Get an accelerator type
Retrieve one accelerator type from the curated catalog, including its workload affinity, placements, and live market fields when available.
Authorizations
Gateway auth: send Authorization: Bearer <mka1-api-key>. For multi-user server-side integrations, you can also send X-On-Behalf-Of: <external-user-id>.
Headers
Optional external end-user identifier forwarded by the API gateway.
Path Parameters
Response
Successful Response
An offered accelerator type in the catalog.
Coarse accelerator class. npu/tpu are reserved for future catalog entries.
gpu, cpu Max accelerators of this type per replica. For class cpu the unit is vCPUs: the maximum rentable vCPU count per replica.
"H100"
Placements where this accelerator exists. All catalog entries are external (marketplace-acquired) today; internal arrives with private-cluster inventory.
internal, external "nvidia"
"amd"
"intel"
"aws"
Per-workload suitability ratings for an accelerator type, seeded from the training hardware matrix.
Banded estimate of how reliably matching capacity can be acquired, computed from availability history. "unknown" means no history, never a guess.
high, medium, low, unknown Configurations acquirable now (0 when offered is false). Absent when live market data is unavailable.
Deprecated: use class. Retained for SDK compatibility until the next breaking conventions pass.
gpu, npu, tpu, other Per-accelerator memory in GB. Absent for class cpu: CPU capacity is sized per request (min_vcpus/min_mem_gb), not by a fixed module.
Whether any capacity acquirable right now matches this type. Absent when live market data is unavailable.
USD-per-hour price band across configurations acquirable now for an accelerator type.