Latest AMDGPU-PRO Ubuntu Linux Performance vs. NVIDIA, Including The GTX 1080

[bug77]

So if I read the graphs right, the new driver yields about the same results as the old driver?

[cRaZy-bisCuiT]

I do hope Metro will be fixed soon. That kind of Game is already much better for a real world comparison.

At the moment I don't know how much left behind an AMD card that is considered to be equal to a certain nVidia card in reality.

[Michael]

I do hope Metro will be fixed soon. That kind of Game is already much better for a real world comparison.

At the moment I don't know how much left behind an AMD card that is considered to be equal to a certain nVidia card in reality.

Feel free to comment your support and get others to as well on: https://steamcommunity.com/app/28739...4767232116861/ so that the developers will hopefully fix the game.

[bridgman]

So if I read the graphs right, the new driver yields about the same results as the old driver?

What do you mean by "new" and "old" here ? If you're saying that the amdgpu hybrid performance is about the same as fglrx, that makes sense because the userspace drivers are pretty much the same - what's different is that the rest of the stack is now based on upstream open source code.

If you're saying that amdgpu hybrid GL performance is about the same as Mesa, it's getting there but that's fairly recent and a bunch of people worked hard to get it there.

[bug77]

What do you mean by "new" and "old" here ? If you're saying that the amdgpu hybrid performance is about the same as fglrx, that makes sense because the userspace drivers are pretty much the same - what's different is that the rest of the stack is now based on upstream open source code.

If you're saying that amdgpu hybrid GL performance is about the same as Mesa, it's getting there but that's fairly recent and a bunch of people worked hard to get it there.

I was looking at year old benchmarks here on Phoronix and in the tests that were run on both occasions (Heaven/Valley and LuxMark), the AMD cards score the same. Iirc, one of the goals of AMDGPU-Pro was to close the gap on Nvidia. It's pretty disappointing (to me, at least), to hear you say that after all the work, the driver is about the same and no performance differences are to be expected

[Nille_kungen]

I'm finally getting a chance to get back to hands-on development work but still very rusty.

Welcome back then code monkey

[bridgman]

I was looking at year old benchmarks here on Phoronix and in the tests that were run on both occasions (Heaven/Valley and LuxMark), the AMD cards score the same.

OK, so fglrx right ?

If you look at the LuxMark results we're already very competitive (a non-X Fury is trading blows with 1080) so not sure why you're disappointed that nothing has changed there. I don't think demo programs are very high on the list for performance work although I expect that work done for other apps will result in some improvements there as well.

BTW when I look at price/performance we already seem to be roughly in line on Heaven/Valley anyways, aren't we ? The 1080 skews the numbers a bit but so does the RX 480.

Iirc, one of the goals of AMDGPU-Pro was to close the gap on Nvidia. It's pretty disappointing (to me, at least), to hear you say that after all the work, the driver is about the same and no performance differences are to be expected

If you mean "close the gap" w.r.t. performance I don't think that was ever a goal of the amdgpu initiative itself, although there are other initiatives (eg driving work on Vulkan & similar APIs, working with game devs on GL/Vulkan performance for example) which are doing that. Neither of those would affect Heaven/Valley or LuxMark though.

The amdgpu initiative was about bringing all the things people liked about the open source stack to the proprietary stack, and allowing us to have both teams contributing to common code rather than having separate open/closed stacks.

[Marc Driftmeyer]

Something is seriously wrong if Nvidia is dominating in OpenCL.

[bridgman]

Something is seriously wrong if Nvidia is dominating in OpenCL.

They're not though... look at Luxmark which is more of a real world app, where a non-X Fury is competitive with a 1080. The amdgpu stack is pretty new, don't think anyone has had time to do much performance tuning for synthetic benchmarks yet. I'll ask though...

[oooverclocker]

I more-or-less stopped looking at the Steam stats when I realized they didn't even include SteamOS and that (IIUC) games played under Wine or similar count as Windows not Linux.

I know but you don't develop games so this won't help no one to get rid of GameWorks neither convincing people to use Vulkan.

I had assumed that 3xx cards were being lumped in with 2xx but if 3xx was just being dropped that would help to explain some of the numbers we're seeing.

That's a logical assumption if you feel a bit lazy. Please excuse me being so inconvenient - I just mean well for you - but I don't think that you can afford to just make assumptions. The Fury isn't represented as well - where should it be? It would really make no sense to put it under R9 200 series while even the HD 8800 series is counted separately - being in use by 1.5 percent??? So more than the HD 7900 and nearly the same like the R9 200?
No matter if it is somewhere counted or not - if this would have prevented CD Project Red to use Gameworks for Witcher III or preventing id Software from not showing options in the beta so you just could react after the release - again, it would help you so much more than 1% more powerful cards...
You have to accept that there is a survey but you have the power to influence it. If you don't care how your firm is represented by others while I could imagine that Nvidia has a full time guy only looking that they get high counts in the survey, you have a great disadvantage. The standard answer if something is buggy with AMD cards in game developer forums(by other forum members) is: "Look at the market share in the steam survey."