16-Way GPU OpenCL Showdown With Radeon ROCm, NVIDIA 384 On Linux

With earlier this week having delivered our very latest NVIDIA GeForce vs. Radeon OpenGL/Vulkan graphics benchmarks, I have now finished up the very latest OpenCL Linux compute numbers for both vendor's graphics cards on their latest drivers. Radeon cards were tested with their latest ROCm packages on Ubuntu 16.04 LTS while on the NVIDIA side was their long-lived 384.59 driver release.

For those curious about the latest Radeon vs. GeForce OpenCL performance (unfortunately no workstation cards for testing today), here is a big batch of the latest performance numbers to complement yesterday's Ethereum Ethminer GPU benchmarks in the same configuration. The hardware tested that was limited by what I had available and also what was supported by the Radeon ROCm driver included the:

- Radeon R9 285
- Radeon R9 290
- Radeon RX 480
- Radeon RX 560
- Radeon RX 580
- Radeon R9 Fury
- GeForce GTX 780 Ti
- GeForce GTX 960
- GeForce GTX 970
- GeForce GTX 980
- GeForce GTX 980 Ti
- GeForce GTX 1050
- GeForce GTX 1060
- GeForce GTX 1070
- GeForce GTX 1080
- GeForce GTX 1080 Ti

All tests happened from the Intel Core i7 7740X box with the ASUS PRIME X299-A motherboard, 2 x 8GB Corsair DDR4-3200MHz memory, and Samsung 950 PRO 256GB NVMe SSD. Ubuntu 16.04 LTS was running on the box; NVIDIA tests happened with the 384.59 driver on Linux 4.13. The ROCm open compute stack on Ubuntu 16.04 LTS currently uses the 4.11.0-kfd kernel and OpenCL 2.0 AMD-APP 2450.0.

LuxMark, Darktable, SHOC, FAHBench, ViennaCL, and clpeak were among the OpenCL compute benchmarks ran for this article in a fully-automated and reproducible manner via the open-source Phoronix Test Suite benchmarking software.