16-Way OpenCL Compute Comparison Of The Latest Polaris & Pascal GPUs

With FinanceBench, only a few Radeon GPUs had successfully run the test while for the other AMD GPUs the system would freeze. It's also interesting from these results that the RX 480 performance is reportedly between the GTX 1070 and GTX 1080, while also being faster than the R9 Fury, but keep in mind it only comes down to a few miliseconds difference between the GTX 1080 and the GTX 980.

Next up is LuxMark, a test that traditionally favors AMD over NVIDIA in performance. Here was also a strange situation with the GTX 1080 performing abnormally poor compared to the GTX 1070/1060 an the GTX 900 Maxwell GPUs. The GTX 1080 was running into some issue and keep in mind that the Phoronix Test Suite runs each benchmark a minimum of three times for statistical accuracy. After all of this OpenCL benchmarking was complete, I was puzzled about the really strange GTX 1080 LuxMark test results so I ended up running all of the tests again on the GTX 1080 and it turned out the same... It's something really strange happening with the GTX 1080. In this particular test, also strange about the R9 290 performance relative to the RX 480.

With the Luxball scene in LuxMark there continued to be the odd GTX 1080 performance issue (again, replicated multiple times) while AMDGPU-PRO continued overall with its strong showing for LuxMark while now at least the RX 480 was outperforming the R9 290.

Yet another scene with LuxMark.

With LuxMark's Hotel scene, the AMDGPU-PRO driver OpenCL stack couldn't run with any of the cards tested. The NVIDIA cards continued to run fine sans the GTX 1080 still coming up short of the GTX 1070.