GPU Support

Every sandbox gets a full dedicated GPU. No sharing, no throttling.

Supported GPUs

GPU ModelVRAMUse Cases
T416 GBInference, fine-tuning small models
A10G24 GBTraining, medium model inference
L40S48 GBLarge model training and inference
H10080 GBLarge-scale training, LLM inference

CPU-only sandboxes

Sandboxes also run without a GPU for general compute workloads. CPU sandboxes are billed per core-hour, with memory billed separately per GiB-hour.