Is this page helpful?

NVIDIA AI Enterprise Infra 4.9 Release Notes#

Release Highlights#

  • Multi-Architecture GPU Support - GPU Data Center Driver 535.288.01 with support for Blackwell, Hopper, Ada Lovelace, and Ampere GPU architectures

  • NVSwitch Fabric Management - NVIDIA Fabric Manager 535.288.01 enables high-bandwidth, low-latency GPU-to-GPU communication for multi-GPU AI workloads

  • Enhanced vGPU Virtualization - vGPU for Compute 16.13 with improved MIG-backed vGPU configurations and enhanced live migration capabilities

  • Kubernetes Automation - GPU Operator 25.10.1 and Network Operator 25.10.0 for streamlined GPU infrastructure management

  • High-Performance Networking - DOCA-OFED Driver 3.2.0 for enhanced networking and infrastructure acceleration

  • Enterprise Management - Base Command Manager 11.31.0 for cluster provisioning and workload orchestration at scale

What’s Included in NVIDIA AI Enterprise Infra 4.9#

Complete list of infrastructure components with versions and documentation links:

Table 1 Supported Infrastructure Software#

Product

Description

Version

Core Infrastructure Drivers

NVIDIA GPU Data Center Driver

Latest GPU driver with Blackwell, Hopper, Ada Lovelace, and Ampere architecture support

535.288.01

NVIDIA Fabric Manager

Manages NVSwitch fabric to enable high-bandwidth, low-latency GPU-to-GPU communication for multi-GPU AI workloads

535.288.01

NVIDIA DOCA-OFED Driver for Networking

High-performance networking for InfiniBand and Ethernet

3.2.0

Virtualization

NVIDIA vGPU for Compute (Virtual GPU Manager and Guest Drivers)

Enterprise GPU virtualization with advanced monitoring and management capabilities, enabling multiple VMs to run AI workloads with near bare metal performance

16.13

Container Platform

NVIDIA Container Toolkit

GPU-accelerated container runtime with enhanced security

1.18.1

Kubernetes Operators

NVIDIA GPU Operator

Automated GPU software lifecycle management for Kubernetes

25.10.1

NVIDIA Network Operator

Streamlined networking for GPU workloads in Kubernetes

25.10.0

Cluster Management & Orchestration

NVIDIA Base Command Manager (BCM)

Enterprise cluster provisioning, workload orchestration, and lifecycle management

11.31.0

Compatibility and Support#

Refer to the NVIDIA AI Enterprise Infrastructure Support Matrix for:

  • Supported GPU architectures (Blackwell, Hopper, Ada Lovelace, Ampere, Turing, Volta)

  • Operating system compatibility (Ubuntu, RHEL, SLES, Debian, Windows Server)

  • Hypervisor and orchestration platform versions (VMware, KVM, Nutanix, Kubernetes distributions)

  • Cloud provider instance types (AWS, Azure, GCP, OCI, Alibaba, Tencent, Volcano Engine)

  • Networking hardware (ConnectX NICs, BlueField SuperNICs/DPUs)