What is supercomputer?
Supercomputer provides GPU and CPU compute sessions that retain code and files between runs, allowing work to be stopped and resumed later. It is aimed at people running coding agents and GPU workloads, including inference, fine-tuning, reinforcement learning, and research experiments. Compute is charged from prepaid monthly credits while a session runs, with GPU options ranging from A10G and A100 to H100, B200, and B300.
supercomputer Features:
Persistent Sessions: GPU and CPU sessions keep code and files between runs, so you can stop work and return to the same session later.
Usage-Based Billing: Compute is billed by the second from your credits, and charges apply only while a session is running.
GPU Selection: Available GPU options include A10G, L40S, A100, H100, H200, B200, B300, and RTX PRO 6000.
Agent GPU Boxes: The service offers dedicated GPUs for coding agents, with a hard $0 stop.
Autoresearch Loops: Autoresearch provides experiment loops that can continue running while you sleep.
Async Workloads: Supercomputer supports asynchronous inference, fine-tuning, and reinforcement learning workloads.
supercomputer Benefits:
Resume Work: Keeping code and files across runs avoids having to recreate a working environment whenever you pause a compute session.
Control Spend: Second-based billing and the ability to stop compute when it is not needed help limit charges to active session time.
Choose Hardware: The range of listed GPUs lets you select from several hardware options for different compute needs.
Run Experiments: Loops that keep working unattended give researchers room to leave experiments running outside their active work hours.
Budget With Credits: Monthly plans add credits to your balance, giving you a set monthly amount to spend on compute.
Use Cases:
Agent Development: Give coding agents dedicated GPU compute while developing or testing agent-driven code.
Overnight Research: Set up autoresearch experiment loops to run while you are away from your workstation or asleep.
Model Inference: Run asynchronous inference workloads on GPU compute without keeping a session active when it is not in use.
Model Fine-Tuning: Use GPU sessions for fine-tuning jobs, then stop compute when the session is finished.
Reinforcement Learning: Run reinforcement learning workloads using the service's GPU compute options.
Forward Deployments: Use Supercomputer for forward deployments, one of the workload categories included among its products.