GPUs you can launch.
Reserved cards stay on RunPod Secure and Hot Aisle. Vast.ai is spot: the host can take the machine back. Rates come from compute gpu list.
Reserved
These stay up for the run.
| GPU | Provider | VRAM | CLI |
|---|---|---|---|
H100-SXM | RunPod | 80 GB | --gpu H100 |
MI300X | Hot Aisle | 192 GB | --gpu MI300X |
Spot
The host can take the machine back.
| GPU | Provider | VRAM | CLI |
|---|---|---|---|
RTX-3090 | Vast.ai | 24 GB | --provider vastai --gpu RTX-3090 |
RTX-6000-ADA | Vast.ai | 48 GB | --provider vastai --gpu RTX-6000-ADA |
L4 | Vast.ai | 24 GB | --provider vastai --gpu L4 |
A100-40GB | Vast.ai | 40 GB | --provider vastai --gpu A100-40GB |
RTX-4090 | Vast.ai | 24 GB | --provider vastai --gpu RTX-4090 |
RTX-5090 | Vast.ai | 32 GB | --provider vastai --gpu RTX-5090 |
A100-80GB | Vast.ai | 80 GB | --provider vastai --gpu A100-80GB |
H100-SXM | Vast.ai | 80 GB | --provider vastai --gpu H100-SXM |
compute gpu list has the current rate. H100-SXM exists on RunPod and Vast.ai — pass --provider when you mean the spot one.