Add some more updates to the helper scripts

Author: ntselepidis

misc/nvidia_uvm/bind.sh

@@ -14,7 +14,7 @@ export CUDA_VISIBLE_DEVICES="$local_rank"
 export MPICH_OFI_NIC_POLICY=USER
 export MPICH_OFI_NIC_MAPPING="0:0; 1:1; 2:2; 3:3"
 
-# Bind to cores ( first core per socket )
+# Bind to cores ( all cores per socket )
 physcores=(0-71 72-143 144-215 216-287)
 
 #echo hostname: $(hostname), rank: $local_rank, cores: ${physcores[$local_rank]}, GPU: $CUDA_VISIBLE_DEVICES, NIC mapping: $MPICH_OFI_NIC_POLICY

misc/nvidia_uvm/nsys.sh

@@ -15,7 +15,7 @@ if [[ "$NSYS" -ne 0 && "$rank" -eq 0 ]]; then
       --cpu-socket-events=61,71,265,273 \
       --cpu-socket-metrics=103,104 \
       --event-sampling-interval=10 \
-      --trace=nvtx,openacc \
+      --trace=nvtx,cuda,openacc \
       --force-overwrite=true \
       -e NSYS_MPI_STORE_TEAMS_PER_RANK=1 \
       -o "$NSYS_FILE" "$@"

toolchain/templates/santis.mako

@@ -6,6 +6,7 @@
 #SBATCH --uenv=icon/25.2:v1@santis
 #SBATCH --nodes=${nodes}
 #SBATCH --ntasks-per-node=${tasks_per_node}
+#SBATCH --cpus-per-task=72
 #SBATCH --job-name="${name}"
 #SBATCH --output="${name}.out"
 #SBATCH --error="${name}.err"