CUDA 3.2RC & 3.0

Dear CUDA fellows,

I’m having troubles with one of my kernels and I change to 3.2RC, now I realize that I don’t have emu staff and I can’t do a proper nsight debug (with another target pc) so the only thing I can do is go back to 3.0 but I don’t want. Do you know if it is possible to have both versions working? and how? (another OS installed? user accounts? anything…?)

I have two computers with win7 and some available space for install an alternative O.S, right now both machines are running CUDA 3.2RC with nsight I want to keep it that way. I’m looking to start debugging with Nsight 1.5 but in this moment emulation can help me more.

Now I passed this faulty kernel (or I will the faulty guy) and I’m fine with 3.2RC, I just want to fix the problem sometime somehow in the meantime.

Any comments? thanks in advance!

Daniel.

Can I for example create another user and install 3.0 there so user account switch would mean cuda version switch?

Can I for example create another user and install 3.0 there so user account switch would mean cuda version switch?

Can I for example create another user and install 3.0 there so user account switch would mean cuda version switch?

Can’t u switch between different CUDA versions just by changing the “CUDA_LIB_PATH” and “CUDA_INC_PATH” and other related envmt variables (like PATH)? This assumes that you need a common driver that will support both CUDA versions and hopefully “toolkit” is your concern and not the driver.

In Linux, probably u can make /usr/local/cuda as a symbolic link to either /usr/local/cuda3.0 or /usr/local/cuda3.2rc (like a crossbar switch) - you can use the toolkit that you need.

Can’t u switch between different CUDA versions just by changing the “CUDA_LIB_PATH” and “CUDA_INC_PATH” and other related envmt variables (like PATH)? This assumes that you need a common driver that will support both CUDA versions and hopefully “toolkit” is your concern and not the driver.

In Linux, probably u can make /usr/local/cuda as a symbolic link to either /usr/local/cuda3.0 or /usr/local/cuda3.2rc (like a crossbar switch) - you can use the toolkit that you need.

Can’t u switch between different CUDA versions just by changing the “CUDA_LIB_PATH” and “CUDA_INC_PATH” and other related envmt variables (like PATH)? This assumes that you need a common driver that will support both CUDA versions and hopefully “toolkit” is your concern and not the driver.

In Linux, probably u can make /usr/local/cuda as a symbolic link to either /usr/local/cuda3.0 or /usr/local/cuda3.2rc (like a crossbar switch) - you can use the toolkit that you need.