The Perf-Per-Watt Of NVIDIA Fermi To Pascal, AMD R700 To Polaris With Newest Linux Drivers

[phoronix]

Phoronix: The Perf-Per-Watt Of NVIDIA Fermi To Pascal, AMD R700 To Polaris With Newest Linux Drivers

Unless you want your graphics card to keep you warm this winter, here's a big comparison of AMD Radeon and NVIDIA GeForce graphics cards under Linux looking at their performance-per-Watt using the latest OpenGL Linux drivers as of the end of 2016. A few days back I posted a 31-way GeForce/Radeon Linux comparison looking at the raw performance with each company's latest Linux drivers going back to the Fermi and R700 days while for this article is looking at the system power consumption and power efficiency for this mass assortment of GPUs.

http://www.phoronix.com/vr.php?view=23916

[bug77]

Hint to Michael: when you have "perf-per-watt" in an article title, maybe tests should lead with fps/W graphs instead of burying them towards the end of each page.

[Michael]

Hint to Michael: when you have "perf-per-watt" in an article title, maybe tests should lead with fps/W graphs instead of burying them towards the end of each page.

It's not really buried at all, simply showing the raw results so people have an idea and then it goes to show the perf-per-Watt.

[tildearrow]

Next up is the demanding Unigine Heaven game.

Wait. Is Unigine Heaven a game?

[GraysonPeddie]

Thanks for not including benchmarks for The Talos Principle. That game has an auto-detection script that makes GTX 960 seem much worse compared to GTX 970.

[Michael]

Thanks for not including benchmarks for The Talos Principle. That game has an auto-detection script that makes GTX 960 seem much worse compared to GTX 970.

There test profile overrides that script

[bug77]

It's not really buried at all, simply showing the raw results so people have an idea and then it goes to show the perf-per-Watt.

It's the third graph, which is either last or right in the middle. Not the best spot for the primary aspect of the review, wouldn't you agree?

Also, Merry Christmas everyone!

[johnc]

Also, Merry Christmas everyone!

Merry Christmas!

[QuImUfu]

I think 3-4% more CPU usage on this high end CPU is a huge amount.
Theoretically it would be ca. 23-30% (according to synthetic CPU benchmarks the CPU Michael uses is 7,5 x faster than mine) on my low end and old CPU. (if my CPU would be good enough for any of these games/benchmarks )

[Marc Driftmeyer]

Still waiting to see that OpenCL Mesa bench test for AMD Polaris series and not the AMDGPU-Pro drivers. Wake me when AMD actually fixes their broken OpenCL with LLVM's own broken libclc that still fails even after LLVM 3.9.1 was released leaving LuxRender, Blender and presumably Darktable 2.2 useless with Polaris GPGPUs.

Or thrill no one with showing more pointless AMDGPU Pro benchmarks on Ubuntu 16.04 or SEL/Redhat/CentOS with Linux 4.7: something even Debian no longer supports outside of Stable.