The results to share this afternoon are tests I did on a variety of Radeon HD 6000 series through Radeon Rx 200/300 series hardware on the open-source driver. Coming tomorrow as a featured article will be these open-source results compared to the Catalyst Linux driver performance!
These tests were done on the HD 6870, HD 7950, R9 285, R9 290, and R7 370. Unfortunately with this build from the Padoka PPA, the OpenCL support for Fiji wasn't recognized so I couldn't run any R9 Fury test. Also, my HD 6950 Cayman tests were foiled due to an error (unsupported call to function __floatunsidf in compute_dp_v1). Aside from that, this open-source comparison was basically limited to the AMD discrete GPUs I had available for testing.
With the results from this morning's presentation slides being done using OpenCL clpeak, I did tests there too. I ended up writing a new test profile system/clpeak for allowing the clpeak benchmark if installed on the system to be driven via the Phoronix Test Suite. This is another example of an easy to craft test profile where now simply running phoronix-test-suite benchmark clpeak is able to automate this OpenCL benchmark.
Some of the SHOC benchmarks were also able to run on the open-source driver stack for today's testing. Unfortunately, our other common OpenCL tests like LuxMark, JuliaGPU, SmallPT-GPU, etc, still fail currently on the open-source Clover Gallium3D driver stack.
The R9 290 performed particularly well compared to the other GPUs on this Clover-based OpenCL stack with Mesa 11.2-devel + LLVM 3.9 SVN and riding off the Linux 4.5 kernel.
What will be most interesting though is how these numbers compare to Catalyst...
Stay tuned for the open vs. closed-source AMD OpenCL benchmarks tomorrow! If you want to see how your own system performs, simply install the Phoronix Test Suite and run phoronix-test-suite benchmark 1602020-GA-OPENCLRAD51. Via this OpenBenchmarking.org result file are a few more open-source R600g/RadeonSI test results!