If users don't add "vblank_mode=0" when they run their glxspheres tests, they might be getting strange results, as the GPU will try to match the screens refresh rate.
When I first ran my test, I thought the new primus backend was only as good as the Intel card, but when adding the "vblank_mode=0" it really showed how much better than virtualgl it was! - great job guys!
This is mentioned on primus' github page. :)
As well as why you shouldn't read into it too much. Oddly, with glxgears I got higher numbers with optirun -c yuv glxgears than primus, though when I played games primus' performance was visually much better.
I'll add that primus could be optimized better for disabled vsync, and VirtualGL does not synchronize frame transfer by default: try VGL_FPS=1 optirun glxgears ;)
VGL_FPS=1 optirun glxgears
The best benchmarks are the games you play, after all.
I just meant that when you tell people to test the performance difference by running a benchmark like glxspheres, like on the ubuntu ppa, it would be nice to add the "vblank_mode=0" info. Else the point of the benchmark would be lost, as the FPS would be limited to the screens refresh rate (in some cases).
This instructions are like this since VGL is still the default. We might consider reworking the doc on those things when primus will be widely used or being the default.