NVIDIA GeForce RTX 3080 Offers Up Incredible Linux GPU Compute Performance

[phoronix]

Phoronix: NVIDIA GeForce RTX 3080 Offers Up Incredible Linux GPU Compute Performance

Yesterday I finally received a GeForce RTX 3080 graphics card from NVIDIA for being able to deliver the first of our Linux benchmarks on the new RTX 30 Ampere series. What is immediately clear is the huge performance uplift for OpenCL and CUDA workloads with the RTX 3080 compared to its predecessors. The raw performance and even performance-per-dollar is staggering out of the GeForce RTX 3080 with the initial tests carried out on Ubuntu Linux. Linux gaming benchmarks will be out in the days ahead but for now is a look at the RTX 3080 compute performance across dozens of benchmarks and going as far back as the GeForce GTX 980 series for comparison.

http://www.phoronix.com/vr.php?view=29572

[chuckula]

Love them or hate them, Nvidia is still the standard by which all others are judged in GPU compute.

[GruenSein]

So the 30 TFlops as claimed by Nvidia are just 30% faster than 13 TFlops of a 2080 Ti. Interesting...

[abu_shawarib]

@phoronix

Unfortunately we do not yet have an RTX 3080 or any ETA on when we may have that higher-tier card, but the RTX 3080 performance already is mighty impressive.

I think you meant 3090.

Seems good overall at compute. Not entirely surprising since this is the same architecture they have in their data center solution. Still curious about their gaming performance.

[bug77]

So the 30 TFlops as claimed by Nvidia are just 30% faster than 13 TFlops of a 2080 Ti. Interesting...

TFlops are TFlops. They're a bit like potential energy: useful if you put it to use, useless otherwise. But it's still there in both cases.

[moilami]

This is bad news about Intel. This indicates Intel is badly out of game. Nvidia managed to retaliate hard when AMD launched Navi. Intel haven't been able to do the same when AMD launched Zen.

[piotrj3]

Those results are... wierd. From one side certain stuff like Mixbench and luxcore seems to show how good new GPUs are I guess in single precision performance, from 2nd side it seems this GPU really suffers more on FP16 and FP64 side. Also I suspect it suffers more on int32 cores, since Nvidia in Ampere architecture made some cores FP32 and some FP32 OR INT32, while Turing was full FP32 OR INT32.

Wish there were some Quadros in test but I doubt Nvidia would provide them.

[Shnatsel]

It would be interesting to see a comparison of x86_64 vs POWER9 performance for GPU compute, Nvidia especially. POWER9 has cache coherency features that might push performance even further; or it might have some driver deficiencies that destroy performance.

[Papicha]

So the 30 TFlops as claimed by Nvidia are just 30% faster than 13 TFlops of a 2080 Ti. Interesting...

Unfortunately, not every algorithm has linear time complexity.

[zxy_thf]

The performance is weird for NN, because NN applications should easily eat up all ALUs and TCs.
Might be some software issue? I know VkFFT and RealSR only use VK but Caffe should be able to exploit TCs.

This is bad news about Intel. This indicates Intel is badly out of game. Nvidia managed to retaliate hard when AMD launched Navi. Intel haven't been able to do the same when AMD launched Zen.

Xe would be DOA if doesn't have 40 TFLOPS theoretical performance.