Q: Graphics Acceleration is not being used?

I have some products that are written to take advantage of GPU-processing (Graphics Acceleration) if so configured.

• When I do not select Graphics Acceleration in the software, in Task Manager the process CPU load is ~2-3%.
• When I do select Graphics Acceleration in the software, the process CPU load is ~14%.

Which is very odd. In both cases GPUShark reports that no GPU processing power is used. The big mystery is why my CPU load rises significantly when I enable Graphics Acceleration in the program.

Everywhere I have looked tells me that the graphics hardware acceleration is enabled. The card is a Gainward 9500 GT (NVIDIA-based, Open GL 3.2.0, Open CL 1.0, CUDA 3.0.1). Old, I know by trouble-free, so far.

Any thoughts that will help me track down the problem? Is there any simple software which will test the GPU functionality (a load tester would not be necessary)?

First of all, try your card in a different slot along with the latest drivers for your card...

I would like to understand the problem on my system first.

Why would the CPU load rise when I select to use graphics acceleration in the software products? I have contacted the developer and it works perfectly on their systems with a three-fold reduction in CPU load.

Also, is there any idiot's guide to how Graphics Acceleration works? Something beyond "the GPU takes over some of the processing" but not as far as "add this ... to your code".

The 9500 GT only supports Cuda 1.1 are you sure your application is written for 1.1?

From what I see it is CUDA 3.0.1 (and that's the version on the nvcuda.dll driver file). Is there something-else I should check?

But that might still be too old -I'll check with the developer.

If your CPU figures are correct, I'd guess that the GPU doesn't support whatever feature is being used and it's resorting to emulation in the CPU.

Just because the CUDA dll says its 3.0.1 doesn't mean the card supports it.

For example, installing DirectX 11 card drivers won't make a DX7 card gain or support new hardware features, said features will either be unavailable or emulated in software instead.

I've looked around and it looks like the CUDA driver version specified (3.0.1) does not correspond with the "Compute capability" version (1.1). so MaddAussie was correct. My apologies for the confusion.

From GPUShark:
http://i.imgur.com/y8C8H0z.png

@ BobF - taking your statement about falling-back to emulation in the CPU, that could well explain why the CPU load goes up when I opt for "Graphics Acceleration". As the graphics processing is not heavy-duty 3D maybe I'll try to persuade the developer to support the older standard (if that is indeed possible and my PC has not been overtaken by progress ;)).