feat(gpu): recognize aks-gpu-cuda in LoadConfig alongside aks-gpu-cuda-lts

.github/renovate.json

@@ -580,6 +580,17 @@
       "enabled": true,
       "ignoreUnstable": false
     },
+    {
+      "matchPackageNames": [
+        "aks/aks-gpu-cuda"
+      ],
+      "groupName": "nvidia-gpu-cuda-legacy",
+      "versioning": "regex:^(?<major>\\d+)\\.(?<minor>\\d+)\\.(?<patch>\\d+)-(?<prerelease>\\d{14})$",
+      "allowedVersions": "/^580\\./",
+      "automerge": false,
+      "enabled": true,
+      "ignoreUnstable": false
+    },
     {
       "matchPackageNames": [
         "aks/aks-gpu-grid"

parts/common/components.json

@@ -742,6 +742,13 @@
         "latestVersion": "580.159.04-20260629214430"
       }
     },
+    {
+      "downloadURL": "mcr.microsoft.com/aks/aks-gpu-cuda:*",
+      "gpuVersion": {
+        "renovateTag": "registry=https://mcr.microsoft.com, name=aks/aks-gpu-cuda",
+        "latestVersion": "580.126.09-20260430040408"
+      }
+    },
     {
       "downloadURL": "mcr.microsoft.com/aks/aks-gpu-grid:*",
       "gpuVersion": {

pkg/agent/baker.go

@@ -1497,7 +1497,7 @@ func GetGPUDriverVersion(size string) string {
 	if isStandardNCv1(size) {
 		return datamodel.Nvidia470CudaDriverVersion
 	}
-	return datamodel.NvidiaCudaDriverVersion
+	return datamodel.NvidiaCudaLTSDriverVersion
 }
 
 func isStandardNCv1(size string) bool {
@@ -1522,7 +1522,7 @@ func GetAKSGPUImageSHA(size string) string {
 	if useGridDrivers(size) {
 		return datamodel.AKSGPUGridVersionSuffix
 	}
-	return datamodel.AKSGPUCudaVersionSuffix
+	return datamodel.AKSGPUCudaLTSVersionSuffix
 }
 
 // GetGPUDriverType maps a GPU VM size to the aks-gpu image variant used to install its driver.

pkg/agent/baker_test.go

@@ -943,7 +943,7 @@ var _ = Describe("GetGPUDriverVersion", func() {
 		Expect(GetGPUDriverVersion("standard_nc6")).To(Equal(datamodel.Nvidia470CudaDriverVersion))
 	})
 	It("should use cuda with nc v3", func() {
-		Expect(GetGPUDriverVersion("standard_nc6_v3")).To(Equal(datamodel.NvidiaCudaDriverVersion))
+		Expect(GetGPUDriverVersion("standard_nc6_v3")).To(Equal(datamodel.NvidiaCudaLTSDriverVersion))
 	})
 	It("should use grid with nv v5", func() {
 		Expect(GetGPUDriverVersion("standard_nv6ads_a10_v5")).To(Equal(datamodel.NvidiaGridDriverVersion))
@@ -958,7 +958,7 @@ var _ = Describe("GetGPUDriverVersion", func() {
 	})
 	// NV V1 SKUs were retired in September 2023, leaving this test just for safety
 	It("should use cuda with nv v1", func() {
-		Expect(GetGPUDriverVersion("standard_nv6")).To(Equal(datamodel.NvidiaCudaDriverVersion))
+		Expect(GetGPUDriverVersion("standard_nv6")).To(Equal(datamodel.NvidiaCudaLTSDriverVersion))
 	})
 })
 
@@ -995,8 +995,8 @@ var _ = Describe("GetAKSGPUImageSHA", func() {
 		Expect(GetAKSGPUImageSHA("standard_nc128ds_xl_rtxpro6000bse_v6")).To(Equal(datamodel.AKSGPUGridV20VersionSuffix))
 		Expect(GetAKSGPUImageSHA("standard_nc128lds_xl_rtxpro6000bse_v6")).To(Equal(datamodel.AKSGPUGridV20VersionSuffix))
 	})
-	It("should use newest AKSGPUCudaVersionSuffix with non grid SKU", func() {
-		Expect(GetAKSGPUImageSHA("standard_nc6_v3")).To(Equal(datamodel.AKSGPUCudaVersionSuffix))
+	It("should use newest AKSGPUCudaLTSVersionSuffix with non grid SKU", func() {
+		Expect(GetAKSGPUImageSHA("standard_nc6_v3")).To(Equal(datamodel.AKSGPUCudaLTSVersionSuffix))
 	})
 })
 

pkg/agent/datamodel/gpu_components.go

@@ -13,9 +13,11 @@ const Nvidia470CudaDriverVersion = "cuda-470.82.01"
 //nolint:gochecknoglobals
 var (
 	NvidiaCudaDriverVersion    string
+	NvidiaCudaLTSDriverVersion string
 	NvidiaGridDriverVersion    string
 	NvidiaGridV20DriverVersion string
 	AKSGPUCudaVersionSuffix    string
+	AKSGPUCudaLTSVersionSuffix string
 	AKSGPUGridVersionSuffix    string
 	AKSGPUGridV20VersionSuffix string
 )
@@ -63,6 +65,12 @@ func LoadConfig() error {
 		// confused by substring matching.
 		switch gpuImageRepo(image.DownloadURL) {
 		case "aks-gpu-cuda-lts":
+			NvidiaCudaLTSDriverVersion = version
+			AKSGPUCudaLTSVersionSuffix = suffix
+		case "aks-gpu-cuda":
+			// Pre-LTS CUDA image, pinned to the R580 line (V100-capable). Loaded so its version is
+			// available if a SKU is ever routed to the "cuda" image in CSE; the default managed CUDA
+			// image is aks-gpu-cuda-lts (see GetGPUDriverType / GetGPUDriverVersion in baker.go).
 			NvidiaCudaDriverVersion = version
 			AKSGPUCudaVersionSuffix = suffix
 		case "aks-gpu-grid":

pkg/agent/datamodel/gpu_components_test.go

@@ -63,6 +63,16 @@ func TestLoadConfig(t *testing.T) {
 	if !suffixPattern.MatchString(AKSGPUGridV20VersionSuffix) {
 		t.Errorf("AKSGPUGridV20VersionSuffix '%s' does not match expected format", AKSGPUGridV20VersionSuffix)
 	}
+
+	// aks-gpu-cuda-lts drives the render, so its version/suffix must be loaded. aks-gpu-cuda
+	// (NvidiaCudaDriverVersion / AKSGPUCudaVersionSuffix, checked above) is the recognized pre-LTS
+	// image, available if a SKU is routed to the "cuda" image in CSE later.
+	if !versionPattern.MatchString(NvidiaCudaLTSDriverVersion) {
+		t.Errorf("NvidiaCudaLTSDriverVersion '%s' does not match expected format", NvidiaCudaLTSDriverVersion)
+	}
+	if !suffixPattern.MatchString(AKSGPUCudaLTSVersionSuffix) {
+		t.Errorf("AKSGPUCudaLTSVersionSuffix '%s' does not match expected format", AKSGPUCudaLTSVersionSuffix)
+	}
 }
 
 // TestGPUImageRepo verifies that the bare repo name is extracted via exact final