gpustack

Tracked

Open-source GPU cluster manager for efficient model training and high-performance inference orchestration.

Author gpustack Open Sourced 2024-05-11 Last Commit Unknown

GPUStack is an open-source GPU cluster manager that configures and orchestrates inference engines like vLLM and SGLang for high-performance AI model deployment. It unifies heterogeneous GPU resources into a single orchestratable pool, providing device discovery, resource abstraction, and centralized scheduling to help teams run distributed training and low-latency inference with improved GPU utilization.

Resource Management

  • Automatic resource pooling and device discovery across CUDA and ROCm stacks
  • Identifies GPU model, memory, and driver details for optimal placement
  • Heterogeneous GPU support combining NVIDIA and AMD hardware in one cluster
  • Resource abstraction layer that simplifies multi-GPU orchestration

Intelligent Scheduling

  • Scheduling policies based on job requirements and priorities
  • Dynamic GPU allocation by request load for cost-effective inference serving
  • Multi-tenant isolation allowing safe GPU sharing across projects
  • Extensible plugin hooks for custom schedulers and monitoring integrations

Observability and Operations

  • Built-in metrics collection with Prometheus and Grafana integration
  • RESTful API and CLI for automation and operational management
  • Modular architecture supporting independent deployment of scheduler, monitoring, and access layers
  • Cloud-native design integrated with container ecosystems

Supported Workloads

  • Research and education clusters sharing GPUs without memory or card conflicts
  • Enterprise training platforms orchestrating large-scale distributed training
  • Online inference fleets requiring low-latency, high-throughput serving
  • Apache-2.0 licensed with comprehensive community documentation