Linux Benchmarks Of NVIDIA's Early 2015 GeForce Line-Up

[phoronix]

Phoronix: Linux Benchmarks Of NVIDIA's Early 2015 GeForce Line-Up

Following yesterday's NVIDIA GeForce GTX 960 launch, here's an 11-way comparison looking at all of NVIDIA's Maxwell GPUs as well as many Kepler and Fermi GeForce graphics cards under Linux. Beyond the raw OpenGL performance, the thermal and power efficiency data is also available for the tested range of GeForce 900/700/600/500/400 series graphics cards.

http://www.phoronix.com/vr.php?view=21400

[peppercats]

Any suggestion on which gtx 970 to buy?

[Licaon]

The linked test ( http://openbenchmarking.org/result/1...DE-PERF9600864 ) does not have the Metro 2033 Redux test.

[Michael]

The linked test ( http://openbenchmarking.org/result/1...DE-PERF9600864 ) does not have the Metro 2033 Redux test.

That's just the result file with the perf-per-Watt and GPU temp metrics.

[Licaon]

I was curious to compare my 660 Ti, yet while Unigine Valley yields the results as expected ( 5% slower than 760 and 15% slower than 960 ), the Metro benches are rather wierd.

First of all the whole bench limitation in Metro is rather meh because I'm not sure what setting your test used, say if the game was never launched, and settings never changed, there isn't any user.cfg in steam/common/Metro/<some-hash>/ so a default one is copied, now changing settings in-game will change them here too.

Now for my results, I can force a desired resolution (my default is 1920x1200) from the control panel, either a smaller one (say 1920x1080) or a larger one ( say 2560x1600 or even 3840x2400 via Advanced->Viewport In/out/Panning) that gets downscaled via de monitors panel (I believe) since there is no DSR/GPU scaling option like in the Windows driver. And the bench will use it (the screen gets resized and you can pan around with the mouse) but considering the performance penalty I believe the game is rendered at the desired resolution although the game does not use the resolution as it should.

For the results, you say in the text that the game is run at 2560x1600:

Metro 2033 Redux
750 Ti: 53.15
my 660 Ti: 47.96
my 660 Ti@1920x1200: 56.67
760: 73.11

Metro Last Light Redux
750 Ti: 46.47
my 660 Ti: 25.98
my 660 Ti@1920x1200: 32.67
760: 68.99

Can you post each user.cfg ?

Is the game so much more CPU dependant that my 4 core i5-4570 slows down the GPU ?

[Michael]

I was curious to compare my 660 Ti, yet while Unigine Valley yields the results as expected ( 5% slower than 760 and 15% slower than 960 ), the Metro benches are rather wierd.

First of all the whole bench limitation in Metro is rather meh because I'm not sure what setting your test used, say if the game was never launched, and settings never changed, there isn't any user.cfg in steam/common/Metro/<some-hash>/ so a default one is copied, now changing settings in-game will change them here too.

Now for my results, I can force a desired resolution (my default is 1920x1200) from the control panel, either a smaller one (say 1920x1080) or a larger one ( say 2560x1600 or even 3840x2400 via Advanced->Viewport In/out/Panning) that gets downscaled via de monitors panel (I believe) since there is no DSR/GPU scaling option like in the Windows driver. And the bench will use it (the screen gets resized and you can pan around with the mouse) but considering the performance penalty I believe the game is rendered at the desired resolution although the game does not use the resolution as it should.

For the results, you say in the text that the game is run at 2560x1600:

Metro 2033 Redux
750 Ti: 53.15
my 660 Ti: 47.96
my 660 Ti@1920x1200: 56.67
760: 73.11

Metro Last Light Redux
750 Ti: 46.47
my 660 Ti: 25.98
my 660 Ti@1920x1200: 32.67
760: 68.99

Can you post each user.cfg ?

Is the game so much more CPU dependant that my 4 core i5-4570 slows down the GPU ?

As Metro doesn't allow modifying the graphical settings from the CLI interface, for my testing I always start with the default / fresh install of the game from scratch whenever starting a new round of testing.

[Licaon]

trimmed results of pts benchmark 1501235-DE-PERF9600864

https://i.imgur.com/inlb7ds.png