Automatic ComputeDomain Management (NVIDIA MNNVL)
Automatic NVIDIA DRA ComputeDomain lifecycle and pod claim injection for Multi-Node NVLink (MNNVL) workloads across LeaderWorkerSet and DisaggregatedSet.
Accelerators
Accelerator-specific integrations across LeaderWorkerSet and DisaggregatedSet.
Core LeaderWorkerSet and DisaggregatedSet primitives — such as group lifecycle management, subgroups, exclusive topology placement, slice placement policies, and gang scheduling — are hardware-neutral and work across GPUs, TPUs, and other AI accelerators using standard Kubernetes resources and node labels.
In addition to those portable primitives, LWS provides built-in integrations for accelerator platforms that require vendor-specific pod wiring or cluster resources across both LeaderWorkerSet and DisaggregatedSet workloads.
Feedback
Was this page helpful?
Glad to hear it! Please tell us how we can improve.
Sorry to hear that. Please tell us how we can improve.