Mixed GPU Partition Layout

Data artifactInfrastructureInfrastructurearc:MixedGPUPartitionLayout

A GPU partition layout that assigns different partition profiles to different GPUs in one cluster to match heterogeneous model sizes or service tiers.

Responsibility. Declares a heterogeneous partition geometry sized to the observed demand per tier.

Also known as: Strategy mixed, Heterogeneous partitioning

Variant of GPU Partition Layout abstract

When to choose. Choose for multi-tier offerings and diverse model sizes (7B-70B) when the team accepts extra scheduling and configuration complexity for better utilisation.

Direct neighbourhood (hover for relationship types)

Relationships

configures structural

GPU Partition Ch7.6

alternative to variability

Design guidance

SHOULD derive the profile mix from the observed demand distribution and keep about 10% headroom for spikes or new tiers.
SHOULD monitor utilisation per profile to detect partitions idling while requests queue for another profile.

Quantitative guidance

As stated by the sources; verify before use.

Three-tier example on 10 A100-80GB: 4x3g.39gb, 9x2g.20gb, 35x1g.10gb for 5/15/40 tenants (Ch7.6).

Classification

Quality attributes: Performance efficiency (ISO/IEC 25010)Flexibility (ISO/IEC 25010)
Risks mitigated: Mismatched partition sizes for larger tenant models

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.