This lesson will work best if done on a real cluster, like Alliance’s Béluga, Cedar, Graham or Narval. If you haven’t got an account on those, our virtual training cluster will also be fine although some of the profiling exercises might not work as described.
ssh userXX@pcsii.ace-net.training
To set up your environment, load the GCC Compiler Collection and CUDA with:
module purge
module load StdEnv/2020 gcc/9.3.0 cuda/11.4
You should have access to the compiler and tools now. Test this with:
which nvcc
which nvprof
On real clusters we can use
StdEnv/2023
withcuda/12.2
insteadAll of our real clusters have been upgraded to also support
StdEnv/2023
withcuda/12.2
instead of the oldercuda/11.4
that is still required on the virtual training clusters.module purge module load StdEnv/2023 cuda/12.2