Since GeForce 9300 has a DirectX 10 engine and according to NVIDIA it supports CUDA (programs compiled with CUDA will run on the graphics engine rather than on the CPU, for improved performance) we decided to test if the graphics engine from GeForce 9300 could really run CUDA-compiled programs.

To test that, we installed the Folding at Home GPU client, which uses the graphics engine to make protein folding calculations.

We had only two problems, none to do with GeForce 9300. The driver ECS shipped with the motherboard didn’t support CUDA, so we had to download the latest driver from NVIDIA’s website to make the Folding@Home client recognize GeForce 9300. We had an issue with Folding@Home installing its .dll files on the wrong directory, but this is a problem with the installer.

After setting everything up Folding@Home started using the graphics engine to do math calculations. It downloaded project 5014, which gives 480 points after a work unit is processed. GeForce 9300 was taking 7 minutes and 46 seconds to process 1% of the work, so it would take 46,600 seconds to process one work unit, or 13 hours. So the maximum theoretical performance you should expect for GeForce 9300 processing Folding@Home calculations is two work units per day (maximum of 960 points/day). Even though this is a low number compared to high-end video cards (which can easily process more than five work units per day), this is two and a half times faster than a Core 2 Duo E6600 running the standard CPU client (118.100 seconds/WU). Note that this comparison is flawed as the kind of processing a work unit targeted to video cards does is different from the kind of processing a work unit targeted to CPUs does. Anyway, you will get more points running your Folding@Home client on GeForce 9300 than running it on a Core 2 Duo CPU (using the standard client).