NVIDIA 375.10 vs. Linux 4.8 + Mesa 13.1-dev AMD GPU Benchmarks

[DIRT]

I been noticing at low resolutions AMD drivers don't really run at a high fps like nvidia cards. But when you start turning up the resolution AMD cards don't drop as much percentage wise.

Also would like to note that some native games like tf2 or gmod run slower than crysis does in wine. L4D2 would make my nvidia card breath fire with vsync off because it would utilize a lot of the gpu. On amd it doesn't radeon top confirms this. I'm running a 270x atm. I still question how well optimized for AMD hardware some programs are.

[airlied]

Two points:
- Please don't even consider testing the Mesa OpenCL driver. It's been abandoned. It won't see any new work. If you want to test OpenCL, the closed driver is the only way.
- Similarly for radv. It's not our driver. If you want to compare radv with nvidia, call it "NVIDIA vs Red Hat". But not AMD.

If you can't test the closed driver, just say that, but falling back on Mesa OpenCL or radv isn't a solution.

Please don't say Red Hat, NVIDIA vs radv or community drivers. calling it Red Hat is kinda unfair in that a lot of it was written by a non-RH employee.

Dave.

[bug77]

Got a particular image/link? Somewhat recall what you might be thinking of, but what makes it harder is that all PTS code is universal and not specific to any specific test profiles, so like no graphics-specific hacks, plus the fact of the visual presentation / design element not being my strong point.

I think this is what he means: http://www.hardocp.com/article/2015/...4#.WA-ytBOMQwE

I found it super annoying at first. Then liked it. And now they've stopped doing that and test largely like everybody else
They also do visual comparisons. Haven't found anything noteworthy in recent years, which is why I believe the previous poster was full of it when talking about IQ.

[dungeon]

That is most probably because of SKL Xeon involved in this one

Yeah that @ 4GHz proably has 20% more singlethread IPC then previously used one... could Zen reach there, is yet to be seen

[valici]

To be honest, these benchmark results are much better than I expected:
- I don't think I'd ever seen 90+ FPS for Bioshock with Mesa on Phoronix.
- CS:GO, Tesseract, and Xonotic results at 4K look competitive with NVIDIA.

Agree. I am impressed by the progress. I still believe there is some optimization possible but maybe after implementing more important features.

If we ever have a lot of free time I would be really tempted to work on a hacked-up set of drivers that cut out most of the API validation code, so we can have a more apples-to-apples comparison..

I think this would be a great idea. Or maybe a switch that can be edited to bypass the API validation code. If it can help with CPU bottleneck.

I for one am happy with the results. Since I've been working a lot overtime in the last year, I'm planning to buy a gaming laptop with a mobile RX 480 and a Zen APU. If there will be one .

[dungeon]

- I don't think I'd ever seen 90+ FPS for Bioshock with Mesa on Phoronix.

That is most probably because of SKL Xeon involved in this one

[marek]

To be honest, these benchmark results are much better than I expected:
- I don't think I'd ever seen 90+ FPS for Bioshock with Mesa on Phoronix.
- CS:GO, Tesseract, and Xonotic results at 4K look competitive with NVIDIA.

[dungeon]

No cheating on Radeon Pro SSG, only raw pictures please - that will cost you $10K

[dungeon]

I think that they all cheat on consumer market, but less on professional

[duby229]

Not when I do not - and will not - touch any images/results manually. I already have the automated post-processing support, but alas limited by the game/engine's ability to get the screenshots reliably. Even just using a timer to say trying to get the same elements visible likely wouldn't work since the games that take much longer to load on Mesa than the binary blobs.

And as long as you not willing it won't happen. As of the status quo, the only thing you've been doing is meaningless fps measurements usually way higher than anyone would play at and with no indication of visual quality. So in the real world gamers do in fact tweak their games to reach close to refresh. and they do in fact care very much about visual quality. But in your world the only thing that matters is whether or not you can reproduce completely pointless numbers.

Nvidia's driver -DOES- cheat on visual fidelity. Absolutely yes. But your numbers don't show that fact. Your numbers show hundreds of FPS but not visual quality or user experience at all. Which is -the- reason why they don't mean anything. What is the point in reproducing numbers that don't represent anyones usage?