With having out most of my NVIDIA graphics cards earlier this week due to running the 27-way OpenGL and performance-per-Watt comparison on NVIDIA graphics cards going back a decade, I took the opportunity to also run a smaller, fresh OpenCL/CUDA GPU compute comparison on various recent NVIDIA GPUs.
If GPU compute is more along your lines rather than OpenGL graphics, this is the article for you. Due to time constraints and the rapid progression of GPGPU performance as of late, for this comparison I used 13 recent NVIDIA GeForce graphics cards from Kepler and Maxwell generations. Unfortunately they were all GeForce graphics cards due to not having access to any current Quadro GPUs.
The graphics cards used for this comparison were the GeForce GTX 650, GTX 680, GTX 750, GTX 750 Ti, GTX 760, GTX 770, GTX 780 Ti, GTX 950, GTX 960, GTX 970, GTX 980, GTX 980 Ti, and GTX TITAN X. All of the tests were done with the NVIDIA 352.39 driver as the official Linux driver release paired with the CUDA 7.5 environment. Ubuntu 15.10 x86_64 was running on the host. My only outstanding gripes with the NVIDIA compute stack at the moment is that CUDA doesn't yet play nicely with GCC 5 (thus having to switch back to GCC 4.9.3) and still the proprietary driver stack lacks OpenCL 2.0+ support.
Besides looking at the raw OpenCL and CUDA compute performance for these 13 graphics cards, I also did some performance-per-Watt benchmarks (the AC system power consumption was measured using the WattsUp Pro power meter) and then for the modern Maxwell graphics cards also listed their current performance-per-dollar. The pricing was based upon the current NVIDIA GeForce GTX 900 series pricing listed at Amazon.com.
Before jumping to all of the numbers, just a quick reminder that if you like all of the Linux hardware tests carried out exclusively at Phoronix.com, consider taking advantage of this week's premium sale for supporting our site and allowing for many more interesting and extensive Linux hardware comparisons in the future. Thanks!