WineD3D Benchmarking Part 3: The Results

At long last, the results are in!

http://openbenchmarking.org/result/1307214-AH42-WINE16941

What can we tell from these?

Firstly, and most obviously, there is a large drop in performance for 3DMark2001 and ShaderMark between wine-1.5.8 and wine-1.5.9. This is not a real performance drop: as I noted in my previous post, prior to wine-1.5.9, XRandR support was incomplete, and some applications (i.e. 3DMark2001 and ShaderMark) were unable to render at 1920×1080.

OpenGL Tests

Furmark: flat-line consistency. OpenGL performance has not changed at all through the entire cycle. Not surprising, but good to know.

UnigineHeaven (openGL): Okay, it’s possible OpenGL performance has changed a little, but we’re back where we started.

Lightsmark (OpenGL): fairly steady, but a slight improvement

Xonotic: flat-line OpenGL performance again

 

Overall, OpenGL performance is unchanged, and is not likely to change much, either. OpenGL calls can be passed through natively. In fact, the wine D3D implementations convert everything to OpenGL calls anyway.

 

DirectX/D3D Tests

UnigineHeaven: nice 10% improvement around wine-1.5.4. Overall 13% improvement over wine-1.4. At no point is this faster than the OpenGL version, but towards the end it gets very close.

streamsrc / drawprim / clear / dynbuffer: I don’t pretend to know what they test, but except for dynbuffer, they all imrpoved. dynbuffer was flat-line.

3DMark2001: apart from the resolution recognition, there wasn’t much change here

ShaderMark: same as 3dmark, but with a 5% boost around wine-1.5.29. Interesting to note that running with PixelShader3.0, the test performed 2% slower than with PixelShader2.0. My understanding is PS3.0 is not fully implemented yet, and so many of the calls fall-back to PS2.0, anyway.

Half-Life2 (dx 8.1):  Slightly decreasing performance over the series.

Half-Life2 (dx 9.5): Also slightly decreasing, but definitely faster than with directx8.1

TrackMania: slight decrease, but an improvement in the fact that it wouldn’t give results prior to 1.5.16!

 

Results Summary

Some things improved, some didn’t, and some got worse. Overall, it probably depends on what application you’re using, but the improvement in XRandR support and many other things suggest that everyone should be using at least 1.5.9.

Comments are closed.