we need a profiler for wsl. Especially to see why OpenACC or Unified Memory is different. I can see from the task manager they behave different.
Am I correct that your Maneged Memory code compile for Windows in the old way?!
It shouldn’t compile at all or gives me at leat some warnings.
OpenACC works great for multicore.
Also, It seems to me that WSL memory copying is slower. Is it because
the driver is somehow bridged?!
Your HPC SDK is great another world lucky Linux guys.
Now it works with me after i installed few days ago CUDA Toolkit v11.7.0
BUT I am confused because in this manual it is said
Developers who require debugging support are encouraged to find alternatives in the meanwhile.
IT WORKS FINE FOR ME
is it ok with everybody or i had someway installed something that made it works for me.
To install i have just used your download webpage instruction for linux runfile
CUDA Toolkit 11.7 Downloads | NVIDIA Developer
sudo sh cuda_11.7.0_515.43.04_linux.run
that is all and then export the new path to make it work
nothing special. it is easy and straightforward
Of course, I have installed first cuda tools 11.7 for windows first, which is newer than
microsoft drivers 11.6.
mabd@LAPTOP-T8DQ9UK0:~/py$ nsight-sys python3 simple_speed_test.py
OpenGL version: “3.1 Mesa 20.0.8”
it is interesting that now mesa driver is installed i.e opengl progs will work
that is great.
nsight-sys launchs without problems but only collects cpu no gpu. i will recheck it latter
nvvp launches but complains about cc>7.2 it has some limited functionality.
Over the time i have installed different graphics support X stuff and even
gimp gummi and may be others