Michael: If you get a chance, throw in a low end Radeon card as well. And some performance per dollar and performance per watt graphs would be cool.
I appreciate your benchmarks
But again, the point is to see how much better each driver utilizes the GPU's power here .. I'd say Catalyst(as expected) wins this one ..
Of course there's tweaks to make things faster, but not everyone tweaks/wants to tweak their drivers, and mostly they don't always work well or at all and can potentially cause problems(that's why they're not tweaked by default).
Was the LLVM compiler enabled? I'd like to see a test on r600 with and without it.
I have Intel X4500HD. This week I get Intel HD 4000 graphics.
X4500HD served fine, except not that fine in some 3D games. HD 4000 will serve very fine even in 3D games I play (in maybe about less than once a month), which are Free Software.
Could not be more happy supporter of Intel
I look down on "me no can reveal docs nor source because me business sucks so much me would be ruined if me's competitors would get that information".
If NVIDIA or AMD can't free the source because it would affect their business, then they deserve to disappear from the face of earth because their business just sucks big time. They are hardware manufacturers to say the least.
I don't like intel but ever since I switched to intel gfx this shit has been silky smooth
Please read the text:
AFAIK, OSS driver does not switch performance levels by default so the card would be running on default freqs. The specs table also says that the driver reports OpanGL 2.1 so any advantages of opengl 3+ are gone. I don't know how much work canonical backports to it's kernels, but things like pcie2.0 were enabled in 3.6 and the test uses 3.5.The stock open-source Radeon Linux graphics driver stack was used with the Unity desktop while the only change was disabling swap buffers wait from the xorg.conf.
All in all, OSS driver was in pretty bad position for this test. IMO the stock configuration is supposed to be stable first anything else second, and all tests ran without lockup or crash (at least none are reported).