NVIDIA GeForce OpenCL Performance Comparison

A while back I performed an OpenCL performance comparison against a range of AMD Radeon graphics cards. In this article, the table has turned as the OpenCL results on NVIDIA's GeForce graphics cards are examined.

This testing was done just like the AMD Radeon OpenCL benchmarks. In fact, it was done at the same time, but only now am I getting around to pushing it through the publishing queue. Aside from the GPU OpenCL benchmarking, also of interest is the Intel Support For OpenCL On Linux With Ivy Bridge and AMD OpenCL APP SDK Beats Intel's Own SDK On Ivy Bridge.

This GeForce OpenCL compute testing is being done using the proprietary NVIDIA Linux graphics driver since the OpenCL/Clover Gallium3D support for the open-source Nouveau driver is still being established.

Benchmarking was done from the usual Intel Core i7-3770K "Ivy Bridge" system with the ECS Z77H2-A2X motherboard. The system was running Ubuntu 12.04 LTS with the Linux 3.2 kernel. System details (and the raw results) are available in full from the OpenBenchmarking.org result page.

Graphics cards that were benchmarked include the NVIDIA GeForce 9600GSO, 9800GT, GT 220, GT 240, GTX 460, GT 520, and GTX 680 Kepler. The number of CUDA/Compute cores and other details for each graphics card is also available from the aforementioned OpenBenchmarking.org page.

Benchmarking was done via the LuxMark test profile.
There's not much to say that's not unexpected: the newer NVIDIA GPUs are much faster with their GPGPU performance. Especially with the GeForce 400/500 "Fermi" and now GeForce 600 "Kepler" graphics cards, NVIDIA's really ramped up its compute potential not only for OpenCL but CUDA too.

From OpenBenchmarking.org you can compare the results with the AMD Radeon GPUs and various other community-contributed results for LuxMark. You can also compare your own system's OpenCL performance numbers on the GPU by having the Phoronix Test Suite installed and simply running phoronix-test-suite benchmark 1206233-SU-OPENCLNVI32. It's as easy as that! (Assuming you have an OpenCL supported GPU and driver setup.)