GPU Forge — operator console

Automatic Scaling Min Ready Workers Target Max Wait (s) Pending-start Window (s)

GPU Speed

Average seconds per image per GPU. Worker capacity = target wait ÷ seconds.

Nebius Cap Nebius Grace Seconds 0 = never release. Nebius GPU Waterfall Nebius Service Account JSON Base64 Nebius Project ID Nebius Worker Port Worker Label Key Worker Label Value Provider Label Key Provider Label Value

RunPod Cap RunPod Grace Seconds 0 = never release. RunPod API Key RunPod Name Prefix RunPod Env Key RunPod Env Value RunPod Deploy Env RunPod Template ID RunPod GPU Waterfall RunPod Region Volumes RunPod Container Disk GB RunPod Ports RunPod Volume Mount

Vast Cap Vast Grace Seconds 0 = never release. Vast API Key Vast GPU Waterfall Vast Excluded GPUs Vast Max $ / Hour Vast Disk GB Vast Image Vast Runtype

Local Cap Local Grace Seconds 0 = keep leased (warm). Seconds to release idle local GPUs, down to Min Ready Workers. Local GPU Waterfall Preference order for picking an idle local GPU (best first). Unlisted GPUs are still used, just last.

GPU workers

Status

Provider Waterfall

Provider Settings

GPU Speed