ViennaCLBench
You might not have heard of ViennaCLBench, but as the name implies this program uses OpenCL to test common linear algebra operations. If you are thinking what purpose does this serve, think along the lines of Python or MATLAB applications. We are going to show all the benchmark results: Blas3 is a dense matrix-matrix multiplication test. Copy is a host-device bandwidth benchmark. Sparse is a sparse matrix-vector performance. Vector tests vector operations and memory bandwidth. We are also testing using Double Precision, the default selection.
There is not a difference using the latest release Adrenalin driver versus the press driver. However, there is a distinct difference and improvement using the Radeon Pro driver. The result is in GFLOPs and higher is better. With the Radeon Pro driver it is pulling higher GFLOPs.
There is also a big difference in the Copy test which tests host-device bandwidth. It actually improves from 7GB/sec to 8GB/sec using the Radeon Pro drivers.
We also see another large performance improvement using the Radeon Pro drivers in the Sparse test. The improvement over Adrenalin drivers is 18% which is very significant.
Finally, in the Vector test Radeon Pro is higher than Adrenalin, but not as big as it was in the previous tests.
Overall we see big gains using the Radeon Pro drivers in the tests provided by ViennaCLBench. It seems those functions and or bandwidth is noticeably better with Radeon Pro drivers versus Adrenalin.