GPUs are also used with professional applications, AI training and inferencing, and more. Along with our usual proviz tests, we've added Stable Diffusion benchmarks on the various GPUs. AI is a fast-moving sector, and it seems like 95% or more of the publicly available projects are designed for Nvidia GPUs. Those Tensor cores aren't just for DLSS, in other words. Let's start with our AI testing and then hit the professional apps.
We've tested Stable Diffusion using two different targets: 512x512 images (the original default) and 768x768 images (HuggingFace's 2.x model). For Nvidia GPUs, mostly we find that the higher resolution target scales the time to render proportionally with the number of pixels: a 768x768 image has 2.25X as many pixels as a 512x512 image and so it should take about 2.25X longer to generate.
For AMD GPUs, Nod.ai's Shark repository offers the best performance we've found so far, but only if you're doing 512x512 images; 768x768 output cuts performance to about one-fourth the speed. And Intel? Yeah, I've got 512x512 working at a pretty decent rate (9.2 images per minute on the Arc A770 16GB), but I haven't managed to get 768x768 working yet.
If you're okay with 512x512 images, the RTX 4070 can crank out around 22 images per minute. That's 84% faster than AMD's RX 7900 XT, and that's not even trying to use the newer FP8 functionality (though Stable Diffusion might not benefit as much from that as other workloads, as it might need the additional precision of FP16). Moving up to 768x768 Stable Diffusion 2.1 images, the RTX 4070 still plugs along at over nine images per minute (59% slower than 512x512), but for now AMD's fastest GPUs drop to around a third of that speed.
SPECviewperf 2020 consists of eight different benchmarks, and we use the geometric mean from those to generate an aggregate "overall" score. Note that this is not an official score, but it gives equal weight to the individual tests and provides a nice high-level overview of performance. Few professionals use all of these programs, however, so it's typically more important to look at the results for the application(s) you plan to use.
Across the eight tests, Nvidia's RTX 4070 basically ties the RTX 3080, just as we saw with our gaming performance results. AMD's more recent drivers provided a substantial boost to performance, something that you can really only match with Nvidia's professional series cards.
AMD score particularly well snx-04 (or if you prefer, Nvidia's consumer RTX cards do very poorly). AMD also tends to score higher in catia-06, creo-03, energy-03, and medical-03, while Nvidia GPUs do better in 3dsmax-07 — with maya-06 and solidworks-07 being more neutral. If you use any of these applications on a regular basis, that could be enough to sway your GPU purchasing decision.
Moving on to 3D rendering, Blender is a popular open-source rendering application, and we're using the latest Blender Benchmark, which uses Blender 3.50 and three tests. Blender 3.50 includes the Cycles X engine that leverages ray tracing hardware on AMD, Nvidia, and even Intel Arc GPUs. It does so via AMD's HIP interface (Heterogeneous-computing Interface for Portability), Nvidia's CUDA or OptiX APIs, and Intel's OneAPI — which means Nvidia GPUs have some performance advantages due to the OptiX API.
The RTX 4070 ends up basically tied with the 3080 Ti this time, winning in two of three individual scenes as well as in the aggregate score. It's interesting that Junkshop seems to favor the 3080 Ti, but we don't have any clear indication why that might be so. AMD's GPUs meanwhile fall far behind, and even the RTX 3070 delivers better performance in our testing.
Our final two professional applications only have ray tracing hardware support for Nvidia's GPUs. OctaneBench pegs the RTX 4070 as being just a touch behind the RTX 3080 Ti but decently ahead of the vanilla RTX 3080. V-ray goes the other route and puts performance just barely ahead of the 3080 but a decent amount behind the 3080 Ti.
- MORE: Best Graphics Cards
- MORE: GPU Benchmarks and Hierarchy
- MORE: All Graphics Content
