Welcome

New to Strong Compute? Start here.

Introduction to Strong Compute

Our vision is to be the command and control for compute—what Kubernetes/VMware should be if they were great for AI. So we've built the Strong Compute Instant Super Computer (ISC).

Instant Super Computer

Strong Compute’s ISC is designed for efficient AI model training, packaged into the ISC Control Plane (website) and ISC Portal (terminal). It offers two main modes: Cycling for pipeline development and Burst for intensive training.

Cycling: For AI Training Pipeline Development

• Resources: Always-on 72x24GB Ampere GPU cluster.

• Performance: Jobs start in 10 seconds (currently 20-120 seconds), with 90-second cycles per job.

• Benefits:

  • Develop your pipeline on a full cluster, saving weeks of time compared to single GPU/node development.

  • Low-cost error troubleshooting (< $10 each) versus $30K/week for a 72 GPU cluster.

  • Easy distributed training with Strong Compute’s tooling.

  • Controlled costs with no unexpected bills.

Burst: For AI Model Training

• Resources: An additional 72 GPU clusters, available as soon as your pipeline is ready.

• Performance: Clusters start in 2 minutes (currently 15 minutes).

• Benefits:

  • Quick training initiation (minutes, not weeks).

  • No GPU sourcing hassle (avoids lengthy negotiations and commitments).

  • Saves time on system administration and cluster configuration.

  • Significantly faster training (10x-100x) compared to always-on GPU sets at a similar cost.

  • Controlled costs with no unexpected bills.

Underlying Technology:

• Multi-Cloud Backend: Sources 72 GPU clusters from the best available options.

• Robust Checkpointing: Saves more than model checkpoints, enabling job suspension/resumption.

• Fast Dataset Transfer: Achieves 30GB/second (target 100GB/sec).

• Cost Control UI: Helps tightly scope and manage expenses.