Merge pull request #27 from OguzPastirmaci/main

Update BM.GPU.H100.8-nccl-test.yaml

Author: arnaudfroidmont

manifests/BM.GPU.H100.8-nccl-test.yaml

@@ -51,19 +51,19 @@ spec:
               -x IB_RX_QUEUE_LEN=8192 \
               -x NCCL_SOCKET_IFNAME=eth0 \
               -x NCCL_IGNORE_CPU_AFFINITY=1 \
-              /workspace/nccl-tests/build/alltoall_perf -b 8 -f 2 -g 1 -e 4G -c 1
+              /workspace/nccl-tests/build/all_reduce_perf -b 8 -f 2 -g 1 -e 4G -c 1
             while :; do { [[ $exit ]] && break; }; sleep 1; done
           ports:
           - { name: mpijob-port, containerPort: 2222, protocol: TCP }
           image: ord.ocir.io/hpc_limited_availability/nccl-tests:pytorch-24.02-nccl-2.20.5-1
           name: mpimaster
           resources:
             limits:
-              ephemeral-storage: 32Gi
+              ephemeral-storage: 16Gi
             requests:
-              cpu: 128
-              ephemeral-storage: 32Gi
-              memory: 512Gi
+              cpu: 4
+              ephemeral-storage: 16Gi
+              memory: 1Gi
           securityContext:
             privileged: true
             capabilities:
@@ -121,4 +121,4 @@ spec:
         - { key: nvidia.com/gpu, operator: Exists }
         volumes:
         - { name: devinf, hostPath: { path: /dev/infiniband }}
-        - { name: shm, emptyDir: { medium: Memory, sizeLimit: 32Gi }}
+        - { name: shm, emptyDir: { medium: Memory, sizeLimit: 32Gi }}