NVIDIA vs. AMDGPU-PRO vs. RadeonSI OpenGL Comparison For Holidays 2016

[M@yeulC]

Interestingly, the open driver results seem to be missing for some cards. Did you encounter some problems, or did you just skip these tests?

Missing cards (sorry for the abbreviations):
R9 290 Open on Bioshock infinite
460, 290, on CS:GO
the same on DX:MD, ET:L, M:LL, Tesseract, Xonotic; as well as Unigine Heaven, where the 290 on the PRO driver isn't here as well.

Interesting results. I hope marek will receive his copy of DX:MD from bridgman soon enough! [1]

[duby229]

i was talking about performance difference between today and years ago.
AMD is working very good and the open source driver is good, nvidia has better performance, but AMD is improved a lot

Very good point, and the fact is AMD's OSS driver tries to be very strict with standards (too strict in fact), and nVidia doesn't give two tiny little shits about standards in any form.

[mannerov]

You either have a funny definition of "really good" and "amazing" or you can't see the green bars

The Green bars are for Nvidia cards of the last generation. The red bars are mostly from old gen, but they match the expected performance ratio. As you can see The RX480 matches the 1060, as is expected.

[iznogood]

Still not clear why so much difference between AMD and Nvidia. Any clues?

[dungeon]

it's a pity that mobile APU performance are really bad, seems that they are optimizing only for desktop GPU and not for mobile hardware

Well if AMD sell less APUs in future, we will know a reason why

Yeah, APU does not see these improvments AT ALL as they are CPU bound much earlier... and does not run 4K so you can't see anything there

[duby229]

Still not clear why so much difference between AMD and Nvidia. Any clues?

No clues? Really? Nvidia doesn't care one tiny little bit about specifications. They have a history of cheating on performance that goes back dozens of situations and decades of time. AMD's driver is very strict on adhering to standards, and as such it exposes bad design choices in OpenGL and many many bugs related to non-compliance in games and applications.

[mannerov]

Still not clear why so much difference between AMD and Nvidia. Any clues?

cards of last gen tested against mostly cards of previous gens

[duby229]

No clues? Really? Nvidia doesn't care one tiny little bit about specifications. They have a history of cheating on performance that goes back dozens of situations and decades of time. AMD's driver is very strict on adhering to standards, and as such it exposes bad design choices in OpenGL and many many bugs related to non-compliance in games and applications.

EDIT: I have no doubt at all the overall situation will improve dramatically once Vulkan titles proliferate.

[mannerov]

No clues? Really? Nvidia doesn't care one tiny little bit about specifications. They have a history of cheating on performance that goes back dozens of situations and decades of time. AMD's driver is very strict on adhering to standards, and as such it exposes bad design choices in OpenGL and many many bugs related to non-compliance in games and applications.

I don't think that explains a potential performance gap.

As we can see from previous performance improvements from the open source driver, games are doing things that are not always optimal, like computing shader outputs that are not used later. By having the driver cleaning the stuff, you get performance improvements. NVidia drivers likely do these tricks from a long time, and probably have more. There's also multithreading that helps when cpu bound. But it's only a matter of time now that the open source drivers get these tricks too.

[dungeon]

Still not clear why so much difference between AMD and Nvidia. Any clues?

If you are asking about AMDGPU-PRO perfromance, they still has performance regression going from FGLRX time

While opensource driver as always is never better on paper, but on reality there is bugzilla