Infrastructure · Data artifact

GPU Partition Reconfiguration Policy

Data artifactInfrastructureInfrastructurearc:GPUPartitionReconfigurationPolicy

A declarative policy governing how partition-layout changes roll out: reconfiguration mode, validation delay between GPUs, and spare capacity preserved for evicted workloads.

Responsibility. Bounds the disruption caused by changing GPU partition geometry.

Also known as: drain-one-at-a-time reconfigure mode, N+1 capacity rule

Direct neighbourhood (hover for relationship types)

Relationships

constrains control

GPU Partition Manager Ch7.6

Design guidance

MUST keep at least N+1 GPU capacity so evicted workloads can reschedule during reconfiguration.
SHOULD keep 10-15% of GPU capacity unused as reconfiguration headroom.

Quantitative guidance

As stated by the sources; verify before use.

Example validation delay: 60 s between GPUs (Ch7.6).

Classification

Quality attributes: Reliability (ISO/IEC 25010 | NIST AI RMF: valid and reliable)
Risks mitigated: Cluster capacity exhaustion while a GPU is drained

Sources

Ch7.6: T. Nguyen, "Multi-Instance GPU," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 7.6. ISBN: 9798244538229.