Application Layer Software#
NVIDIA AI Enterprise application software—including AI frameworks, SDKs, NIMs, and pre-trained models—is distributed through three release branch types, each offering a different balance of innovation speed, API stability, and support duration.
Note
The release cadence of each NVIDIA AI Enterprise release branch type is provided for general guidance only. The actual security update and release cadence can change at NVIDIA’s discretion.
Feature Branch (FB)#
An FB contains the latest versions of NVIDIA NIMs, AI frameworks, workflows, and SDKs. FBs are typically supported for one month. Some products follow different release cycles depending on the software product. Bug fixes and security updates are made available in future FBs.
A new FB is released monthly for most products and can change at NVIDIA’s discretion.
All FB components are listed in the NGC Catalog.
Feature Branch - Latest#
For the NVIDIA AI Enterprise Feature Branch, use the latest NVIDIA AI Enterprise Infra 7 or NVIDIA AI Enterprise Infra 6 collections as your supported configuration.
Because FB releases are superseded monthly, there is no archived product table for Feature Branch. The NGC Catalog collection always reflects the current FB release. For guidance on choosing a release branch, refer to Choosing the Right Release Branch.
Production Branch (PB)#
A PB contains production-ready AI frameworks and SDK branches to provide API stability and a secure environment for building mission-critical AI applications. Each PB is supported for 9 months. During a PB’s support period, it receives monthly bug fixes and security updates for high and critical software vulnerabilities.
A new PB is released every 6 months.
Products available as a PB can be found in the PB Collections section in the NGC Catalog.
Production Branch - October 2025 (PB 25h2)#
The following table provides release information and support timelines for Production Branch - October 2025 (PB 25h2):
PB Collection on NGC |
|
|---|---|
Government Ready PB Collection |
Production Branch Government Ready - October 2025 (PB 25h2) |
First planned release [3] |
October 2025 |
Last planned release |
June 2026 |
Planned End of Life (EOL) |
July 2026 |
Government Ready Versions |
STIG hardened, FIPS enabled versions are available for some products. Refer to the table below. |
Compatible Infrastructure Release |
Use the latest NVIDIA AI Enterprise Infra Release 7 on the NGC Catalog. |
Support Matrix |
The following table lists all products included in Production Branch - October 2025 (PB 25h2) with their versions, Government Ready availability, and product documentation links:
Name |
Version |
Government Ready Versions |
Product Documentation |
|---|---|---|---|
CUDA Deep Learning |
25.08 |
Yes for x86 |
|
Multi-LLM NIM |
1.14 |
Yes for x86 |
|
NVIDIA NIM Llama-3.1-70b-instruct |
1.14 |
Yes for x86 |
|
NVIDIA NIM Llama-3.3-nemotron-super-49b-v1.5 [5] |
1.14 |
Yes for x86 |
|
NVIDIA Holoscan SDK |
3.3 |
No |
|
NVIDIA Retrieval QA Llama 3.2 1B Embedding v2 [5] |
1.11.0 |
Yes for x86 |
|
NVIDIA Retrieval QA Llama 3.2 1B Reranking v2 [5] |
1.9.0 |
Yes for x86 |
|
NVIDIA TensorRT |
25.08-py |
Yes for x86 |
|
NVIDIA Triton Inference Server |
|
Yes for x86 |
|
NVIDIA TAO |
|
Yes for x86 |
|
PyTorch |
25.08-py |
Yes for x86 |
|
RAPIDS [5] |
25.08.02-runtime |
Yes for x86 |
|
RAPIDS Accelerator for Apache Spark [5] |
v25.10 |
No |
New Features - Government Ready
This release introduces a significant new baseline for security, Government Ready, for most x86 container images. Government Ready containers meet software security requirements for use within FedRAMP High or equivalent Sovereign use cases while providing matching functionality with standard NVIDIA software.
Government Ready containers are built on Canonical’s Ubuntu 24.04 STIG-hardened base image and include FIPS-enabled versions of common cryptography libraries. These containers can be deployed the same as normal containers and require a FIPS-enabled Linux kernel on the host machine to utilize FIPS mode.
For detailed information about Government Ready containers, including technical implementation details, STIG and FIPS 140-3 standards, deployment instructions, and verification procedures, refer to Government Ready Containers.
EOL - Production Branch#
Production Branch - May 2025 (PB 25h1) - EOL#
Important
This branch reached end-of-life in January 2026. No further updates will be provided.
The following table provides release information and support timelines for Production Branch - May 2025 (PB 25h1):
PB Collection on NGC |
|
|---|---|
First planned release |
May 2025 |
Last planned release |
December 2025 |
Planned End of Life (EOL) |
January 2026 |
Compatible Infrastructure Release |
Use the latest NVIDIA AI Enterprise Infra Release 6 on the NGC Catalog. |
Support Matrix |
The following table lists products that were included in Production Branch - May 2025 (PB 25h2):
Name |
Version |
Product Documentation |
|---|---|---|
NVIDIA NIM Llama-3.1-8b-instruct [2] |
1.10 |
|
NVIDIA NIM Llama-3.1-70b-instruct [2] |
1.10 |
|
NVIDIA DeepStream SDK [2] |
N/A |
|
NVIDIA Holoscan SDK [2] |
3.3.0 |
|
NVIDIA Morpheus |
25.02-runtime |
|
NVIDIA TensorRT |
25.03-py |
|
NVIDIA Triton Inference Server |
|
|
NVIDIA NIM Retrieval QA E5 Embedding v5 [2] |
1.8.0 |
|
PyTorch |
25.03-py |
|
RAPIDS |
25.02-runtime |
|
RAPIDS Accelerator for Apache Spark |
25.02.1 |
Production Branch - October 2024 (PB 24h2) - EOL#
Important
This branch reached end-of-life in July 2025. No further updates will be provided.
The following table provides archived release information for Production Branch - October 2024 (PB 24h2):
PB Collection on NGC |
|
|---|---|
First planned release |
October 2024 |
Last planned release |
June 2025 |
Planned End of Life (EOL) |
July 2025 |
Compatible Infrastructure Release |
Use the latest NVIDIA AI Enterprise Infra Release 6 or NVIDIA AI Enterprise Infra Release 5 on the NGC Catalog. |
Support Matrix |
The following table lists products that were included in Production Branch - October 2024 (PB 24h2):
Name |
Version |
Product Documentation |
|---|---|---|
Deep Graph Library (DGL) |
24.08-py3 |
|
NVIDIA NIM Llama-3.1-8b-instruct [1] |
1.3 |
|
NVIDIA NIM Llama-3.1-70b-instruct [1] |
1.3 |
|
NVIDIA DeepStream SDK |
7.1-triton-x86 |
|
NVIDIA Holoscan SDK |
24.08 |
|
NVIDIA Morpheus |
24.06-runtime |
|
NVIDIA TensorRT |
24.08-py3 |
|
NVIDIA Triton Inference Server |
|
|
NVIDIA NIM Retrieval QA E5 Embedding v5 [1] |
1.2 |
|
PyTorch |
24.08-py3 |
|
PyTorch Geometric (PyG) |
24.08-py3 |
|
RAPIDS |
24.06-runtime |
|
RAPIDS Accelerator for Apache Spark |
24.06.02 |
|
TensorFlow 2 |
24.08-tf2-py3 |
Production Branch - May 2024 (PB 24h1) - EOL#
Important
This branch reached end-of-life in January 2025. No further updates will be provided.
The following table provides archived release information for Production Branch - May 2024 (PB 24h1):
PB Collection on NGC |
|
|---|---|
First planned release |
May 2024 |
Last planned release |
December 2024 |
Planned End of Life (EOL) |
January 2025 |
Compatible Infrastructure Release |
Use the latest NVIDIA AI Enterprise Infra Release 5 as a supported configuration. |
Support Matrix |
The following table lists products that were included in Production Branch - May 2024 (PB 24h1):
Name |
Version |
Product Documentation |
|---|---|---|
NVIDIA MONAI Toolkit |
24.03-py3 |
|
NVIDIA Morpheus |
24.02-runtime |
|
NVIDIA TensorRT |
24.03-py3 |
|
NVIDIA Triton Inference Server |
|
|
PyTorch |
24.03-py3 |
|
RAPIDS |
24.02-runtime |
|
RAPIDS Accelerator for Apache Spark |
24.02 |
|
TensorFlow 2 |
24.03-tf2-py3 |
Production Branch - October 2023 (PB 23h2) - EOL#
Important
This branch reached end-of-life in July 2024. No further updates will be provided.
The following table provides archived release information for Production Branch - October 2023 (PB 23h2):
PB Collection on NGC |
|
|---|---|
First planned release |
October 2023 |
Last planned release |
June 2024 |
Planned End of Life (EOL) |
July 2024 |
Compatible Infrastructure Release |
Use the latest NVIDIA AI Enterprise Infra Release 5 as a supported configuration. |
The following table lists products that were included in Production Branch - October 2023 (PB 23h2):
Name |
Version |
Product Documentation |
|---|---|---|
NVIDIA Holoscan SDK |
23.10 |
|
NVIDIA TensorRT |
23.08-py3 |
|
NVIDIA Triton Inference Server |
|
|
PyTorch |
23.08-py3 |
|
RAPIDS |
23.06-runtime |
|
TensorFlow 2 |
23.08-tf2-py3 |
Long-Term Support Branch (LTSB)#
An LTSB contains selected AI frameworks and SDKs for highly regulated industries that require longer application lifecycle management. Each LTSB is supported for 3 years and receives quarterly bug fixes and security updates for high- and critical software vulnerabilities.
A new LTSB is released every 30 months.
Products available as an LTSB can be found in the LTSB Collections section of the NGC Catalog.
Long-Term Support Branch 2 (LTSB 2)#
The following table provides release information and support timelines for Long-Term Support Branch 2 (LTSB 2):
LTSB Collection on NGC |
|
|---|---|
First planned release |
November 2024 |
Last planned release |
August 2027 |
Planned End of Life (EOL) |
October 2027 |
Compatible Infrastructure Release |
Use the latest NVIDIA AI Enterprise Infra Release 4 as a supported configuration. |
The following table lists all products included in Long-Term Support Branch 2 (LTSB 2) with their versions and product documentation links:
Name |
Version |
Product Documentation |
|---|---|---|
NVIDIA TensorRT |
|
|
NVIDIA Triton Inference Server |
|
|
PyTorch |
|
|
RAPIDS |
23.06-runtime |
Long-Term Support Branch 1 (LTSB 1) - EOL#
Important
This branch reached end-of-life in June 2024. No further updates will be provided.
The following table provides archived release information for Long-Term Support Branch 1 (LTSB 1):
First planned release |
August 2021 |
|---|---|
Last planned release |
February 2024 |
Planned End of Life (EOL) |
June 2024 |
The following table lists products that were included in Long-Term Support Branch 1 (LTSB 1):
Name |
Version |
Product Documentation |
|---|---|---|
NVIDIA TensorRT |
21.08 |
|
NVIDIA Triton Inference Server |
21.08 |
|
PyTorch |
21.08-py3 |
|
RAPIDS |
21.08 |
|
TensorFlow |
21.08 |
Government Ready Containers#
NVIDIA AI Enterprise offers Government Ready container images that meet software security requirements for use within FedRAMP High or equivalent Sovereign use cases, while maintaining matching functionality with standard NVIDIA software.
What is Government Ready?#
Government Ready designation indicates that container images:
Meet software security requirements for use within FedRAMP High or equivalent Sovereign use cases
Provide matching functionality with NVIDIA software without the Government Ready designation
Are built on Canonical’s Ubuntu 24.04 STIG-hardened base image
Include FIPS-enabled versions of common cryptography libraries, such as OpenSSL
Technical Implementation#
Security Technical Implementation Guides (STIGs)
Security Technical Implementation Guides (STIGs) are configuration standards consisting of cybersecurity requirements for specific products developed by the U.S. Department of Defense. STIGs provide a methodology for standardized secure installation and maintenance of DOD IA and IA-enabled devices and systems, helping organizations harden their systems against security vulnerabilities through detailed technical configuration guidance.
FIPS 140-3
FIPS 140-3 is the U.S. government computer security standard used to approve cryptographic modules, with FIPS 140-3 superseding FIPS 140-2 for new submissions as of April 1, 2022. The goal of the CMVP is to promote the use of validated cryptographic modules and provide Federal agencies with a security metric to use in procuring equipment containing validated cryptographic modules. These standards ensure that cryptographic implementations meet rigorous security requirements for government and regulated environments.
Deployment#
Government Ready containers can be deployed the same as normal containers. To make use of FIPS mode, your host machine must have a FIPS-enabled Linux kernel.
If you run into problems integrating your application with FIPS-enabled libraries, check the documentation for each library whether FIPS mode can be toggled. For example, for OpenSSL you can use OPENSSL_FORCE_FIPS_MODE=0 to disable FIPS mode if needed for testing.
Verifying FIPS Mode on Your Host System#
To verify that your host machine is running in FIPS mode, check the /proc/sys/crypto/fips_enabled file and ensure it is set to 1. If it is set to 0, the FIPS modules will not run in FIPS mode. If the file is missing, the FIPS kernel is not installed. You can verify this with the shell command:
cat /proc/sys/crypto/fips_enabled
Additionally, you can check your kernel version using uname -a to confirm you’re running a FIPS-enabled kernel. Refer to Canonical’s FIPS documentation as an example of setting up a FIPS kernel. Any Linux distribution with a FIPS-enabled kernel should provide similar verification methods through the /proc/sys/crypto/fips_enabled flag.
Additional Resources#
Learn more about NVIDIA’s hardened image in the AI Software for Regulated Environments white paper.
Footnotes