fix(ncp-local): size fake GPUs for one-GPU workers

Author: mikeyrcamp

tools/ncp-local-cluster/apps/fake-gpu-operator/static/fake-gpu-operator.yaml

@@ -63,7 +63,7 @@ data:
     nodePoolLabelKey: run.ai/simulated-gpu-node-pool
     nodePools:
       default:
-        gpuCount: 8
+        gpuCount: 2
         gpuMemory: 81559
         gpuProduct: NVIDIA-H100-SXM4-80GB
 kind: ConfigMap
@@ -418,7 +418,7 @@ spec:
     metadata:
 
       annotations:
-        checksum/topology: 4e5af8859971722ac4a3aa64e8f38f255630bad818277aec4825b6c1d1f79cb1
+        checksum/topology: d4293b3607dec297d688bfb0dfe5ffa78eb892015c926bf1d5b61500788468c4
       labels:
         app: device-plugin
         component: device-plugin
@@ -496,7 +496,7 @@ spec:
   template:
     metadata:
       annotations:
-        checksum/topology: 4e5af8859971722ac4a3aa64e8f38f255630bad818277aec4825b6c1d1f79cb1
+        checksum/topology: d4293b3607dec297d688bfb0dfe5ffa78eb892015c926bf1d5b61500788468c4
       labels:
         app: mig-faker
         component: mig-faker
@@ -652,7 +652,7 @@ kind: Deployment
 metadata:
   name: kwok-gpu-device-plugin
   annotations:
-    checksum/topology: 4e5af8859971722ac4a3aa64e8f38f255630bad818277aec4825b6c1d1f79cb1
+    checksum/topology: d4293b3607dec297d688bfb0dfe5ffa78eb892015c926bf1d5b61500788468c4
   labels:
     app: kwok-gpu-device-plugin
 spec:
@@ -664,7 +664,7 @@ spec:
   template:
     metadata:
       annotations:
-        checksum/topology: 4e5af8859971722ac4a3aa64e8f38f255630bad818277aec4825b6c1d1f79cb1
+        checksum/topology: d4293b3607dec297d688bfb0dfe5ffa78eb892015c926bf1d5b61500788468c4
       labels:
         app: kwok-gpu-device-plugin
         component: kwok-gpu-device-plugin
@@ -770,7 +770,7 @@ kind: Deployment
 metadata:
   name: status-updater
   annotations:
-    checksum/topology: 4e5af8859971722ac4a3aa64e8f38f255630bad818277aec4825b6c1d1f79cb1
+    checksum/topology: d4293b3607dec297d688bfb0dfe5ffa78eb892015c926bf1d5b61500788468c4
   labels:
     app: status-updater
 spec:
@@ -782,7 +782,7 @@ spec:
   template:
     metadata:
       annotations:
-        checksum/topology: 4e5af8859971722ac4a3aa64e8f38f255630bad818277aec4825b6c1d1f79cb1
+        checksum/topology: d4293b3607dec297d688bfb0dfe5ffa78eb892015c926bf1d5b61500788468c4
       labels:
         app: status-updater
         component: status-updater

tools/ncp-local-cluster/apps/fake-gpu-operator/values.yaml

@@ -6,9 +6,9 @@ topology:
     default:
       gpuProduct: NVIDIA-H100-SXM4-80GB
       # Keep a one-GPU local instance schedulable after NVCA infrastructure
-      # overhead. A 128-way slice on k3d nodes is too small and suppresses
-      # the required *_1x registration type.
-      gpuCount: 8
+      # overhead. Higher-density slices on k3d nodes suppress the required
+      # *_1x registration type.
+      gpuCount: 2
       gpuMemory: 81559
   nodePoolLabelKey: run.ai/simulated-gpu-node-pool
   migStrategy: mixed

tools/ncp-local-cluster/tests/test-multicluster-make.sh

@@ -83,7 +83,7 @@ assert_missing_docker_config_blocks_target start
 assert_missing_docker_config_blocks_target build-and-deploy-cluster
 
 fake_gpu_count="$(yq -r '.topology.nodePools.default.gpuCount' "$ROOT_DIR/apps/fake-gpu-operator/values.yaml")"
-assert_eq "8" "$fake_gpu_count" "fake GPU count"
+assert_eq "2" "$fake_gpu_count" "fake GPU count"
 
 print_directory_clusters="$(MAKEFLAGS=--print-directory; export MAKEFLAGS; run_make print-compute-clusters)"
 assert_eq "ncp-local-compute-1" "$print_directory_clusters" "print-directory compute cluster output"