I was quite curious about the OpenCL performance of these latest cards. AMD has really been pushing OpenCL as of late, especially in advance of the release of the Kaveri APUs. They have a pretty solid OpenCL 1.2 implementation for their GCN based chips. NVIDIA of course has been blowing the GPGPU horn as loudly as possible over the years and focused on CUDA. Oddly enough, Kepler made compute performance take a back seat to enable better performance and efficiency in graphics applications. I used the 64 bit version of Luxmark 2.0 with the Sala test.
Yeah, that ain’t pretty. The GTX 780 is slower than the 270X when it comes to this particular OpenCL application. It looks as though the upcoming Maxwell architecture improves GPGPU/OpenCL type performance to a pretty large degree. Unfortunately for the GTX 780, it is not based on the Maxwell architecture. The Kepler architecture does some things very well, but this type of computing is not one of them.