AMDGPU Linux 4.5 DRM Tuning Tests With DRI2/DRI3, PowerPlay, Semaphores, Scheduler
Complementing yesterday's AMDGPU tests with the new DRM-Next code that has PowerPlay support where the speed of this latest open-source driver code was compared to the proprietary driver, here are some tests showing the AMDGPU driver performance under a few different scenarios.
Using the Radeon R9 285 (Tonga) graphics card were extra runs in the following conditions:
- The system with the AMDGPU DRM-Next code for Linux 4.5 and Mesa 11.2-devel when there were no out-of-the-box tweaks (sans the always doing no swap buffers wait)... As mentioned already, with Linux 4.5 the PowerPlay support isn't enabled by default. So this is basically the out-of-the-box reference run.
- The system when PowerPlay was enabled (with having the appropriate PowerPlay Kconfig option enabled and booting the kernel with amdgpu.powerplay=1) while leaving the rest to the defaults.
- The prior PowerPlay-enabled configuration, where DRI2 is the default, but instead switching over to DRI3. As shown by previous tests, in some environments DRI3 can be much faster than DRI2.
- The PowerPlay-enabled, DRI3-using configuration from above but enabling semaphores. Semaphores aren't enabled by default in the AMDGPU driver but require setting amdgpu.semaphores=1.
- The PowerPlay-enabled, DRI3-using configuration while disabling the new AMDGPU scheduler that was recently turned on by default. The scheduler can be disabled via the amdgpu.enable_scheduler=0 module parameter.
- Finally was the PowerPlay-enabled, DRI3 configuration while increasing the sched_jobsfrom 32 to 64. This was just done for testing as I haven't seen any information lately whether increasing the maximum number of jobs for the software queue would have any impact on the performance.
The rest of the system remained the same during this AMDGPU testing driven by curiosity.
All of the benchmarks were done using the open-source Phoronix Test Suite benchmarking software.