The OpenGL Speed & Perf-Per-Watt From The Radeon HD 2000/3000 Series Through The R9 Fury

With the fourteen graphics cards used, they were tested on a Core i7 5960X system with the Gigabyte X99-UD4-CF motherboard, 16GB of RAM, and 120GB Samsung 850 EVO SSD.

Ubuntu 15.10 x86_64 was installed on the system while I manually upgraded to the very latest stack possible. That stack included Mesa 11.2-devel Git master and LLVM 3.8 SVN from the Padoka PPA as of this past weekend. On the kernel side, I was using the DRM-Next driver code for the Radeon/AMDGPU kernel drivers that will land in the Linux 4.5 kernel. The Padoka PPA provides the xf86-video-ati 7.6.99 and xf86-video-amdgpu 1.0.99 DDX drivers.

Additionally, when it came to testing the R9 285 and R9 Fury with that Direct Rendering Manager driver being AMDGPU, I used a kernel with the PowerPlay support enabled and amdgpu.powerplay=1 from the kernel command-line in order to support power management / re-clocking with these very latest AMD GPUs. Unfortunately for Linux 4.5, this PowerPlay support is still being disabled by default.

Lastly, DRI3 (Direct Rendering Infrastructure 3) was enabled during all of this benchmarking.

The Phoronix Test Suite was used for carrying out all of the OpenGL benchmarking on these 14 different GPUs. For the performance-per-Watt and system power consumption monitoring, the Phoronix Test Suite benchmarking software was interfacing with a USB-based WattsUp Pro power meter for polling the AC system power consumption in realtime.

For the runs with BioShock Infinite, Counter-Strike: Global Offensive, Furmark, Metro Last Light Redux, OpenArena, Tesseract, Team Fortress 2, and Xonotic, the monitor's native resolution of 2560 x 1600 was used in order to keep the newer graphics cards as busy as possible without these GCN GPUs becoming CPU bound while a 2560 x 1600 resolution was still supported by the older RV600 GPUs.

With all of the bases covered, let's see how the performance goes from the Radeon HD 3650 through the the R9 Fury!