Infrastructure · Infrastructure resource

Dedicated GPU Device

Infrastructure resourceInfrastructureInfrastructurearc:DedicatedGPUDevice

A whole, unpartitioned physical GPU allocated exclusively to one workload, giving it all compute, memory and memory bandwidth.

Responsibility. Provides the full capacity of one GPU to a single workload.

Also known as: Full GPU allocation, MIG strategy none

Variant of GPU Allocation Unit abstract

When to choose. Choose for large models (70B+), high-concurrency single-tenant deployments, or latency-critical workloads needing full memory bandwidth and large continuous batches.

Direct neighbourhood (hover for relationship types)

Relationships

hosts structural

Inference Server Ch7.6

is configured by structural

Unpartitioned GPU Layout Ch7.6

alternative to variability

Quantitative guidance

As stated by the sources; verify before use.

One GPU per tenant for 50 tenants needs 50 A100s ($750,000) at ~12% utilisation each (Ch7.6).
A single agent serving 10,000 concurrent users with sub-100ms SLA needs the full A100 for 32-64-request batches (Ch7.6).

Classification

Quality attributes: Performance efficiency (ISO/IEC 25010)

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.