OpenGL Performance & Perf-Per-Watt From The Radeon HD 3850 Through R9 Fury

First of all, the tests done for this article were done at 1920 x 1080 max in order to go back and test the old R600 era hardware. So for some of the tests there are some CPU-bound results with the newer graphics hardware, but with complementing that data by the power consumption and performance-per-Watt data makes this rather interesting. First up is Valve's Team Fortress 2.

TF2 is one of the examples where basically all of the AMD GCN GPUs are around being CPU bound. (Just so happens to be 120 FPS, but no vsync'ing or other factors were done during testing, PTS enforces FPS max of 0.) Anyhow, even from this data it's fun to see how much faster modern Radeon GPUs are compared to old hardware.

A look at the TF2 power usage during testing...

The performance-per-Watt is what makes this article more fun and for here we see how going from a Radeon HD 4870 to Radeon R9 Fury is more than three times more efficient. The top performer here was the R7 260X with the best performance-per-Watt.

Dota 2 with its OpenGL renderer also becomes rather CPU bound at 1080p, but the performance-per-Watt puts it into greater perspective for how the performance has evolved.

The R9 Fury is eleven times more power efficient than the least efficient card tested (Radeon HD 3850). Going from all of the pre-GCN GPUs to the more modern AMD GPUs represent big wins for this Linux 4.6 + Mesa 12.1-dev driver stack.