Via OpenBenchmarking.org for the Phoronix Test Suite is the Dota 2 test profile that can be automated and doesn't run into any automation issues. So first on my agenda, of course, was seeing the latest performance of the open-source driver stack, including RadeonSI Gallium3D paired with the new AMDGPU DRM driver for Fiji and Tonga.
For this initial comparison I did tests with RadeonSI using a Radeon HD 7950, R9 285, R9 290, R7 370, and R9 Fury. For R600g coverage was the Radeon HD 6870 and HD 6950 as well as the HD 5830. Tests for this posting are just of the open-source AMD Linux driver stack... Still working on the binary blob tests.
I tested the cards at a variety of resolutions along with some PTS system sensor monitoring at 4K. Via the OpenBenchmarking.org result files linked at the end of the article you can explore these results in more detail.
At 1080p, the tested GCN GPUs on RadeonSI were certainly playable.
The older R600g cards are only playable when dialing back the resolution; see the OpenBenchmarking.org result file for more of the resolution tests.
At 4K, only the R9 290 and R9 Fury were scoring well with Dota 2.
Then I was doing a second run later on with the system monitoring results with the GCN GPUs at 4K and I noticed something...
Compare the R9 Fury result from this graph to the same 4K result in the previous page... See how this one is much better? It was all the same resolution, same test, same automation, etc. It's nothing to do with the test. I think I discovered one of the AMDGPU bottlenecks / issues! Stay tuned for another post shortly after I finish my investigation ;)
Of course, if you enjoy these fresh open-source Linux driver game benchmarks of Dota 2 and the recent Talos Principle benchmarks, consider doing your share by taking advantage of our Valentine's weekend deal.
You can explore this performance data in more detail via the multi-resolution test results and the 4K system monitoring test data. Or explore more in general via the new (still 2.0 beta) OpenBenchmarking.org.