NVIDIA vGPU for Compute Features

NVIDIA vGPU (Virtual GPU) for Compute virtualizes NVIDIA GPUs for AI, machine learning, and high-performance computing. The subsections below describe MIG (Multi-Instance GPU) partitioning, provisioning, data paths, migration, multi-GPU guests, interconnects, scheduling, power state, and unified memory.

Table 18 Feature Summary

Feature	Description
MIG-Backed vGPU	Hardware-level GPU partitioning with spatial isolation for multi-tenant workloads
Device Groups	Automated topology-aware device provisioning for multi-GPU and GPU-NIC pairs
GPUDirect RDMA and Storage	Direct memory access and storage I/O bypass between GPUs and network/storage devices
Heterogeneous vGPU	Mixed vGPU profiles with different framebuffer sizes on a single GPU
Live Migration	Zero-downtime VM migration between physical hosts
Multi-vGPU and P2P	Multiple vGPUs per VM with peer-to-peer NVLink communication
NVIDIA NVSwitch	High-bandwidth GPU-to-GPU interconnect fabric for HGX systems
NVLink Multicast	Efficient one-to-many data distribution across NVLink-connected GPUs
Scheduling Policies	Workload-specific GPU scheduling algorithms (Best Effort, Equal Share, Fixed Share)
Suspend-Resume	VM state preservation and resumption without losing GPU context
Unified Virtual Memory	Single memory address space across CPU and GPU for simplified CUDA programming