NVIDIA CUDA 8.0 Benchmarks For Ending 2016 + cuDNN Caffe AlexNet/Googlenet

With running a ton of end of year benchmarks for showing the latest Linux graphics driver performance at the end of 2016, it's mostly focused upon OpenGL and Vulkan, but for those desiring some fresh NVIDIA CUDA numbers, here they are for your viewing pleasure.

This article provides some fresh reference benchmarks of various GeForce Kepler, Maxwell, and Pascal graphics cards when using the latest public Linux driver (NVIDIA 375.26) as well as the latest CUDA (v8.0) and cuDNN (in the case of the Caffe benchmarks). Unfortunately no Quadro/Tesla accelerator benchmarks due to NVIDIA only providing GeForce hardware for testing at Phoronix.

Tested hardware included the GeForce GTX 750, GTX 760, GTX 780 Ti, GTX 950, GTX 960, GTX 970, GTX 980, GTX 980 Ti, GTX 1050, GTX 1050 Ti, GTX 1060, GTX 1070, and GTX 1080. All tests were carried out on Ubuntu 16.04 LTS. There are no big surprises from the driver side but these numbers offer a fresh glimpse at how these three generations of NVIDIA hardware are performing as we end out 2016. Plus thanks to all of the benchmarks being done in an automated and reproducible manner using the Phoronix Test Suite, you can simply install it and then run phoronix-test-suite benchmark 1612261-TA-CUDA2016N88 on any CUDA Linux box to compare that system's performance against these numbers in a fully-automated, side-by-side performance comparison with our open-source benchmarking framework.