Which driver / cuda? instrumented cuda driver? backward comp?

[MarkVanHeeswijk]

Hello,

i’m currently using the instrumented driver that came with NVPerfKit for my work, but for school i would want to start learning CUDA (in emulated mode) with the Matlab plugin for CUDA 1.1.

I’m using Windows XP SP2, and my video card is a Geforce Go 7600 (so only emulated mode possible, right?)

It would be thankful if someone could point me in the right direction on the following questions:

[*] Is there an instrumented driver with CUDA support?

[*] If not, is it possible to quickly switch between drivers somehow? (ugly hacks allowed ;) )

[*] Are newer drivers like the one that is listed with CUDA 2.0b2 backwards compatible with older CUDA versions, or is it a driver per version?

[*] Is CUDA 2.0 backwards compatible with plugins developed for older versions of CUDA (like the Matlab plugin)?

Thanks!

[MisterAnderson42]

Is there an instrumented driver with CUDA support?

I am not aware of any.

If not, is it possible to quickly switch between drivers somehow? (ugly hacks allowed ;) )

On Windows? I doubt it. There are ways to do it on linux that only require restarting X instead of the whole machine.

Are newer drivers like the one that is listed with CUDA 2.0b2 backwards compatible with older CUDA versions, or is it a driver per version?

The beta CUDA drivers are one version per driver. With the released CUDAs, I honestly don’t know. I never tried to run a CUDA 1.0 app on a CUDA 1.1 driver.

Is CUDA 2.0 backwards compatible with plugins developed for older versions of CUDA (like the Matlab plugin)?

I’ve never used the matlab plugin system, but my large application written for CUDA 1.1 compiled and ran on 2.0 without any modifications. CUDA 2.0 only adds new features and doesn’t change the API in any way that breaks old code.