GPU types

The GPU catalog an application's configuration selects from, with pricing.

Introduction

This is the catalog an app's configuration selects from. A gpuType the catalog does not carry is rejected when you create or scale an app, and the sizes each entry allows are what gpusPerWorker is checked against.

Rates are per GPU-hour and are what pay-as-you-go time is billed at. See Pricing for how that time is measured and Compute and scaling for choosing between them.

List GPU types

GETapi.serverless.runware.ai/v1/gpu-types

Returns the global GPU type catalog and pricing. The request requires authentication. Customer principals receive only GPU types with capacity currently offered to customers. A type whose hardware is not yet cleared for customer workloads is omitted. The Runware principal receives the full catalog, including retired types and types not yet offered. Retired entries carry deletedAt. Each entry's pricing is the price currently in effect.

Response

data object[]required

Array items6 properties each

id stringrequiredmin: 1max: 64: Public catalog code referenced by gpuType in worker configuration (not the internal database row UUID).

name stringrequiredmin: 1

memory stringrequiredmin: 1: Human-readable memory capacity.

availability stringrequired: How the catalog currently offers this GPU type. restricted means the SKU is not generally available. It is not purchased reserved capacity.
Possible values3 values

deletedAt string (date-time)nullabledate-time: Retirement time, returned only in the Runware catalog view.

pricing objectrequired

The price currently in effect, the most recent price whose effectiveFrom is not in the future. The Runware principal can read the full price history, including scheduled future changes, from GET /v1/gpu-types/{gpuTypeId}/prices.

Properties3 properties

id string (uuid)requiredUUID v4

perSecond stringrequiredmin: 1: Price per GPU per second in USD major units as an exact decimal string (e.g. "0.000767"). Stored and billed from nanodollars internally. The platform bills in USD only.

effectiveFrom string (date-time)requireddate-time: From when this price is effective.

Errors

Status	When
`401`	Missing or invalid credentials
`500`	Unexpected server error
`503`	A required service is temporarily unavailable

Get a GPU type

GETapi.serverless.runware.ai/v1/gpu-types/{gpuTypeId}

Returns an active entry from the global GPU type catalog. Retired entries return 404. The result is not organization-specific, but the request still requires authentication.

Request

Path

gpuTypeId stringrequiredmin: 1max: 64: Public catalog code for a GPU type (e.g. h100), not the internal row UUID.

Response

id stringrequiredmin: 1max: 64: Public catalog code referenced by gpuType in worker configuration (not the internal database row UUID).

name stringrequiredmin: 1

memory stringrequiredmin: 1: Human-readable memory capacity.

availability stringrequired: How the catalog currently offers this GPU type. restricted means the SKU is not generally available. It is not purchased reserved capacity.
Possible values3 values

deletedAt string (date-time)nullabledate-time: Retirement time, returned only in the Runware catalog view.

pricing objectrequired

Properties3 properties

id string (uuid)requiredUUID v4

perSecond stringrequiredmin: 1: Price per GPU per second in USD major units as an exact decimal string (e.g. "0.000767"). Stored and billed from nanodollars internally. The platform bills in USD only.

effectiveFrom string (date-time)requireddate-time: From when this price is effective.

Errors

Status	When
`401`	Missing or invalid credentials
`404`	Resource not found
`422`	The request was well-formed but semantically invalid (e.g. a missing or out-of-range field). A request that could not be parsed returns `400`. The `errors` array carries one entry per offending field.
`500`	Unexpected server error
`503`	A required service is temporarily unavailable