the consumer parts. If you want to try using a less expensive consumer card for certain professional apps, AMD GPUs are worth considering.
For GPU-accelerated 3D rendering, the only readily accessible app that currently supports all three GPU vendors is Blender. It now leverages ray tracing hardware to boost performance, and that puts Nvidia back in the pole position. We're using Blender Benchmark, version 3.6.0 — Blender 4.0.0 is also available, but the workloads aren't the same and we haven't yet decided to retest all of the GPUs on the latest release.
The results here are effectively the reverse of the SPECviewperf numbers. Nvidia's RTX 4070 is over twice as fast as the RX 7900 GRE in Blender, and there are other commercial 3D rendering apps that also support Nvidia hardware, like Otoy Octane and V-Ray. Given Nvidia basically pioneered modern ray tracing hardware more than two years ahead of AMD, these results shouldn't be too surprising.
AMD RX 7900 GRE Content Creation Summary
Professional workloads aren't normally the primary use case for consumer GPUs. Yes, you can run them, and some of them do just fine, but content creation professionals will usually opt for the additional support and performance offered by professional GPUs — or at least their employers will.
The RX 7900 GRE ends up delivers an interesting mix of results, particularly when compared to the RX 7800 XT. In some cases, its performance tracks closely with the theoretical compute, while in others the memory bandwidth appears to be a bigger factor. In general, though, compute is more critical for these applications and the GRE makes a great alternative to the 7800 XT, where it's up to 25% faster in some cases.
The exception to the above is in AI workloads, or at least Stable Diffusion, where performance basically matches the less expensive 7800 XT. Not that we'd recommend AMD GPUs for people primarily interested in AI, unless they specifically want to work on improving AMD support for such workloads. Nvidia currently rules the roost when it comes to consumer GPUs and AI in general.
- MORE: Best Graphics Cards
- MORE: GPU Benchmarks and Hierarchy
- MORE: All Graphics Content
Intel reportedly demands all board partners implement Intel Default (Baseline) Profile by May 31 — company hopes to fix issues with some Core i9 chips
'Digital Twins' research gains $285 million in CHIPS ACT funding — virtual chip tech could revolutionize semiconductor manufacturing
Explosive HBM demand fueling an expected 20% increase in DDR5 memory pricing — demand for AI GPUs drives production cuts for standard PC memory
