
The Heimdall Automated Productivity Service (APS) is a suite of cloud-native services developed by NVIDIA to enhance the productivity and reliability of AI training on large-scale supercomputers. It enables efficient AI cluster management by detecting, isolating, and recovering from failures autonomously, reducing the manual effort required for large training jobs that can involve thousands of GPUs. Heimdall’s architecture integrates seamlessly with Slurm, a popular workload manager, providing high fault tolerance and scalability.
Licenses for included OSS components: See "Third-Party Open Source Software Licenses" section of the Autonomous Job Recovery Installation Guide.
Source code for included OSS components: Located inside the containers and at https://developer.nvidia.com/downloads/ajr-nmc/ajr_1.5.0_oss_sources.tar.gz