feat(nvidia): support RTX PRO 6000 Blackwell dynamic MIG

charts/hami/templates/scheduler/device-configmap.yaml

@@ -230,6 +230,23 @@ data:
               core: 100
               memory: 184320
               count: 1
+      - models: [ "RTX PRO 6000 Blackwell Server Edition" ]
+        allowedGeometries:
+          -
+            - name: 1g.24gb
+              core: 25
+              memory: 24576
+              count: 4
+          -
+            - name: 2g.48gb
+              core: 50
+              memory: 49152
+              count: 2
+          -
+            - name: 4g.96gb
+              core: 100
+              memory: 98304
+              count: 1
     cambricon:
       resourceCountName: {{ .Values.mluResourceName }}
       resourceMemoryName: {{ .Values.mluResourceMem }}

docker/Dockerfile

@@ -8,7 +8,7 @@ ADD . /k8s-vgpu
 #RUN --mount=type=cache,target=/go/pkg/mod \
 #    cd /k8s-vgpu && make all
 RUN cd /k8s-vgpu && make all VERSION=$VERSION
-RUN go install github.com/NVIDIA/mig-parted/cmd/nvidia-mig-parted@v0.12.2
+RUN go install github.com/NVIDIA/mig-parted/cmd/nvidia-mig-parted@v0.12.3
 
 FROM $NVIDIA_IMAGE AS nvbuild
 RUN dnf install -y cmake git && dnf clean all

docker/Dockerfile.hamimaster

@@ -8,7 +8,7 @@ ADD . /k8s-vgpu
 ARG VERSION
 RUN go env -w GO111MODULE=on
 RUN cd /k8s-vgpu && make all VERSION=$VERSION
-RUN go install github.com/NVIDIA/mig-parted/cmd/nvidia-mig-parted@v0.12.2
+RUN go install github.com/NVIDIA/mig-parted/cmd/nvidia-mig-parted@v0.12.3
 
 FROM nvidia/cuda:13.3.0-base-ubuntu22.04
 RUN apt-get update && \

docker/Dockerfile.withlib

@@ -8,7 +8,7 @@ ARG GOPROXY=https://goproxy.cn,direct
 ARG VERSION
 RUN go env -w GO111MODULE=on
 RUN cd /k8s-vgpu && make all VERSION=$VERSION
-RUN go install github.com/NVIDIA/mig-parted/cmd/nvidia-mig-parted@v0.12.2
+RUN go install github.com/NVIDIA/mig-parted/cmd/nvidia-mig-parted@v0.12.3
 
 FROM nvidia/cuda:13.3.0-base-ubuntu22.04
 RUN apt-get update && \

docs/develop/dynamic-mig.md

@@ -87,6 +87,20 @@ data:
             - name: 7g.79gb
               memory: 80896
               count: 1
+      - models: [ "RTX PRO 6000 Blackwell Server Edition" ]
+        allowedGeometries:
+          -
+            - name: 1g.24gb
+              memory: 24576
+              count: 4
+          -
+            - name: 2g.48gb
+              memory: 49152
+              count: 2
+          -
+            - name: 4g.96gb
+              memory: 98304
+              count: 1
       nodeconfig: 
           - name: nodeA
             operatingmode: hami-core

pkg/device-plugin/nvidiadevice/nvinternal/plugin/util_test.go

@@ -60,6 +60,14 @@ func TestGenerateMigTemplate(t *testing.T) {
 					{device.MigTemplate{Name: "7g.80gb", Core: 100, Memory: 81920, Count: 1}},
 				},
 			},
+			{
+				Models: []string{"RTX PRO 6000 Blackwell Server Edition"},
+				Geometries: []device.Geometry{
+					{device.MigTemplate{Name: "1g.24gb", Core: 25, Memory: 24576, Count: 4}},
+					{device.MigTemplate{Name: "2g.48gb", Core: 50, Memory: 49152, Count: 2}},
+					{device.MigTemplate{Name: "4g.96gb", Core: 100, Memory: 98304, Count: 1}},
+				},
+			},
 		},
 	}
 
@@ -134,6 +142,23 @@ func TestGenerateMigTemplate(t *testing.T) {
 				"1g.5gb": 7,
 			},
 		},
+		{
+			// The full NVML model string must match the shorter configured
+			// model via substring matching (RTX PRO 6000 Blackwell Server Edition).
+			name:      "rtx pro 6000 blackwell 1g.24gb template",
+			model:     "NVIDIA RTX PRO 6000 Blackwell Server Edition",
+			deviceIdx: 0,
+			containerDev: device.ContainerDevice{
+				Idx:     0,
+				UUID:    "ccccdddd[0-3]",
+				Usedmem: 20000,
+			},
+			expectedPos:   3,
+			expectedReset: true,
+			expectedMig: map[string]int32{
+				"1g.24gb": 4,
+			},
+		},
 	}
 
 	for _, tc := range testCases {