Conclusion
We did not really know what to expect when we thought about performance advantages to GPGPU, Compute, and Workstation workloads. Given the increase in CUDA cores, floating-point performance, and memory bandwidth we thought it would be good, but we didn’t know it would be this good.
3D Rendering
We are highly impressed with 3D and rendering performance. In LuxMark the GeForce RTX 3080 Founders Edition is pretty much 2.5x the performance of GeForce RTX 2080 FE in the “Food” scene, which is 100% brute force path tracing. Then, the RTX 3080 FE is double the performance of even the GeForce RTX 2080 Ti FE in this scene. Further, in the “Hall Bench” RTX 3080 FE is more than double the performance of RTX 2080 FE and 63% faster than RTX 2080 Ti FE. This shows huge gains when it comes to path tracing, proving at the architecture level how well improved Ray Tracing is.
Then V-RAY goes on to prove it again and provides double the performance out of RTX 3080 FE compared to RTX 2080 FE. 61% more performance on RTX 3080 FE versus RTX 2080 Ti FE.
Blender Open Data Benchmark goes on to further prove it. Rendering times were greatly reduced even just using the CUDA API in Blender 2.90. Basically, render times were reduced in half compared to the RTX 2080 FE. It also improves upon the GeForce RTX 2080 Ti FE render time, by a great deal. Then, when you switch on OPTIX and really take advantage of those RT Cores the results just shoot through the roof. Barcelona Pavillion on OPTIX took 53 seconds, less than a minute to render! Victor takes less than a minute and a half. These render times are nothing short of extraordinary.
We saw no differences in HandBrake encoding an H.264 video with NVENC. Though optimizations via application updates for the encoder or architecture, or driver optimizations may be necessary to benefit more from that. Video Transcoding and Encoding is going to be very application-specific. Though at first, it seems there are no outright architecture changes that affect video encoding speed.
GPGPU and Compute
AIDA64 GPGPU Benchmark is a great suite of tests that push the GPU in various aspects. Notably, we saw the memory speed advantage of PCI-Express 4.0. But we also saw some other interesting results. Single-Precision FLOPS seemed to stand out quite a bit, being 176% faster than GeForce RTX 2080 FE and 89% faster than GeForce RTX 3080 Ti FE. That’s a 2.5x increase over the RTX 2080 FE. Double-Precision was 50% faster than RTX 2080 FE. Integer IOPS seemed to be about 50% faster than RTX 2080 FE. Another major standout was AES-256 performance. In this benchmark, it is using OpenCL to measure AES-256 encryption performance of the GPU. The result was an incredible 208% faster than RTX 2080 FE, or 3x the performance!
CompuBench also tests various 3D features and techniques that games do utilize. We saw many tests that put the GeForce RTX 3080 FE in a good position above the rest. This suite of tests showed where the advantages lie in the Ampere architecture. A test that stood out was the Vertex Connection and Merging test where it was more than double the performance of RTX 2080 FE. We also saw more than double the performance in the TV-L1 Optical Flow test.
Workstation
Finally, SPECViewperf 13 showed that in workstation applications, even at 1080p, the GeForce RTX 3080 Founders Edition will provide performance faster than the GeForce RTX 2080 Ti FE. It of course is a lot faster than GeForce RTX 2080 FE. Performance advantages to RTX 2080 FE were 20-60% depending on the application. We saw up to 30% advantage to RTX 2080 Ti. At 4K we continued to see the GeForce RTX 3080 Founders Edition provide better performance than even the GeForce RTX 2080 Ti FE.
Final Points
The new GeForce RTX 3080 Founders Edition, while a gaming powerhouse, is also a GPGPU, Compute, and Workstation powerhouse. We experienced very fast 3D rendering performance, especially where Ray Tracing or Path Tracing is used. Utilizing CUDA was fast, and OPTIX even faster. Compute and GPGPU tests pointed to what we expected, fast performance in the areas of the architecture that are strong, and maybe not as fast in some areas that weren’t buffed as much.
However, one thing is for sure, this $699 video card is faster in every manner of GPGPU, Compute and Workstation performance compared to the more expensive $1200 GeForce RTX 2080 Ti FE. This means, if you are a business on a budget, here is a $699 video card that will now boost your workloads way beyond that of the GeForce RTX 2080 FE and also beyond the GeForce RTX 2080 Ti FE.
You can get superior GPGPU, Compute and Workstation class performance to the GeForce RTX 2080 Ti FE now for a much-reduced price, $500 less than the last generation. That’s an incredible savings, especially if you had to outfit multiple computers in your organization. Combine this video card with the special NVIDIA Studio Driver and you have a very compelling, low cost, fast performing, option for workloads other than gaming.