NVIDIA GeForce RTX 3090 - Windows vs. Linux GPU Compute Performance

[phoronix]

Phoronix: NVIDIA GeForce RTX 3090 - Windows vs. Linux GPU Compute Performance

Following the recent RTX 30 series Linux gaming benchmarks and RTX 30 compute comparison, I was curious how the Linux performance for the flagship GeForce RTX 3090 graphics card compares to the Windows 10 performance in various GPU compute workloads. Well, here are those benchmarks for those wondering about Vulkan / OpenCL / CUDA / OptiX compute performance between Windows and Linux with the very latest NVIDIA drivers.

https://www.phoronix.com/vr.php?view=30180

[marios]

Linux vs windows on nvidia gaming benchmarks would be interesting. If I remember correctly, there are already some results for AMD gpus available.

[schmidtbag]

Interesting how consistent some of these are. Perhaps that's a good thing, where the GPU is a bottleneck rather than the OS.

[hiryu]

Something stuck out to me... "Waifu"?! I had to look this up. Apparently it's some sorta image upscaler: https://linuxreviews.org/Waifu2x

[JackLilhammers]

After this I can imagine birdie's schizophrenia getting worse. His beloved nvidia betrayed him and together with Linux destroyed his favorite spy OS windows in graphics benchmark.

The geometric mean comparison is 76.71 vs 76.05.
0.66 points, it's less than 1%. I wouldn't say Linux destroyed Windows, but I still think it's a great result for its consistency across the board

[Volta]

The geometric mean comparison is 76.71 vs 76.05.
0.66 points, it's less than 1%. I wouldn't say Linux destroyed Windows, but I still think it's a great result for its consistency across the board

Geometric mean is probably the least valuable summary. Summing up percent advantages it's: 114,3 Linux vs 82,3 Windows.

[JackLilhammers]

Geometric mean is probably the least valuable summary. Summing up percent advantages it's: 114,3 Linux vs 82,3 Windows.

I don't think so. The sum of the percentages is misleading.
If Windows had been twice as fast as Linux in one test and tied in all the others, by your logic it would have destroyed Linux...

[Paradigm Shifter]

Linux vs windows on nvidia gaming benchmarks would be interesting. If I remember correctly, there are already some results for AMD gpus available.

They would, but I've already had fun with trying to get Quake 2 RTX working on Linux.

Works fine in Windows (for the all of an hour I had it installed) but it crashes complaining while loading on Arch/Ubuntu for me.

edit: OK, I finished what I was running and decided to check was the error was. Version 1.4.1; "FATAL: Could not find a suitable Vulkan queue family!" Nothing more helpful than that.

If anyone has any ideas, I'd be interested in hearing them, as I've installed, tweaked, uninstalled and generally messed around with just about everything I can find that is remotely Vulkan or nVidia driver related without any success.

[Volta]

I don't think so. The sum of the percentages is misleading.
If Windows had been twice as fast as Linux in one test and tied in all the others, by your logic it would have destroyed Linux...

Yes, but both are quite misleading. In current case Windows could win by winning one benchmark more by beating Linux by 1% in each of the won benchmarks. Meanwhile Linux would beat Windows in 49% of other benchmarks by let's say 60% margin and the winner would be Windows. Geometric mean would be far from showing such disproportion. There's a need for average margin of the advantage in each benchmark. Like: (15% + 23%) / 2 = 19 vs (2% + 24%) / 2 = 13. It addresses the case you mentioned, I think.

[Alexmitter]

Something stuck out to me... "Waifu"?! I had to look this up. Apparently it's some sorta image upscaler: https://linuxreviews.org/Waifu2x

Yes, it is a upscaler specialized for anything digitally drawn. It is really good.