Release Notes

This section provides detailed information for releases and upgrades available for the NVIDIA DGX™ Software Stack for Red Hat Enterprise Linux 9 and Rocky Linux 9.

Current Contents of the Repositories

The following table shows the current version information of the software packages provided in the NVIDIA repositories for the NVIDIA DGX Software Stack.

Current Software Versions (Last Updated on September 17, 2024)

Component

Version

Additional Information

GPU Driver

550.90.07
EL8: RPM installer
EL9: RPM installer

GPU Driver

535.183.06
EL8: RPM installer
EL9: RPM installer

GPU Driver

470.256.02
For EL8 and DGX OS 5 only.
EL8: RPM installer

CUDA Toolkit

12.4 Update 1

R550: 12.4 Update 1 download

CUDA Toolkit

12.2 Update 2

R535: 12.2 Update 2 download

CUDA Toolkit

11.4 Update 4

For EL8 and DGX OS 5 only.

R470: 11.4 Update 4 download

MLNX_OFED

23.10-3.2.2.0 LTS

23.10-3.2.2.0 download

DOCA OFED

2.8.0

2.8.0 download

Inbox OFED

39.0-1

For DGX OS 6 only.

NCCL

2.22.3

cuDNN

9.3.0

DCGM

3.3.8

GPUDirect Storage (GDS)

  • 1.11 for CUDA Toolkit 12.6

  • 1.10 for CUDA Toolkit 12.5

  • 1.9 for CUDA Toolkit 12.4

  • 1.8 for CUDA Toolkit 12.3

  • 1.7 for CUDA Toolkit 12.2

  • 1.0 for CUDA Toolkit 11.4

NVIDIA Container Toolkit

1.16.1-1

NVIDIA Container Toolkit includes the following packages:

  • nvidia-container-toolkit: 1.16.1-1

  • libnvidia-container-tools: 1.16.1-1

  • libnvidia-container1: 1.16.1-1

nvidia-peer-memory

1.3

Note

  • CUDA Toolkit is installed by default only for DGX stations and is optional for DGX servers. Refer to the CUDA Release Notes for driver compatibility information.

  • For CUDA Toolkit minor version compatibility and the minimum required driver version, refer to CUDA Compatibility.

The following table provides information about the matching firmware versions for NVIDIA® OpenFabrics Enterprise Distribution for Linux (MLNX_OFED) version 23.10-3.2.2.0. For more information about this long-term support (LTS) release, refer to

Matching Firmware Versions (Last Updated on September 17, 2024)
DGX-1, DGX-2
ConnectX-4 (CX-4) or
ConnectX-5 (CX-5)
DGX A100
ConnectX-6
DGX A100
ConnectX-7
DGX H100/H200
ConnectX-7

CX-5: 16.35.4030

CX-4: 12.28.2006

20.39.3560

28.39.3560

28.39.3560

For installation instructions, refer to

Note

For information about LTS software versions for related networking components, refer to the Networking Long-Term Support Releases page.

Release Information

This section provides details of each DGX Software for Red Hat Enterprise Linux release. These include mostly new NVIDIA features and accumulated bug fixes and security updates.

  • To check the latest Red Hat Enterprise Linux 9 version, refer to Red Hat Knowledgebase article 3078.

  • To check the MLNX_OFED package OS support, visit Mellanox and click the latest NVIDIA MLNX_OFED software version. Use the side menu to navigate to Release Notes > General Support and view Supported Operating Systems.

Important

Installing or updating to the DGX Software also updates the installed Red Hat Enterprise Linux 9 distribution to the latest version.

If you use NVIDIA MLNX_OFED, then before installing or updating to EL9-23.08, be sure that there is a MLNX_OFED package version available that supports the latest Red Hat Enterprise Linux 9 version.

EL9-24.06 Release

Release Date: July 11, 2024

Release Highlights

  • Added support for Red Hat Enterprise Linux 9.4 and Rocky 9.4.

  • Introduced support for the NVIDIA DOCA™ Host package with the doca-ofed installation profile v2.7.0.

  • Included support for the NVIDIA® OpenFabrics Enterprise Distribution for Linux (MLNX_OFED) v24.04-0.6.6.0

  • Continued support for single-port ConnectX-7 VPI adapter card for DGX A100 System.

  • Updated the DGX Software Stack.

Known Issues

Qualified Software Stack

The following table shows the current version information of the software packages provided in the NVIDIA repositories for the NVIDIA DGX Software Stack.

Component

Latest versions in the repositories

DGX Base OS

EL9-24.06

OS

Red Hat Enterprise Linux 9.4 and Rocky Linux 9.4

Kernel

5.14.0-427.18.1.el9_4.x86_64

GPU Driver

CUDA Toolkit

NCCL

2.21.5

cuDNN

9.1.1

DCGM

3.3.6

GPU Direct Storage

  • 1.9.1 for CUDA 12.4

  • 1.7.2 for CUDA 12.2

NVIDIA System Management (NVSM)

24.03.03

Docker CE

26.1.3

NVIDIA Container Runtime

  • nvidia-container-toolkit: 1.15.0

  • libnvidia-container-tools: 1.15.0

  • libnvidia-container1: 1.15.0

MIG Configuration Tool

0.7.0

GDRCopy

2.4.1

DLFW (Deep Learning Frameworks)

24.05

The following table provides information about the supported OS and matching firmware versions for the NVIDIA® OpenFabrics Enterprise Distribution for Linux (MLNX_OFED) v24.04-0.6.6.0 and the NVIDIA DOCA™ Host package with the doca-ofed installation profile v2.7.0.

OS
DGX-1, DGX-2
ConnectX-4 (CX-4) or
ConnectX-5 (CX-5)
DGX A100
ConnectX-6
DGX A100
ConnectX-7
DGX H100
ConnectX-7
RHEL 9

CX-5: 16.35.3502

CX-4: 12.28.2006

20.41.1000

28.41.1000

28.41.1000

Supported DGX Systems

The EL9-24.06 release supports the following DGX systems:

  • DGX H100

  • DGX A100 640 GB

  • DGX A100 320 GB

  • DGX A800 640 GB

  • DGX-2

  • DGX-1 32 GB

  • DGX Station A100 320 GB

  • DGX Station A100 160 GB

  • DGX Station A800 320 GB

  • DGX Station 32 GB

EL9-23.12 Release

Release Date: December 19, 2023

Release Highlights

  • Added support for Red Hat Enterprise Linux 9.3 and Rocky 9.3.

  • Continued support for Red Hat Enterprise Linux 9.2 and Rocky Linux 9.2.

  • Added support for single-port ConnectX-7 VPI adapter card for DGX A100 System.

  • Added support for NVIDIA® OpenFabrics Enterprise Distribution for Linux (MLNX_OFED) version 23.10-1.1.9.0 - a long-term support (LTS) release.

  • Continued support for DGX H100.

Known Issues

Qualified Software Stack

The following table shows the current version information of the software packages provided in the NVIDIA repositories for the NVIDIA DGX Software Stack.

Component

Latest versions in the repositories

DGX Base OS

EL9-23.12

OS

Red Hat Enterprise Linux 9.3 and Rocky Linux 9.3

Kernel

5.14.0-362.8.1.el9_3

GPU Driver and CUDA Toolkit

CUDA Toolkit 12.2 and GPU Driver 535.129.03 (Default)

NCCL

2.19.3

cuDNN

8.9.6

DCGM

3.3.0-002

GPU Direct Storage

1.7.2 or later

NVIDIA System Management (NVSM)

23.09.02

Docker-CE

24.0.7-1

NVIDIA Container Runtime

  • nvidia-docker2: 2.13.0-1

  • nvidia-container-toolkit (and base): 1.14.3-1

  • libnvidia-container-tools: 1.14.3-1

  • libnvidia-container1: 1.14.3-1

MIG Configuration Tool

0.5.4-1

NGC CLI

3.17.0-1

DLFW (Deep Learning Frameworks)

23.10

The following table provides information about the supported OS and matching firmware versions for NVIDIA® OpenFabrics Enterprise Distribution for Linux (MLNX_OFED) version 23.10-1.1.9.0.

OS
DGX-1, DGX-2
ConnectX-4 (CX-4) or
ConnectX-5 (CX-5)
DGX A100
ConnectX-6
DGX A100
ConnectX-7
DGX H100
ConnectX-7
RHEL 9

CX-5: 16.35.3006

CX-4: 12.28.2006

20.39.1002

28.39.1002

28.39.1002

Supported DGX Systems

NVIDIA has validated and tested EL9-23.12 with the following DGX systems:

  • DGX H100

  • DGX A100 640 GB

  • DGX A100 320 GB

  • DGX A800 640 GB

  • DGX-2

  • DGX-1 32 GB

  • DGX Station A100 320 GB

  • DGX Station A100 160 GB

  • DGX Station 32 GB

Resolved Issues

The following issues have been resolved in the EL9-23.12 release:

Bug ID

Issue

4108242

Running joc tests resulted in an unrecognized arguments: --local-rank error with GPU driver R525.105.17.

4386925

GPUDirect RDMA bandwidth test failed with the Xid (PCI:0000:0f:00): 79, pid='<unknown>', name=<unknown>, GPU has fallen off the bus message.

EL9-23.08 Release

Release Highlights

  • Add support for NVIDIA DGX H100 System. Support is limited to the Red Hat Enterprise Linux 9.1 release.

  • Add support for Red Hat Enterprise Linux 9.2 and Rocky Linux 9.2.

Qualified Software Stack

The following table provides version information for EL9-23.08 and the software it has been qualified:

Component

Latest versions in the repositories

Linux Distribution

Red Hat Enterprise Linux 9.2 and Rocky Linux 9.2

For NVIDIA DGX H100 Systems, only Red Hat Enterprise Linux 9.1 is supported.

GPU Driver

535.86.10

CUDA Toolkit

12.2.0

NCCL

2.18.3

CuDNN

8.9.2.26

DCGM

3.1.8

MLNX OFED

  • ConnectX-7 with DGX H100: 5.9-0.5.6.0.125

  • ConnectX-7 with DGX A100: 5.4-3.7.5.0

  • ConnectX-6 with DGX A100: 5.8-3.0.7.0

  • ConnectX-5 and ConnectX-4: 5.8-3.0.7.0

MLNX FW

  • ConnectX-7 and DGX H100: 28.36.2050

  • ConnectX-7 and DGX A100: 28.34.4000

  • ConnectX-6 and DGX A100: 20.35.4000

  • ConnectX-5: 16.35.3006

  • ConnectX-4: 12.28.2006

GPU Direct Storage

1.7.2

NVIDIA System Management (NVSM)

23.06.04

Docker Engine

23.0.4

NVIDIA Container Runtime

  • nvidia-docker2: 2.13.1-1

  • nvidia-container-toolkit (and base): 1.13.1-1

  • libnvidia-container-tools: 1.13.1-1

  • libnvidia-container1: 1.13.1-1

MIG Configuration Tool

0.5.1

NGC CLI

3.17.0

DLFW (Deep Learning Frameworks)

23.07

The following table provides information about the supported OS and matching firmware versions for Mellanox OFED.

OS

DGX-1, DGX-2

ConnectX-4 or ConnectX-5

DGX A100

ConnectX-6 (CX-6)

DGX A100

ConnectX-7 (CX-7)

DGX H100

ConnectX-7 (CX-7)

RHEL 8

5.8-3.0.7.0

  • CX-5: 16.35.3006

  • CX-4: 12.28.2006

  • RHEL 8.8

5.8-3.0.7.0

  • CX-6: 20.35.3006

  • RHEL 8.8

5.4-3.7.5.0

  • CX-7: 28.34.4000

  • RHEL 8.8

5.9-0.5.6.0.127

  • CX-7: 28.36.2050

  • RHEL 8.7

RHEL 9

5.8-3.0.7.0

  • CX-5: 16.35.3006

  • CX-4: 12.28.2006

  • RHEL 9.2

5.8-3.0.7.0

  • CX-6: 20.35.3006

  • RHEL 9.2

5.4-3.7.5.0

  • CX-7: 28.34.4000

  • RHEL 9.2

5.9-0.5.6.0.127

  • CX-7: 28.36.2050

  • RHEL 9.1

Hardware Compatibility

NVIDIA has validated and tested EL9-23.08 with the following DGX systems:

  • NVIDIA DGX H100

  • NVIDIA DGX A100

  • NVIDIA DGX Station A100

  • NVIDIA DGX Station

  • NVIDIA DGX-2

  • NVIDIA DGX-1

EL9-23.01 Release

Initial release of the DGX Software Stack for Red Hat Enterprise Linux 9.

Qualified Software Stack

The following table provides version information for EL9-23.01 and the software it has been qualified:

Component

Versions in this release

Linux Distribution

Red Hat Enterprise Linux 9.1 and Rocky Linux 9.1

GPU Driver

525.105.17

CUDA Toolkit

12.0

NCCL

2.18.1

CuDNN

8.9.1.23

DCGM

3.1.8

NVIDIA MLNX_OFED

5.8-2.0.3.0

NVIDIA ConnectX Firmware

  • CX-4: 12.28.2006

  • CX-5: 16.35.2000

  • CX-6: 20.35.2000

NVIDIA System Management (NVSM)

22.12.04

Docker Engine

23.0.4

NVIDIA Container Runtime

  • nvidia-docker2: 2.13.0-1

  • nvidia-container-toolkit (and base): 1.13.1-1

  • libnvidia-container-tools: 1.13.1-1

  • libnvidia-container1: 1.13.1-1

MIG Configuration Tool

0.5.1

NGC CLI

3.17.0

DLFW (Deep Learning Frameworks)

23.03

Hardware Compatibility

NVIDIA has validated and tested EL9-23.01 with the following DGX systems:

  • DGX-1

  • DGX-2

  • DGX Station

  • DGX A100

  • DGX Station A100