feat(recipe): register h200 as first-class accelerator type

.claude/skills/analyzing-snapshots/SKILL.md

@@ -96,6 +96,8 @@ Key fields from `GPU.smi`:
 | `gb300` | gb200 class (Blackwell NVL family) |
 | `b200` | b200 |
 | `h100` | h100 |
+| `gh200` | unresolved — Grace Hopper Superchip, not the discrete H200 GPU (check before h200) |
+| `h200` | h200 (discrete H200 GPU) |
 | `a100` | a100 |
 | `l40` | l40 |
 | `rtx pro 6000` | rtx-pro-6000 |
@@ -272,7 +274,7 @@ aicr recipe \
 | Criteria | Extracted From | Valid Values |
 |----------|---------------|--------------|
 | service | K8s.node.provider / K8s.server.version | eks, gke, aks, oke, kind, lke |
-| accelerator | GPU.smi.gpu.model | h100, gb200, b200, a100, l40, rtx-pro-6000 |
+| accelerator | GPU.smi.gpu.model | h100, h200, gb200, b200, a100, l40, rtx-pro-6000 |
 | os | OS.release.ID | ubuntu, rhel, cos, amazonlinux |
 | intent | User-specified | training, inference |
 | platform | User-specified | dynamo, kubeflow, nim, runai, slurm |

.github/ISSUE_TEMPLATE/bug_report.yml

@@ -117,7 +117,7 @@ body:
         - Kubernetes version:
         - OS (ubuntu/cos/other) + version:
         - Kernel version:
-        - GPU type (h100/gb200/b200/a100/l40/rtx-pro-6000/other):
+        - GPU type (h100/h200/gb200/b200/a100/l40/rtx-pro-6000/other):
         - Workload intent (training/inference):
       value: |
         - AICR version (CLI `aicr version`, API image tag, or commit SHA):
@@ -126,7 +126,7 @@ body:
         - Kubernetes version:
         - OS (ubuntu/cos/other) + version:
         - Kernel version:
-        - GPU type (h100/gb200/b200/a100/l40/rtx-pro-6000/other):
+        - GPU type (h100/h200/gb200/b200/a100/l40/rtx-pro-6000/other):
         - Workload intent (training/inference):
     validations:
       required: true

api/aicr/v1/server.yaml

@@ -120,15 +120,15 @@ paths:
           description: GPU/accelerator type. If omitted, treated as "any" (wildcard).
           schema:
             type: string
-            enum: [h100, gb200, b200, a100, l40, rtx-pro-6000, any]
+            enum: [h100, h200, gb200, b200, a100, l40, rtx-pro-6000, any]
             default: any
         - name: gpu
           in: query
           required: false
           description: Alias for accelerator parameter (backwards compatibility).
           schema:
             type: string
-            enum: [h100, gb200, b200, a100, l40, rtx-pro-6000, any]
+            enum: [h100, h200, gb200, b200, a100, l40, rtx-pro-6000, any]
             default: any
         - name: intent
           in: query
@@ -487,15 +487,15 @@ paths:
           description: GPU/accelerator type. If omitted, treated as "any" (wildcard).
           schema:
             type: string
-            enum: [h100, gb200, b200, a100, l40, rtx-pro-6000, any]
+            enum: [h100, h200, gb200, b200, a100, l40, rtx-pro-6000, any]
             default: any
         - name: gpu
           in: query
           required: false
           description: Alias for accelerator parameter (backwards compatibility).
           schema:
             type: string
-            enum: [h100, gb200, b200, a100, l40, rtx-pro-6000, any]
+            enum: [h100, h200, gb200, b200, a100, l40, rtx-pro-6000, any]
             default: any
         - name: intent
           in: query
@@ -1262,7 +1262,7 @@ components:
         accelerator:
           type: string
           description: GPU/accelerator type
-          enum: [h100, gb200, b200, a100, l40, rtx-pro-6000, any]
+          enum: [h100, h200, gb200, b200, a100, l40, rtx-pro-6000, any]
           example: h100
         intent:
           type: string

docs/README.md

@@ -8,7 +8,7 @@ NVIDIA AI Cluster Runtime (AICR) is a suite of tooling designed to automate the
 |------|-------------|
 | **Snapshot** | A captured state of a system including OS, kernel, Kubernetes, GPU, and SystemD configuration. Created by `aicr snapshot` or the Kubernetes agent. |
 | **Recipe** | A generated configuration recommendation containing component references, constraints, and deployment order. Created by `aicr recipe` based on criteria or snapshot analysis. |
-| **Criteria** | Query parameters that define the target environment: `service` (eks/gke/aks/oke/kind/lke/bcm), `accelerator` (h100/gb200/b200/a100/l40/rtx-pro-6000), `intent` (training/inference), `os` (ubuntu/rhel/cos/amazonlinux/talos), `platform` (dynamo, kubeflow, nim, runai, slurm), and `nodes`. |
+| **Criteria** | Query parameters that define the target environment: `service` (eks/gke/aks/oke/kind/lke/bcm), `accelerator` (h100/h200/gb200/b200/a100/l40/rtx-pro-6000), `intent` (training/inference), `os` (ubuntu/rhel/cos/amazonlinux/talos), `platform` (dynamo, kubeflow, nim, runai, slurm), and `nodes`. |
 | **Overlay** | A recipe metadata file that extends the base recipe for specific environments. Overlays are matched against criteria using asymmetric matching. |
 | **Mixin** | A composable recipe fragment (`kind: RecipeMixin`) that carries only `constraints` and `componentRefs`. Mixins live in `recipes/mixins/`, are excluded from overlay discovery, and are referenced by leaf overlays via `spec.mixins` to share orthogonal content (e.g., OS constraints, platform components) without duplication. See [ADR-005](design/005-overlay-refactoring.md). |
 | **Bundle** | Deployment artifacts generated from a recipe: Helm values files, Kubernetes manifests, installation scripts, and checksums. |

docs/contributor/api-server-extending.md

@@ -1087,7 +1087,7 @@ var validate = validator.New()
 type RecipeRequest struct {
     OS       string `validate:"required,oneof=ubuntu rhel cos"`
     OSVersion string `validate:"omitempty,semver"`
-    GPU      string `validate:"required,oneof=h100 gb200 b200 a100 l40 rtx-pro-6000"`
+    GPU      string `validate:"required,oneof=h100 h200 gb200 b200 a100 l40 rtx-pro-6000"`
     Service  string `validate:"omitempty,oneof=eks gke aks oke kind lke bcm"`
 }
 

docs/contributor/api-server.md

@@ -263,7 +263,7 @@ Supported content types:
 | Parameter | Type | Validation | Example |
 |-----------|------|------------|--------|
 | `service` | ServiceType | Enum: eks, gke, aks, oke, kind, lke, bcm, any | `service=eks` |
-| `accelerator` | AcceleratorType | Enum: h100, gb200, b200, a100, l40, rtx-pro-6000, any | `accelerator=h100` |
+| `accelerator` | AcceleratorType | Enum: h100, h200, gb200, b200, a100, l40, rtx-pro-6000, any | `accelerator=h100` |
 | `gpu` | AcceleratorType | Alias for accelerator | `gpu=h100` |
 | `intent` | IntentType | Enum: training, inference, any | `intent=training` |
 | `os` | OSType | Enum: ubuntu, rhel, cos, amazonlinux, talos, any | `os=ubuntu` |

docs/contributor/cli.md

@@ -1700,7 +1700,7 @@ OUTPUT_DIR="recipes"
 mkdir -p "$OUTPUT_DIR"
 
 # GPU types from NVIDIA product line
-GPU_TYPES=("h100" "gb200" "b200" "a100" "l40" "rtx-pro-6000")
+GPU_TYPES=("h100" "h200" "gb200" "b200" "a100" "l40" "rtx-pro-6000")
 
 # Kubernetes services
 K8S_SERVICES=("eks" "gke" "aks" "oke" "kind" "lke" "bcm")

docs/contributor/data.md

@@ -124,7 +124,7 @@ Criteria define when a recipe matches a user query:
 | Field | Type | Description | Example Values |
 |-------|------|-------------|----------------|
 | `service` | String | Kubernetes platform | `eks`, `gke`, `aks`, `oke`, `kind`, `lke`, `bcm` |
-| `accelerator` | String | GPU hardware type | `h100`, `gb200`, `b200`, `a100`, `l40`, `rtx-pro-6000` |
+| `accelerator` | String | GPU hardware type | `h100`, `h200`, `gb200`, `b200`, `a100`, `l40`, `rtx-pro-6000` |
 | `os` | String | Operating system | `ubuntu`, `rhel`, `cos`, `amazonlinux` |
 | `intent` | String | Workload purpose | `training`, `inference` |
 | `platform` | String | Platform/framework type | `dynamo`, `kubeflow`, `nim`, `runai`, `slurm` |
@@ -801,7 +801,7 @@ fingerprint:
     value: eks                       # eks | gke | aks | oke | kind | lke | bcm
     source: k8s.node.provider
   accelerator:
-    value: h100                      # h100 | gb200 | b200 | a100 | l40 | rtx-pro-6000
+    value: h100                      # h100 | h200 | gb200 | b200 | a100 | l40 | rtx-pro-6000
     source: gpu.smi.gpu.model
   os:
     value: ubuntu                    # ubuntu | rhel | cos | amazonlinux | talos

docs/contributor/validations.md

@@ -94,7 +94,7 @@ conditions:
 **Supported Condition Keys:**
 - `intent`: Workload intent (training, inference)
 - `service`: Kubernetes service (eks, gke, aks, oke, kind, lke, bcm)
-- `accelerator`: GPU type (h100, gb200, b200, a100, l40, rtx-pro-6000)
+- `accelerator`: GPU type (h100, h200, gb200, b200, a100, l40, rtx-pro-6000)
 - `os`: Operating system (ubuntu, rhel, cos, amazonlinux, talos)
 - `platform`: Platform/framework (dynamo, kubeflow, nim, runai, slurm)
 

docs/user/api-reference.md

@@ -117,7 +117,7 @@ Generate an optimized configuration recipe based on environment parameters.
 | Parameter | Type | Default | Description |
 |-----------|------|---------|-------------|
 | `service` | string | any | K8s service: `eks`, `gke`, `aks`, `oke`, `kind`, `lke`, `bcm`, `any` |
-| `accelerator` | string | any | GPU type: `h100`, `gb200`, `b200`, `a100`, `l40`, `rtx-pro-6000`, `any` |
+| `accelerator` | string | any | GPU type: `h100`, `h200`, `gb200`, `b200`, `a100`, `l40`, `rtx-pro-6000`, `any` |
 | `gpu` | string | any | Alias for `accelerator` |
 | `intent` | string | any | Workload: `training`, `inference`, `any` |
 | `os` | string | any | Node OS: `ubuntu`, `rhel`, `cos`, `amazonlinux`, `talos`, `any` |
@@ -821,7 +821,7 @@ openapi-generator-cli generate -i openapi.yaml -g typescript-fetch -o ./ts-clien
 
 **"Invalid accelerator type" error:**
 ```shell
-# Use valid values: h100, gb200, b200, a100, l40, rtx-pro-6000, any
+# Use valid values: h100, h200, gb200, b200, a100, l40, rtx-pro-6000, any
 curl "http://localhost:8080/v1/recipe?accelerator=h100"
 ```
 

docs/user/cli-reference.md

@@ -379,7 +379,7 @@ Generate recipes using direct system parameters:
 | Flag | Short | Type | Description |
 |------|-------|------|-------------|
 | `--service` | | string | K8s service: eks, gke, aks, oke, kind, lke, bcm |
-| `--accelerator` | `--gpu` | string | Accelerator/GPU type: h100, gb200, b200, a100, l40, rtx-pro-6000 |
+| `--accelerator` | `--gpu` | string | Accelerator/GPU type: h100, h200, gb200, b200, a100, l40, rtx-pro-6000 |
 | `--intent` | | string | Workload intent: training, inference |
 | `--os` | | string | OS family: ubuntu, rhel, cos, amazonlinux, talos |
 | `--platform` | | string | Platform/framework type: dynamo, kubeflow, nim, runai, slurm |

pkg/api/doc.go

@@ -65,7 +65,7 @@
 //
 // The /v1/recipe endpoint accepts these query parameters for GET requests:
 //   - service: Kubernetes service (eks, gke, aks, oke, kind, lke, bcm, any)
-//   - accelerator: GPU type (h100, gb200, b200, a100, l40, rtx-pro-6000, any)
+//   - accelerator: GPU type (h100, h200, gb200, b200, a100, l40, rtx-pro-6000, any)
 //   - gpu: Alias for accelerator (back-compat)
 //   - intent: Workload intent (training, inference, any)
 //   - os: Operating system (ubuntu, rhel, cos, amazonlinux, talos, any)

pkg/cli/recipe.go

@@ -100,7 +100,7 @@ func recipeCmd() *cli.Command {
 		Usage:    "Create optimized recipe for given intent and environment parameters.",
 		Description: `Generate configuration recipe based on specified environment parameters including:
   - Kubernetes service type (e.g. eks, gke, aks, oke, kind, lke, bcm)
-  - Accelerator type (e.g. h100, gb200, b200, a100, l40, rtx-pro-6000)
+  - Accelerator type (e.g. h100, h200, gb200, b200, a100, l40, rtx-pro-6000)
   - Workload intent (e.g. training, inference)
   - GPU node operating system (e.g. ubuntu, rhel, cos, amazonlinux, talos)
   - Number of GPU nodes in the cluster

pkg/fingerprint/doc.go

@@ -18,7 +18,7 @@
 //
 // A Fingerprint records the cluster-identity dimensions used to bind
 // a snapshot to a recipe — service (eks/gke/aks/oke/kind/lke/bcm),
-// accelerator (h100/gb200/b200/a100/l40/rtx-pro-6000), OS
+// accelerator (h100/h200/gb200/b200/a100/l40/rtx-pro-6000), OS
 // (ubuntu/rhel/cos/amazonlinux/talos plus raw VERSION_ID), Kubernetes
 // server version, region, total node count, and GPU node count. Each
 // dimension records the resolved value plus an optional source string

pkg/fingerprint/gpu_sku.go

@@ -28,6 +28,13 @@ var gpuSKURegistry = []struct {
 	{"GB200", "gb200"},
 	{"B200", "b200"},
 	{"H100", "h100"},
+	// GH200 (Grace Hopper Superchip) contains the substring "H200" but is a
+	// distinct Grace-Hopper SKU, not the discrete H200 GPU. It has no
+	// first-class accelerator enum, so match it explicitly before the "H200"
+	// rule and leave it unresolved ("") rather than mislabeling it as h200 —
+	// same reason "GB200" precedes "B200" above.
+	{"GH200", ""},
+	{"H200", "h200"},
 	{"A100", "a100"},
 	{"RTX PRO 6000", "rtx-pro-6000"},
 	{"L40", "l40"},

pkg/fingerprint/gpu_sku_test.go

@@ -24,6 +24,10 @@ func TestParseGPUSKU(t *testing.T) {
 	}{
 		{"H100 80GB HBM3", "NVIDIA H100 80GB HBM3", "h100"},
 		{"H100 PCIe", "NVIDIA H100 PCIe", "h100"},
+		{"H200 NVL", "NVIDIA H200 NVL", "h200"},
+		{"H200 141GB HBM3e", "NVIDIA H200 141GB HBM3e", "h200"},
+		{"GH200 not matched as h200", "NVIDIA GH200 480GB", ""},
+		{"GH200 Grace Hopper", "NVIDIA GH200", ""},
 		{"GB200 NVL72", "NVIDIA GB200", "gb200"},
 		{"GB200 wins over B200 substring", "NVIDIA GB200 NVL72", "gb200"},
 		{"B200", "NVIDIA B200", "b200"},

pkg/fingerprint/types.go

@@ -70,7 +70,7 @@ type Fingerprint struct {
 	// k8s.node.provider (parsed from spec.providerID).
 	Service Dimension `json:"service" yaml:"service"`
 
-	// Accelerator is the GPU SKU (h100, gb200, b200, a100, l40,
+	// Accelerator is the GPU SKU (h100, h200, gb200, b200, a100, l40,
 	// rtx-pro-6000). Parsed from gpu.smi.gpu.model.
 	Accelerator Dimension `json:"accelerator" yaml:"accelerator"`
 

pkg/recipe/criteria.go

@@ -114,6 +114,7 @@ type CriteriaAcceleratorType string
 const (
 	CriteriaAcceleratorAny        CriteriaAcceleratorType = "any"
 	CriteriaAcceleratorH100       CriteriaAcceleratorType = "h100"
+	CriteriaAcceleratorH200       CriteriaAcceleratorType = "h200"
 	CriteriaAcceleratorGB200      CriteriaAcceleratorType = "gb200"
 	CriteriaAcceleratorB200       CriteriaAcceleratorType = "b200"
 	CriteriaAcceleratorA100       CriteriaAcceleratorType = "a100"
@@ -130,6 +131,8 @@ func ParseCriteriaAcceleratorType(s string) (CriteriaAcceleratorType, error) {
 		return CriteriaAcceleratorAny, nil
 	case "h100":
 		return CriteriaAcceleratorH100, nil
+	case "h200":
+		return CriteriaAcceleratorH200, nil
 	case "gb200":
 		return CriteriaAcceleratorGB200, nil
 	case "b200":
@@ -152,7 +155,7 @@ func ParseCriteriaAcceleratorType(s string) (CriteriaAcceleratorType, error) {
 // types sorted alphabetically. For the union of static + registry, use
 // AllCriteriaAcceleratorTypes.
 func GetCriteriaAcceleratorTypes() []string {
-	return []string{"a100", "b200", "gb200", "h100", "l40", "rtx-pro-6000"}
+	return []string{"a100", "b200", "gb200", "h100", "h200", "l40", "rtx-pro-6000"}
 }
 
 // AllCriteriaAcceleratorTypes returns the union of the static OSS list
@@ -347,7 +350,7 @@ type Criteria struct {
 	// Service is the Kubernetes service type (eks, gke, aks, oke, kind, lke, bcm).
 	Service CriteriaServiceType `json:"service,omitempty" yaml:"service,omitempty"`
 
-	// Accelerator is the GPU/accelerator type (h100, gb200, b200, a100, l40, rtx-pro-6000).
+	// Accelerator is the GPU/accelerator type (h100, h200, gb200, b200, a100, l40, rtx-pro-6000).
 	Accelerator CriteriaAcceleratorType `json:"accelerator,omitempty" yaml:"accelerator,omitempty"`
 
 	// Intent is the workload intent (training, inference).

pkg/recipe/criteria_registry_parse_test.go

@@ -102,16 +102,16 @@ func TestParseCriteriaServiceType_StrictAcceptsEmbedded(t *testing.T) {
 
 func TestParseCriteriaAcceleratorType_RegistryFallback(t *testing.T) {
 	withRegistry(t, func() {
-		if _, err := ParseCriteriaAcceleratorType("h200"); err == nil {
+		if _, err := ParseCriteriaAcceleratorType("mi300x"); err == nil {
 			t.Fatal("expected error before registration")
 		}
-		DefaultRegistry().Register(FieldAccelerator, "h200", OriginExternal)
-		got, err := ParseCriteriaAcceleratorType("h200")
+		DefaultRegistry().Register(FieldAccelerator, "mi300x", OriginExternal)
+		got, err := ParseCriteriaAcceleratorType("mi300x")
 		if err != nil {
 			t.Fatalf("error = %v", err)
 		}
-		if got != CriteriaAcceleratorType("h200") {
-			t.Errorf("got = %q, want h200", got)
+		if got != CriteriaAcceleratorType("mi300x") {
+			t.Errorf("got = %q, want mi300x", got)
 		}
 	})
 }
@@ -181,7 +181,7 @@ func TestAllCriteriaServiceTypes_UnionWithRegistry(t *testing.T) {
 
 func TestAllCriteriaTypes_AllDimensions(t *testing.T) {
 	withRegistry(t, func() {
-		DefaultRegistry().Register(FieldAccelerator, "h200", OriginExternal)
+		DefaultRegistry().Register(FieldAccelerator, "mi300x", OriginExternal)
 		DefaultRegistry().Register(FieldIntent, "fine-tuning", OriginExternal)
 		DefaultRegistry().Register(FieldOS, "bottlerocket", OriginExternal)
 		DefaultRegistry().Register(FieldPlatform, "nvmesh", OriginExternal)
@@ -192,7 +192,7 @@ func TestAllCriteriaTypes_AllDimensions(t *testing.T) {
 				t.Errorf("AllCriteria%sTypes() missing %q; got %v", field, want, got)
 			}
 		}
-		assertContains("Accelerator", AllCriteriaAcceleratorTypes(), "h200")
+		assertContains("Accelerator", AllCriteriaAcceleratorTypes(), "mi300x")
 		assertContains("Intent", AllCriteriaIntentTypes(), "fine-tuning")
 		assertContains("OS", AllCriteriaOSTypes(), "bottlerocket")
 		assertContains("Platform", AllCriteriaPlatformTypes(), "nvmesh")
@@ -212,11 +212,11 @@ func TestMergeCriteriaTypes_NoMutationOfInput(t *testing.T) {
 func TestCriteriaValidate_AdmitsRegisteredValues(t *testing.T) {
 	withRegistry(t, func() {
 		DefaultRegistry().Register(FieldService, "ncp-customer-x", OriginExternal)
-		DefaultRegistry().Register(FieldAccelerator, "h200", OriginExternal)
+		DefaultRegistry().Register(FieldAccelerator, "mi300x", OriginExternal)
 
 		c := &Criteria{
 			Service:     "ncp-customer-x",
-			Accelerator: "h200",
+			Accelerator: "mi300x",
 			Intent:      CriteriaIntentTraining,
 			OS:          CriteriaOSUbuntu,
 		}

pkg/recipe/criteria_registry_test.go

@@ -314,7 +314,7 @@ func TestSeedCriteriaRegistry(t *testing.T) {
 
 			c := &Criteria{
 				Service:     "ncp-x",
-				Accelerator: "h200",
+				Accelerator: "mi300x",
 				Intent:      "fine-tuning",
 				OS:          "bottlerocket",
 				Platform:    "nvmesh",
@@ -326,7 +326,7 @@ func TestSeedCriteriaRegistry(t *testing.T) {
 				value string
 			}{
 				{FieldService, "ncp-x"},
-				{FieldAccelerator, "h200"},
+				{FieldAccelerator, "mi300x"},
 				{FieldIntent, "fine-tuning"},
 				{FieldOS, "bottlerocket"},
 				{FieldPlatform, "nvmesh"},

pkg/recipe/criteria_test.go

@@ -72,6 +72,8 @@ func TestParseCriteriaAcceleratorType(t *testing.T) {
 		{"any", "any", CriteriaAcceleratorAny, false},
 		{"h100", "h100", CriteriaAcceleratorH100, false},
 		{"H100 uppercase", "H100", CriteriaAcceleratorH100, false},
+		{"h200", "h200", CriteriaAcceleratorH200, false},
+		{"H200 uppercase", "H200", CriteriaAcceleratorH200, false},
 		{"gb200", "gb200", CriteriaAcceleratorGB200, false},
 		{"b200", "b200", CriteriaAcceleratorB200, false},
 		{"a100", "a100", CriteriaAcceleratorA100, false},
@@ -725,7 +727,7 @@ func TestGetCriteriaAcceleratorTypes(t *testing.T) {
 	types := GetCriteriaAcceleratorTypes()
 
 	// Should return sorted list
-	expected := []string{"a100", "b200", "gb200", "h100", "l40", "rtx-pro-6000"}
+	expected := []string{"a100", "b200", "gb200", "h100", "h200", "l40", "rtx-pro-6000"}
 	if len(types) != len(expected) {
 		t.Errorf("GetCriteriaAcceleratorTypes() returned %d types, want %d", len(types), len(expected))
 	}