New to CUDA under Windows, I soon came across the run-time limits on kernels, & the reg edit workarounds to increase the timeout there.
Now new to CUDA under Linux, when porting that code to L4T on the nano, I quickly realised I might be hitting a similar restriction. I’m guessing it’s this, as reducing the kernel workload / iterations per kernel makes things run as expected. But rather than exit with an error as it does under windows, my code seems to hang at the point where the kernel workload has exceeded the imagined limit.
- Is the hung code (L4T) v early exit (windows) the correct symptom re too long a kernel run time?
- Googling seems to indicate no way under linux/L4T to increase the timeout - correct?