Nouveau With Boost Patches Are Now Competitive To Radeon/AMDGPU With RadeonSI

[bug77]

So, whether AMD throws its weight behind open source or Nvidia releases some standard docs, we still end up in the same place.
Of course, proper reclocking is not mainlined yet, but hey, it's getting there.

[Ancurio]

Yeah right, but to stay a benchmark is garbage (and that's what he tries to pull), just because high resolutions were used, which are actually used, then he clearly just wants to try to find _any_ reasons why this benchmark is stupid and doesn't fit in his view of the world (most likely)

Where did I use the word "garbage", or even imply that the benchmark was useless? All I was saying was to take the results with a pinch of salt. Just because the FPS counts are close doesn't mean that the drivers perform at a similar level. Although the shader compilers must be roughly on par.

[karolherbst]

Where did I use the word "garbage", or even imply that the benchmark was useless? All I was saying was to take the results with a pinch of salt. Just because the FPS counts are close doesn't mean that the drivers perform at a similar level. Although the shader compilers must be roughly on par.

Yeah I know I overreacted a bit, sorry for that. It is just that your phrasing wasn't exactly the best and that your comment is only true for a few of those benchmarks. pixmark_piano for example only needs fast shader cores, everything else is no problem (60 gl calls per frame... and still only performs at < 10fps at this resolution, also no significant memory access or texture loads or anything)

[bug77]

Where did I use the word "garbage", or even imply that the benchmark was useless? All I was saying was to take the results with a pinch of salt. Just because the FPS counts are close doesn't mean that the drivers perform at a similar level. Although the shader compilers must be roughly on par.

Funny, I thought it means precisely that, if the image quality is the same (which seems to be the case here).

[smitty3268]

Funny, I thought it means precisely that, if the image quality is the same (which seems to be the case here).

Ancurio's point is that the benchmarks shown here may not be representative of real world performance, and he went on to explain why he believes that may be the case. You guys are blowing it way out of proportion.

[karolherbst]

Ancurio's point is that the benchmarks shown here may not be representative of real world performance, and he went on to explain why he believes that may be the case. You guys are blowing it way out of proportion.

which is also true for every benchmark anyway. His post just has a big trolling character due to the words he used.

But Nouveau has a big performance problems with all eon based games, for some reasons they perform just bad and bioshock is one of the best here.

But I would also wished for a few more complex stuff, because I am sure that's where the performance is bad currently. It is nice to have some competitice results in micro benchmarks and not complex games.

[theriddick]

Looks good except for the Bioshock scores, seems they are having a issue there?

Soon it looks like AMD won't be even leading in Open-source department.... Wonder if they can add freesync to this open-source driver? that would be AMAZING!

[smitty3268]

Soon it looks like AMD won't be even leading in Open-source department....

Lets wait until they actually have all their cards working with automatic reclocking support before going there. I think it's still a long way off.

[Luke]

I've known for years this was possible, ever since the days of non dynamic power management on Radeon. I tested my HD6750 on mid clocks then on Scorched3d and Critter, then switched to the GTS450 "nouveau card" and which boots to mid clocks and got very similar results. Both cards ran at about 40 C at idle too. Dynamic clocks for Nouveau on Fermi will cause me to bench both cards and run whichever is faster, sending the other to my #2 system.

[VikingGe]

Soon it looks like AMD won't be even leading in Open-source department....

Except that it took Nouveau like 4 years to get Kepler working properly - we aren't even talking about Maxwell here. And since I don't see NVGPU (or actually any kind of official support) happening any time soon, Nouveau will always be at a disadvantage.