I installed the latest driver for Nsight 5.5 in order to use the next-gen debugger, but unfortunately it’s not working. The legacy debugger is working fine.
I get the following message:
“Could not initialize driver for debugging.
Debugging has been automatically stopped.”
I can only say that I’m never using the Quadro when selecting the device. Moreover, when running my application with the profiler, only the Tesla shows usage. Quadro context is never mentioned in any call.
Next-gen debugger only work with pascal above card, so you cannot debug it on k20c, profiler should work on any device, could you please check if your app only runs on one GPU? I guess your app only use one card, you can set CUDA_VISIBLE_DEVICES to let it run on another card.
I’m facing the same issue here.
I want to debug a Windows CLI project. From [url]NVIDIA Nsight VSE Documentation I learned that that’s only possible with Nsight 5.5. However, I “only” have one GPU in my machine a no option to upgrade.
Thanks a lot for the answer. If I use the legacy debugger it doesn’t hit any break point.
I have three layers in my project:
.Net (CS) for testing
.cu for actual functions
and a C++/CLI layer to connect the two
Quaro K2100M is a kepler card actually GK106, I’m not sure if the latest nsight can work with it as we are only testing gk104 and gk107 now.
AFAIK, the legacy debugger can debug the .net exe + cuda dll, you must make sure that your cuda codes are contained in a local c++ not cli dll file, then use your .net exe to load it via DllImport.
You can set the bp in cuda dll then use legacy debugger to debug the .net exe, the bp should be hit.
would you generally say, that’s best practice for .Net and Cuda? I’m starting a new project so wouldn’t mind exploring a totally different work flow. I was looking into things like Alea GPU, managedCuda or altimesh, but can’t quite tell, which is the best bet in terms of performance and ease of development / debugging. I want to write something like a physics engine for .Net with minimal data transfer between host and device, so would like to have access to device memory to manage these things myself…
Sorry, my mistake, I just saw you only have on GPU, nsight cannot debug the CLR project, I’m afraid you cannot debug the code like managedCuda, nsight can only debug the code in local c++ project, the .net exe + cuda dll I told you before is actually the same thing, cu codes are contained in a local dll and the .net exe project is only a wrapper.
Could you use “c:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi.exe” to check if your TitanV is under TCC mode? I remember only quadro and tesla can work under tcc mode.