Merge pull request #52 from oracle-quickstart/rccl-test-update

Update BM.GPU.MI300X.8.yaml

Author: arnaudfroidmont

manifests/rccl-tests/BM.GPU.MI300X.8.yaml

@@ -35,7 +35,7 @@ spec:
             --mca btl ^openib \
             -x NCCL_DEBUG=VERSION \
             -x NCCL_IB_HCA==mlx5_0,mlx5_2,mlx5_3,mlx5_4,mlx5_5,mlx5_7,mlx5_8,mlx5_9 \
-            -x NCCL_SOCKET_IFNAME=eth0 \
+            -x UCX_NET_DEVICES=eth0 \
             -x NCCL_IB_TC=41 \
             -x NCCL_IB_SL=0 \
             -x NCCL_IB_GID_INDEX=3 \
@@ -47,14 +47,14 @@ spec:
             /workspace/rccl-tests/build/all_reduce_perf -b 1G -e 16G -f 2 -g 1
           ports:
           - { name: mpijob-port, containerPort: 2222, protocol: TCP }
-          image: iad.ocir.io/hpc_limited_availability/oke/rccl-tests:rocm-6.2.1-ofed-5.9-0.5.6.0.127
+          image: iad.ocir.io/hpc_limited_availability/oke/rccl-tests:rocm-6.3.2-OFED-24.10-1.1.4.0
           imagePullPolicy: Always
           name: mpimaster
           resources:
             limits:
               ephemeral-storage: 32Gi
             requests:
-              cpu: 8
+              cpu: 2
               ephemeral-storage: 32Gi
               memory: 2Gi
           securityContext:
@@ -82,7 +82,7 @@ spec:
           - sysctl --system; mkdir -p /var/run/sshd; /usr/sbin/sshd -D -p 2222
           ports:
           - { name: mpijob-port, containerPort: 2222, protocol: TCP }
-          image: iad.ocir.io/hpc_limited_availability/oke/rccl-tests:rocm-6.2.1-ofed-5.9-0.5.6.0.127
+          image: iad.ocir.io/hpc_limited_availability/oke/rccl-tests:rocm-6.3.2-OFED-24.10-1.1.4.0
           imagePullPolicy: Always
           name: mpiworker
           resources:
@@ -108,4 +108,4 @@ spec:
         tolerations:
         - { key: amd.com/gpu, operator: Exists }
         volumes:
-        - { name: shm, emptyDir: { medium: Memory, sizeLimit: 128Gi }}
+        - { name: shm, emptyDir: { medium: Memory, sizeLimit: 128Gi }}