GPU SKUs
Available GPU configurations and how to select them.
A SKU picks how many GPUs of which type your session runs on. It follows the
pattern gpu_{count}x_{model} — for example gpu_8x_b200 is 8× NVIDIA B200.
Bigger model or higher count means more memory and more cost per hour. CPU-only
SKUs exist too (cpu_generic). The RTX PRO 6000 was previously listed as
gpu_{count}x_rtx6000; those names still work and resolve to
gpu_{count}x_rtx_pro_6000.
List everything currently available from the terminal:
super skusFind your SKU
Filter by GPU count, compare accelerator memory, and click any SKU to copy its
--sku value:
| SKU | GPUs | Memory |
|---|---|---|
| CPU only | — | |
| 1 × T4 | 16 GB | |
| 1 × A10G | 24 GB | |
| 1 × L4 | 24 GB | |
| 1 × L40S | 48 GB | |
| 4 × T4 | 64 GB(16 GB/GPU) | |
| 1 × H100 | 80 GB | |
| 4 × A10G | 96 GB(24 GB/GPU) | |
| 4 × L4 | 96 GB(24 GB/GPU) | |
| 1 × RTX PRO 6000 Blackwell | 96 GB | |
| 8 × T4 | 128 GB(16 GB/GPU) | |
| 1 × H200 | 141 GB | |
| 1 × B200 | 180 GB | |
| 8 × L4 | 192 GB(24 GB/GPU) | |
| 4 × L40S | 192 GB(48 GB/GPU) | |
| 2 × RTX PRO 6000 Blackwell | 192 GB(96 GB/GPU) | |
| 4 × RTX PRO 6000 Blackwell | 384 GB(96 GB/GPU) | |
| 8 × A100 80GB | 640 GB(80 GB/GPU) | |
| 8 × H100 | 640 GB(80 GB/GPU) | |
| 8 × RTX PRO 6000 Blackwell | 768 GB(96 GB/GPU) | |
| 8 × H200 | 1.1 TB(141 GB/GPU) | |
| 8 × B200 | 1.4 TB(180 GB/GPU) |
--sku valueUsing a SKU
Pass it as --sku (CLI) or sku= (SDK). Pick a GPU below to see the exact command:
pick a GPU above · compare all SKUs
# Python SDK
s = project.session(sku="gpu_1x_l4")
s.run(["nvidia-smi"])
s.close()The dispatcher routes your request to an available host matching the requested
SKU. If none are free, the session waits in preparing until one is provisioned.