> ## Documentation Index > Fetch the complete documentation index at: https://docs.costgraph.ai/llms.txt > Use this file to discover all available pages before exploring further. # Get an accelerator > Returns one accelerator held by the tenant selected by the X-CostGraph-Tenant-ID header: its model, vendor and provider, where it sits, its cost_summary where the catalog prices it, whether it is carved into MIG partitions with how many of them ran no compute, its compute and memory utilisation, its MIG partitions with the ones that ran no compute marked idle, and reports carrying every measurement recorded against it with a row per partition where the card is carved up. Every measured field is null rather than zero where nothing was measured, and every field the card also carries in the list carries the same name and the same value here. placement says where the accelerator sits and carries the ids a caller links on: placement.instance_id always, and placement.cluster_id, placement.node_id and placement.node_name where the machine is a Kubernetes node we hold, which is also what makes placement.kind workload rather than instance. cost_summary carries the currency every amount is in, and the hourly and projected monthly rate the catalog prices the card at; mtd and change_percent are always null with mtd_reason saying why, since spend to date is billed for the whole machine, so it is never attributed to one accelerator and never rendered as zero. A uuid the tenant has never reported gets a 404, and the accelerators on a machine that have not reported are only ever counted in the list's aggregate item, so they have no detail of their own. Where the same uuid has reported against more than one machine, the card the list ranks first wins and its reports are the ones from that same machine; filter the list by instance_id to reach the others. serving names the model-serving deployment running on the card, resolved from the pod the accelerator itself reports rather than from any name: serving.deployment_id is set only in state resolved, and addresses the same deployment /ai/deployments lists, so its serving series can be read from /ai/deployments/{id}/metrics. Every other state leaves deployment_id null and says why in serving.reason - shared when more than one serving deployment holds the card, not_serving when the pod holding it serves no model, unheld when no pod reported against it, partitioned when the card is carved up and its partitions report no pod, and unavailable when the lookup could not be made at all, which includes a card whose machine is not a Kubernetes node we hold. serving_holders carries every holder of the card rather than the single one serving resolves to, so a shared card can be charted in full. Each holder is named the way its placement names it: a card on a Kubernetes node we hold names its holders by namespace, pod and container with process_name and process_id null, and a card on any other machine names them by the process on the host, carrying process_name and process_id with namespace, pod and container null. mig_instance_id names the partition the holder occupies and is null where the holder holds the whole card. deployment_id and deployment_name are set only where the holder is a model-serving deployment we measure, and address the same deployment /ai/deployments lists. A card nothing holds carries an empty list rather than null, with serving.reason saying why. ## OpenAPI ````yaml /api-reference/costgraph/openapi.json get /api/v1/tenant/gpus/{gpu_uuid} openapi: 3.0.0 info: description: Read and manage your CostGraph organization, spend, alerts, and settings. title: CostGraph API contact: {} version: '1.0' servers: - url: https://api.costgraph.ai security: [] tags: - name: ai x-group: AI - name: ai-serving x-group: AI serving - name: anomalies x-group: Anomalies - name: auth x-group: Auth - name: billing x-group: Billing - name: billing-export x-group: Billing export - name: budgets x-group: Budgets - name: ci x-group: CI - name: compute-recommendations x-group: Compute recommendations - name: config x-group: Config - name: cost x-group: Cost - name: gpus x-group: GPUs - name: graphai x-group: Graph AI - name: infracost x-group: Infracost - name: integrations x-group: Integrations - name: invitations x-group: Invitations - name: kubernetes-clusters x-group: Kubernetes clusters - name: marketplace x-group: Marketplace - name: network-requests x-group: Network requests - name: notifications x-group: Notifications - name: oauth x-group: OAuth - name: oauth-clients x-group: OAuth clients - name: opencost x-group: OpenCost - name: organization x-group: Audit log - name: organizations x-group: Organizations - name: placement-alternatives x-group: Placement alternatives - name: reports x-group: Reports - name: service-map x-group: Service map - name: settings x-group: Settings - name: sso x-group: Single sign-on - name: tenants x-group: Tenants - name: user x-group: Users - name: virtual-machines x-group: Virtual machines - name: virtual-tags x-group: Virtual tags - name: workflows x-group: Workflows paths: /api/v1/tenant/gpus/{gpu_uuid}: get: tags: - gpus summary: Get an accelerator description: >- Returns one accelerator held by the tenant selected by the X-CostGraph-Tenant-ID header: its model, vendor and provider, where it sits, its cost_summary where the catalog prices it, whether it is carved into MIG partitions with how many of them ran no compute, its compute and memory utilisation, its MIG partitions with the ones that ran no compute marked idle, and reports carrying every measurement recorded against it with a row per partition where the card is carved up. Every measured field is null rather than zero where nothing was measured, and every field the card also carries in the list carries the same name and the same value here. placement says where the accelerator sits and carries the ids a caller links on: placement.instance_id always, and placement.cluster_id, placement.node_id and placement.node_name where the machine is a Kubernetes node we hold, which is also what makes placement.kind workload rather than instance. cost_summary carries the currency every amount is in, and the hourly and projected monthly rate the catalog prices the card at; mtd and change_percent are always null with mtd_reason saying why, since spend to date is billed for the whole machine, so it is never attributed to one accelerator and never rendered as zero. A uuid the tenant has never reported gets a 404, and the accelerators on a machine that have not reported are only ever counted in the list's aggregate item, so they have no detail of their own. Where the same uuid has reported against more than one machine, the card the list ranks first wins and its reports are the ones from that same machine; filter the list by instance_id to reach the others. serving names the model-serving deployment running on the card, resolved from the pod the accelerator itself reports rather than from any name: serving.deployment_id is set only in state resolved, and addresses the same deployment /ai/deployments lists, so its serving series can be read from /ai/deployments/{id}/metrics. Every other state leaves deployment_id null and says why in serving.reason - shared when more than one serving deployment holds the card, not_serving when the pod holding it serves no model, unheld when no pod reported against it, partitioned when the card is carved up and its partitions report no pod, and unavailable when the lookup could not be made at all, which includes a card whose machine is not a Kubernetes node we hold. serving_holders carries every holder of the card rather than the single one serving resolves to, so a shared card can be charted in full. Each holder is named the way its placement names it: a card on a Kubernetes node we hold names its holders by namespace, pod and container with process_name and process_id null, and a card on any other machine names them by the process on the host, carrying process_name and process_id with namespace, pod and container null. mig_instance_id names the partition the holder occupies and is null where the holder holds the whole card. deployment_id and deployment_name are set only where the holder is a model-serving deployment we measure, and address the same deployment /ai/deployments lists. A card nothing holds carries an empty list rather than null, with serving.reason saying why. parameters: - description: Tenant ID name: X-CostGraph-Tenant-ID in: header required: true schema: type: string - description: Accelerator UUID as the device reports it name: gpu_uuid in: path required: true schema: type: string responses: '200': description: OK content: application/json: schema: allOf: - $ref: '#/components/schemas/responses.SuccessResponse' - type: object properties: data: $ref: '#/components/schemas/db.GPUDetail' '400': description: Bad Request content: application/json: schema: $ref: '#/components/schemas/responses.ErrorResponse' '401': description: Unauthorized content: application/json: schema: $ref: '#/components/schemas/responses.ErrorResponse' '403': description: Forbidden content: application/json: schema: $ref: '#/components/schemas/responses.ErrorResponse' '404': description: Not Found content: application/json: schema: $ref: '#/components/schemas/responses.ErrorResponse' '500': description: Internal Server Error content: application/json: schema: $ref: '#/components/schemas/responses.ErrorResponse' security: - BearerAuth: [] components: schemas: responses.SuccessResponse: type: object required: - message - status properties: data: {} message: type: string example: some message status: type: string example: success db.GPUDetail: type: object properties: cost_summary: $ref: '#/components/schemas/db.GPUCost' gpu_availability: type: object additionalProperties: type: string gpu_uuid: type: string idle_partition_count: type: integer instance_id: type: string memory_utilization_percent: type: number mig_enabled: type: boolean model: type: string partition_count: type: integer partitions: type: array items: $ref: '#/components/schemas/db.GPUPartition' placement: $ref: '#/components/schemas/db.GPUPlacement' provider: type: string reports: type: array items: $ref: '#/components/schemas/db.GPUReportRow' serving: $ref: '#/components/schemas/db.GPUServing' serving_holders: type: array items: $ref: '#/components/schemas/db.GPUServingHolder' utilization_percent: type: number vendor: type: string responses.ErrorResponse: type: object required: - message - status properties: message: type: string example: some message status: type: string example: error db.GPUCost: type: object properties: change_percent: type: number currency: type: string hourly_rate: type: number mtd: type: number mtd_reason: type: string projected_monthly_spend: type: number db.GPUPartition: type: object properties: idle: type: boolean mig_instance_id: type: string mig_profile: type: string utilization_band: type: string utilization_p95: type: number db.GPUPlacement: type: object properties: cluster: type: string cluster_id: type: string instance_id: type: string kind: type: string name: type: string namespace: type: string node_id: type: string node_name: type: string db.GPUReportRow: type: object properties: action: type: string created_at: type: string data_age_hours: type: number data_quality_state: type: string gpu_count: type: number gpu_index: type: number gpu_model: type: string gpu_uuid: type: string id: type: string instance_id: type: string last_sample_age_seconds: type: number metric_presence_flags: type: array items: type: string mig_instance_id: type: string mig_profile: type: string observed: $ref: '#/components/schemas/schema.StatsRow' observed_stats_id: type: string partial_windows_flags: type: array items: type: string qos_value_type_id: type: string reasons: type: object stats_30d_id: type: string stats_7d_id: type: string tenant_id: type: string updated_at: type: string urgency: type: string utilization_band: type: string value_type: type: string window_30d: $ref: '#/components/schemas/schema.StatsRow' window_7d: $ref: '#/components/schemas/schema.StatsRow' db.GPUServing: type: object properties: deployment_id: type: string deployment_name: type: string namespace: type: string reason: type: string state: type: string db.GPUServingHolder: type: object properties: container: type: string deployment_id: type: string deployment_name: type: string mig_instance_id: type: string namespace: type: string pod: type: string process_id: type: string process_name: type: string schema.StatsRow: type: object properties: id: type: string last: type: number max: type: number mean: type: number min: type: number p50: type: number p90: type: number p95: type: number p99: type: number std_dev: type: number tenant_id: type: string securitySchemes: BearerAuth: description: Enter "Bearer {token}" type: apiKey name: Authorization in: header ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.