Am I right in assuming that an overclocked CUDA GPU will perform BOINC projects better than CUDA standard GPUs that are not overclocked, as a general rule?

I keep an old computer always on and running climateprediction on a GPU I bought for it (nVidia Geforce 8400GS). Question is whether I should overclock the GPU to do project tasks faster. It might not make much difference for this GPU, but I'm wondering about overclocking in general for distributed computing projects.

...running climateprediction on a GPU.

I doubt that very much. Climate Prediction doesn't do any work yet on GPUs, as at this time it's still too costly and difficult to port the Fortran base-code to CUDA code. Nothing said about the code used being proprietary to the Met Office.

So... whatever you're using the GPU for, outside of CPDN, will an overclock speed it up? Think about why tasks are being crunched faster on a GPU, when compared to on a CPU. A CPU has one core, a GPU's core consists of many stream processors. All these processors do part of the task in parallel and that's what speeds up the calculations.

Does it then matter how fast these streaming processors are? Probably, at 1-10%, but not much (in my opinion). Not when you think that a GPU with 256 stream processors will hack tasks faster than a GPU with 80 stream processors.

---

Thanks.

BTW -- Message tab in BOIC shows Climateprediction project is requesting work for the GPU, and also show that it tests the CPU occasionally, so something is working on the GPU, if not Climateprediction (the only project I am running).

Would BOINC request work from a project for a GPU even if the project was not designed to take advantage of GPUs?

Would BOINC request work from a project for a GPU even if the project was not designed to take advantage of GPUs?

Yes, it would request work - at random times, but roughly once per day. It won't ever get any, though.