Massed Compute

What is Massed Compute?

Massed Compute provides on-demand and dedicated NVIDIA GPU infrastructure for machine learning, rendering, and high-performance computing. It is relevant to developers and teams needing access to enterprise GPUs ranging from A30 to B300 without bandwidth caps or contracts. Access methods include hourly web instances, single-tenant bare metal servers, node clusters, and a programmatic API.

What are the features of Massed Compute?

• On-Demand Instances

  Hourly GPU rentals starting at $0.43 per hour with pre-installed CUDA, free egress, and an included VDI session.
• Bare Metal Servers

  Single-tenant physical hardware isolated on-premises, with setup taking one to three days.
• GPU Clusters

  Multi-node configurations of 8 to over 512 nodes connected via InfiniBand topology, requiring one to two weeks for setup.
• API & MCP

  REST endpoint and MCP server for automated provisioning using bearer keys, scoped tokens, and per-project quotas.
• NVIDIA Hardware Catalog

  Direct access to Blackwell, Hopper, Ada Lovelace, and Ampere SKUs shipping with vendor-tested drivers.
• Pre-built Templates

  Curated instance images including PyTorch with DeepSpeed, vLLM inference, ComfyUI with Flux.1, Ollama, and JupyterLab.

What are the use cases of Massed Compute?

• Training machine learning models using the PyTorch template configured with CUDA 12.4, DeepSpeed, and FlashAttention 2.
• Running large language model inference locally through the vLLM template with OpenAI API compatibility and paged attention.
• Automating GPU deployment within a CI pipeline or Python notebook using the REST API or provided SDKs for Python, Node.js, Go, and Terraform.

How to use Massed Compute?

Web workflow:

• Select a curated template such as PyTorch + CUDA or bring a custom image.
• Launch the GPU instance through the web interface.
• Access the running environment via the terminal or the included one-click VDI session.

Massed Compute Pricing Information:

• On-Demand: $0.43+ per hour; minimum term of 1 hour; approximately 90-second setup time; includes VDI, pre-installed CUDA, and free egress.
• Bare Metal: Price not stated; single-tenant tenancy; 1–3 day setup time; HIPAA and SOC 2 compliance supported.
• GPU Clusters: Quote required; 8–512+ nodes; InfiniBand topology; 1–2 week setup time; dedicated support included.
• API & MCP: Wholesale pricing; full catalog access via REST endpoint authenticated with an API key.