22-Way Comparison Of NVIDIA & AMD Graphics Cards On SteamOS For Steam Linux Gaming

[Michael]

Welp. Here it comes.

Thanks!

[dungeon]

Read their reviews and you will know. The [H] lists the driver versions used as well as the settings. They do apples to apples testing, where the game settings are the same as well as adjusting the settings to the highest playable for the cards involved. Since you mention Witcher 3 here's the gameplay review they did that included 9 cards from AMD and Nvidia.

You might also want to read http://www.hardocp.com/article/2008/...g_benchmarks/1

Ah OK, nice reviewing there it seems... so render gets improved by game patching after couple months, cool.

[asdfblah]

So, people in this very same thread are saying that the CPU is the limiting factor on AMD cards... Is multithreading the reason nVidia performs better? Can PTS monitor/record CPU usage?

[Michael]

So, people in this very same thread are saying that the CPU is the limiting factor on AMD cards... Is multithreading the reason nVidia performs better? Can PTS monitor/record CPU usage?

It's as easy as setting MONITOR=cpu.usage environment variable prior to running tests to get CPU usage graphs for each test. Last time I did such tests, IIRC, it was less than a 10% difference between Catalyst and NVIDIA blob....

[dungeon]

So, people in this very same thread are saying that the CPU is the limiting factor on AMD cards... Is multithreading the reason nVidia performs better? Can PTS monitor/record CPU usage?

Maybe with some earlier driver , but all those games Michael benchmarked now with Catalyst 15.9 are fine threaded i think, either by engine or by profile. Dirt is one that is already threaded, all other runs some driver profile on the matter.

For Bioshock i mentioned it can be faster with game switch to disable workarounds needed on earlier drivers... And i think CS:GO should be already faster then benchmark shows, but not sure if something changed recently for that game

[humbug]

So, people in this very same thread are saying that the CPU is the limiting factor on AMD cards... Is multithreading the reason nVidia performs better? Can PTS monitor/record CPU usage?

I don't know but if you look at Michael's power consumption figures for the R9 290 under Linux gaming loads you will notice that the GPU is consuming much less power than expected. Less wattage than it consumes under a Windows gaming load. This is a sure sign that the GPU is not being utilized properly, not being worked hard, not being fed due to some other external bottleneck.

[faldżip]

It would be better to see tests on some i7 6700k or 4790k OCed to 4.5GHz or even i5 6600k (also 4.5GHz) as they are what typical gamer has, and also games like faster cores and usually do not benefit from more cores. And almost nobody has 8c/16t processor. I undestand that when testing GPU we don't want tests limited by the CPU, but the results seems exactly like that (sometimes GTX 960 is really close to Titan X while it is 5x cheaper). Maybe because they are ran in games' unusual environment.

[artivision]

All the facts about bad Catalyst are nonsense. The truth is that RadeonSI performs as good as Catalyst and Nouveau with re-clocking as good as Nvidia blob. The difference is that Nvidia has 2x the OGL frame rate with the same class hardware, same horsepower. The problem here is that this difference is also there on RadeonSI vs Nouveau. Those drivers they use the same OGL program (state tracker), the same Intermediate program (Gallium, TGSI), the same Compiler program (LLVM), the only difference is the HW program. So just all of you accept that Radeon Gpu (big units) is not good for High level Apis where data are heavily depended between them but will be grate for Low level Asynchronous data. And anyway we have Native D3D9 and in a wile D3D11 and Vulcan (even D3D12 over Vulcan will be here faster than you think), so just by a Radeon R7-260 (2Tflops) and use RadeonSI and get done with it.

[duby229]

if it works perfectly fine on Nvidia, then it has nothing to do with eON.
How about AMD gets its shit together?

That's just not true. Nvidia's driver allows behavior that is out of spec and eON is especially guilty of not being careful enough to avoid those conditions.

[duby229]

Yes and some people will look at these numbers and say not to use catalyst. But the sad part is the open source drivers are even slower, if they can run the game at all...

Is there a chance that AMD's openGL performance will ever be fixed (as in competitive with their directX performance)?
Or are we beating a dead horse? I know they will do great with Vulkan but openGL is not going away...

Your info is just old. Very new sources can outperform Catalyst in many cases. Soon enough it will make it to stable releases. I can't wait until newer distro's get released with benchmarks available that show it clearly.