Open Nav

Multi-Instance GPU Support for ML Workloads with on NVIDIA A100‚Äč

Provision, allocate, monitor & manage MIG instances with meta-scheduler

Deliver Accelerated ML Workloads of All Sizes with Multi-Instance GPU (MIG) is the first ML platform to integrate the NVIDIA multi-instance GPU (MIG) functionality of the NVIDIA A100 Tensor Core GPU. The MIG support is certified for NVIDIA DGX A100 systems as part of the NVIDIA DGX-Ready Software program and delivers on-demand access to MIG instances of one or more NVIDIA A100 GPUs for ML/DL workloads in one click. delivers MLOps, utilization, assignment and scheduling of MIG resources for ML workloads.

Maximize GPU Utilization

  • Achieve up to 7X more GPU instances on a single NVIDIA A100 GPU and up to 56X more on the 8 A100 GPUs in NVIDIA DGX A100
  • Allocate right-sized GPU with guaranteed quality of service (QoS) for every job
  • Enable inference, training, and HPC workloads to run at the same time on a single GPU
  • Automate MIG pool management and make instances available immediately after job is completed
MIG Flow

Accelerate ML Workflow

  • Deliver data scientists one click self-service access to MIG instances
  • Partition instances on-demand with pre-configured resource templates
  • Run simultaneous mixed workloads on independent instances
  • Automatically schedule containerized workloads within any MIG instance
MIG Flow

Maintain Flexibility and Control

  • Track utilization and constantly monitor the pool of MIG resources
  • Set a library of pre-configured MIG instance templates of different sizes
  • Standardize UI for all compute resources for simplified use
  • Leverage meta-scheduling of MIG resources for optimized availability
  • Simplify utilization of NVIDIA A100 and resource management for IT and DevOps teams