I’m going to start on a new CUDA development project. The target device will be some mobile Kepler chip, but we won’t have access to that platform during the early stages of development. So my plan is to install some other GPU card on my desktop PC for initial testing/debugging.
Preferably, this card should feature a chip from the Kepler generation, in case I need to test/debug code using compute capability 3.x features.
Problem is though, that my PC is rather old. It doesn’t have a very strong power supply, and consequently no dedicated power plug for a graphics card.
That list is probably outdated. I have a GT 640 and it is supported:
Because I have 2 GPUs, (GT 640 connected to display and GTX Titan) Nsight would always try to attach to the GTX Titan by default if launched from the Nsight → CUDA Debugging menu (maybe because Nsight picks the ‘fastest’ CUDA device by default? not sure). To get it to attach to the process without having to connect my monitor to the GTX Titan and rebooting, I used the MSVC Debug → Attach to process with NSight GPU debugger. I modified the simplePrintf example to give me a pause so I could attach to the process before continuing.
You probably won’t have to bother with that nonsense that I had to do to prove it works as long as you have only the 1 GPU on your machine.
Thanks, I was hoping to get some answer like that ;-)
Unfortunately, I came across another problem. Win 7 is listed as a minium, while my PC is still running on XP. Can anyone tell if this would be working, too?
Presumably not, chances are the implementation of Nsight depends on WDDM, which postdates the XP driver-model. Chances are that your old computer will run fine on Windows 7.
