Steam for Linux > 综合讨论 > 主题详情
Kano 2013年3月21日上午10:32
Killing Floor Benchmark
I recorded some simple demos (could add others or replace em if needed) and made a little script to select em. It always uses the default res (can only parse 1 monitor) and writes that into the default config. In the comments you see what to change if you want to use your standard settings instead. I used 1 level with nvidia rendering bugs, 1 normal speed one, 2 with speed problems. It should be possible to use those demos as win as well (or just record your own) and use in the console:
demoplay <yourdemoname>?timedemo
When you open the console again you see the result. The script does not require that steam runs, but you can change it that even the overlay is active - more realistic but some message windows could pop up. Have fun!

http://kanotix.com/files/fix/tmp/kf/bench-killing-floor.sh
正在显示第 1 - 11 条,共 11 条留言
< >
Kano 2013年3月21日下午6:34 
I created a new variant that just lists the demos you have recorded. As you usually have got timecodes in the name this seems to be simpler than renaming em to use manually. I also changed the default to use the current settings and not to base on the defaults to benchmark the way you just played it. Can be easyly changed, just remove the # for the alternative OPTIONS.

http://kanotix.com/files/fix/tmp/kf/bench-own-demos-killing-floor.bash
Kano 2013年3月22日下午8:11 
As nobody else wanted to post some benchmark results I begin with AMD ones with 4 renderers - 3 on Win and 1 on Linux. On Win it is more easy to access the console as it works anytime, on Linux only when you are in a game. Therefore i used my first script for testing on Linux. On Win i forced Highest mode (default is High on Win) then I checked all 3 possible renderer settings - OpenGL is only possible via direct KillingFloor.ini change. As there is no ~/.killingfloor directory on Win you have to put the Demos inside the main Killing Floor directory. Here are the results:

i7-3770S - AMD HD 5670 - 8 GB RAM - 1920x1200

AMD Catalyst 13.3 Beta 3 - Windows+Linux

Windows 7 - 64 bit
Linux - Kanotix - 64 bit - KDE 4.8 - Kernel 3.8
Wine 1.5.26

demo-bioticslabs

Windows - Direct 3D 9.0 - Highest - 93 fps
Windows - Direct 3D - Highest - 59 fps
Windows - OpenGL - Highest - 110 fps
Linux - OpenGL - Highest - 74 fps
Wine - OpenGL - Highest - 67 fps

demo-hillbilly

Windows - Direct 3D 9.0 - Highest - 45 fps
Windows - Direct 3D - Highest - 55 fps
Windows - OpenGL - Highest - 53 fps
Linux - OpenGL - Highest - 33 fps
Wine - OpenGL - Highest - 36 fps

demo-icecave

Windows - Direct 3D 9.0 - Highest - 78 fps
Windows - Direct 3D - Highest - 57 fps
Windows - OpenGL - Highest - 94 fps
Linux - OpenGL - Highest - 75 fps
Wine - OpenGL - Highest - 60 fps

demo-suburbia

Windows - Direct 3D 9.0 - Highest - 80 fps
Windows - Direct 3D - Highest - 59 fps
Windows - OpenGL - Highest - 63 fps
Linux - OpenGL - Highest - 38 fps
Wine - OpenGL - Highest - 44 fps
最后由 Kano 编辑于; 2013年3月24日下午4:50
Kano 2013年3月23日上午7:53 
So this time I did all tests twice with the fastest Nvidia gfx card I have got. For Win I tested with single and multi gpu, as the OpenGL multigpu results have been slower than single gpu ones I tested with __GL_THREADED_OPTIMIZATIONS=1 on Linux instead a second time:

i7-3770S - Nvidia GTX 295 (single gpu on Linux only) - 8 GB RAM - 1920x1200

Nvidia 316.26 (vsync off) - Linux - __GL_THREADED_OPTIMIZATIONS=1 in ()
Nvidia 314.31 - Windows - multi gpu in ()


Windows 7 - 64 bit
Linux - Kanotix - 64 bit - KDE 4.8 - Kernel 3.8
Wine 1.5.26 - __GL_THREADED_OPTIMIZATIONS=1 in ()

* missing textures on floor
+ stalkers are visable too early - wrong transparency (there are not always stalkers in the demos!)

demo-bioticslabs

Windows - Direct 3D 9.0 - Highest - 73 (135) fps
Windows - Direct 3D - Highest - 59 (45) fps
Windows - OpenGL - Highest - 151 (143) fps *
Linux - OpenGL - Highest - 104 (88) fps *
Wine - OpenGL - Highest - 123 (100) fps *

demo-hillbilly

Windows - Direct 3D 9.0 - Highest - 42 (78) fps
Windows - Direct 3D - Highest - 58 (28) fps
Windows - OpenGL - Highest - 83 (78) fps +
Linux - OpenGL - Highest - 54 (44) fps +
Wine - OpenGL - Highest - 76 (56) fps +

demo-icecave

Windows - Direct 3D 9.0 - Highest - 50 (93) fps
Windows - Direct 3D - Highest - 58 (36) fps
Windows - OpenGL - Highest - 143 (135) fps +
Linux - OpenGL - Highest - 109 (93) fps +
Wine - OpenGL - Highest - 126 (105) fps +

demo-suburbia

Windows - Direct 3D 9.0 - Highest - 54 (102) fps
Windows - Direct 3D - Highest - 59 (30) fps
Windows - OpenGL - Highest - 96 (91) fps
Linux - OpenGL - Highest - 54 (45) fps
Wine - OpenGL - Highest - 82 (62) fps

It looks like the alternative "Direct 3D" setting enforces vsync.
最后由 Kano 编辑于; 2013年3月24日下午4:52
Kano 2013年3月23日下午3:02 
So now I tested the Intel HD 4000, the Direct 3D results are very bad with it, OpenGL (even on Linux) is much faster.

i7-3770S - Intel HD 4000 - 8 GB RAM - 1920x1200

Mesa 9.1 - OpenGL 3.0 - Linux - vblank_mode=0
Intel 15.28.12.64.2932 - Windows

Windows 7 - 64 bit
Linux - Kanotix - 64 bit - KDE 4.8 - Kernel 3.8
Wine 1.5.26 - vblank_mode=0

demo-bioticslabs

Windows - Direct 3D 9.0 - Highest - 20 fps
Windows - Direct 3D - Highest - 34.5 fps
Windows - OpenGL - Highest - 64 fps
Linux - OpenGL - Highest - 54 fps
Wine - OpenGL - Highest - 65 fps

demo-hillbilly

Windows - Direct 3D 9.0 - Highest - 9 fps
Windows - Direct 3D - Highest - 24 fps
Windows - OpenGL - Highest - 32 fps
Linux - OpenGL - Highest - 29 fps
Wine - OpenGL - Highest - 35 fps

demo-icecave

Windows - Direct 3D 9.0 - Highest - 13 fps
Windows - Direct 3D - Highest - 27 fps
Windows - OpenGL - Highest - 48 fps
Linux - OpenGL - Highest - 53 fps
Wine - OpenGL - Highest - 58 fps

demo-suburbia

Windows - Direct 3D 9.0 - Highest - 13 fps
Windows - Direct 3D - Highest - 29 fps
Windows - OpenGL - Highest - 46 fps
Linux - OpenGL - Highest - 37 fps
Wine - OpenGL - Highest - 47 fps
最后由 Kano 编辑于; 2013年3月24日下午4:53
[Linux]sulman 2013年3月23日下午6:22 
Interesting. I've felt KF is a little rough around the edges on Linux (like Red Orchestra) and your experience of Hillbilly mirrors mine - it is manifestly the worst map for performance.

I've got also got an HD4000 in Macbook Pro and I've got to say for an integrated Intel GPU it is absolutely superb.
Kano 2013年3月24日下午5:02 
So, today I tested maybe the most interesting part and used Wine 1.5.26 (with dwrite set to disabled) and use Steam/Windows with Wine. You can easyly see that with 2 exceptions (fglrx on the "faster" levels) Wine was faster. It was even the fastest way to play KF on Intel - the Mesa devs did a great job! Also Nvidia improved very much using Wine, but had of course the same rendering issues. I would say Tripwire should really fix the OpenGL rendering bugs with Nvidia hardware and improve the Linux native client speed. It does not seem to be a driver problem with Nvidia or Intel but definitely a bug in the native code. The startup time was slower with Wine but gameplay is definitely better using Wine than native.
Shark 2013年3月25日上午3:56 
Wow, wine is faster than native code....
[Linux]sulman 2013年3月25日上午6:07 
That's the port. Ryan Gordon's watch...
Turambar 2013年3月25日上午11:06 
Ouch
Kamek 2013年3月25日上午11:39 
引用自 Wouter
Wow, wine is faster than native code....
Yeah, that's crazy oO
I'd like to see some benchmark of Valve's Source based games, I hope they perform better than that on Linux :/
Kano 2013年3月25日上午11:56 
I tested TF 2 with Linux only on the same cards, but due to the size and my small 128 GB ssd i did not install it again on Win. But it is not that hard to benchmark, i wrote that in another thread already how it works. The only problematic point is that there is no specified way to use default settings, you have to check em manually. KF has not that much settings, when you set it to Highest then everything should be similar. Interestingly the default on Linux is Highest and the Windows the default ist High. At least for Intel users OpenGL should be selectable via the GUI on Windows too and the stupid Direct 3D selection for the Linux binary should be removed.
正在显示第 1 - 11 条,共 11 条留言
< >
每页显示数: 15 30 50
发帖日期: 2013年3月21日上午10:32
帖子数: 11