Format of runpod.create_endpoint gpu_ids | Runpod | Page 1

snow spindle Jul 19, 2025, 9:34 AM

#

It's unclear what the format is, the website is using this format: {
"data": {
"saveEndpoint": {
"gpuIds": "['NVIDIA GeForce RTX 4090'],ADA_24",
"id": "ncipvwdj6cuz8y",
"idleTimeout": 5,
"locations": null,
"name": "7891c33062e411f095a9000000b9f2e1",
"type": "QB",
"networkVolumeId": null,
"scalerType": "QUEUE_DELAY",
"scalerValue": 4,
"templateId": "tsef8csm02",
"userId": "google-oauth2|100781309549972343660",
"workersMax": 3,
"workersMin": 0,
"gpuCount": 1,
"__typename": "Endpoint"
}
}
}

When I just pass NVIDIA GeForce RTX 4090 to gpuIds and check the UI, nothing is selected, only when I select a GPU manually it persists.

inner hazel Jul 19, 2025, 9:42 AM

#

yeah graphql / REST API request doesnt automatically interact with your UI, what are you trying to do?

snow spindle Jul 19, 2025, 9:47 AM

#

I have a serverless deploy script that creates endpoints programmatically, it creates the endpoint but gpu IDs is not set correctly

inner hazel Jul 19, 2025, 10:10 AM

#

What's your request like

snow spindle Jul 19, 2025, 12:41 PM

#

mutation {
saveEndpoint(
input: {
name: "7891c33062e411f095a9000000b9f2e1", templateId: "tsef8csm02", gpuIds: "['NVIDIA GeForce RTX 4090']", networkVolumeId: "", locations: "", idleTimeout: 5, scalerType: "QUEUE_DELAY", scalerValue: 4, workersMin: 0, workersMax: 3, gpuCount: 1
}
) {
id
name
templateId
gpuIds
networkVolumeId
locations
idleTimeout
scalerType
scalerValue
workersMin
workersMax
allowedCudaVersions
gpuCount
}
}

snow spindle Jul 19, 2025, 7:20 PM

#

figured out the IDs by reverse engineering: class GPUType(Enum):
"""GPU types available in RunPod using their official GPU ID codes."""
# Ada Lovelace Architecture
ADA_24 = "ADA_24" # L4, RTX 4000 series consumer cards
ADA_32_PRO = "ADA_32_PRO" # Professional Ada cards with 32GB
ADA_48_PRO = "ADA_48_PRO" # L40, L40S, RTX 6000 Ada
ADA_80_PRO = "ADA_80_PRO" # High-end Ada professional cards

# Ampere Architecture
AMPERE_16 = "AMPERE_16"  # RTX 3060, A2000, A4000
AMPERE_24 = "AMPERE_24"  # RTX 3070/3080/3090, A4500, A5000
AMPERE_48 = "AMPERE_48"  # A40, RTX A6000
AMPERE_80 = "AMPERE_80"  # A100 80GB

# Hopper Architecture
HOPPER_141 = "HOPPER_141"  # H200 with 141GB memory

inner hazel Jul 20, 2025, 2:14 AM

#

I think it's in the docs which gpu id's you can use

#

For next time might worth searching

snow spindle Jul 20, 2025, 9:15 AM

#

ok, just realized the REST API is documented and I was using the GraphQL API as this is being used by the runpod sdk. is probably worth updating the runpod sdk to match the REST API or is the sdk depcrated for making API calls?

smoky cosmos Jul 21, 2025, 3:49 PM

#

We're working on getting upgrades out for runpod-python to use the REST API 👍

#Format of runpod.create_endpoint gpu_ids