v1.3.0-rc.1

NVIDIA Fleet Intelligence Agent v1.3.0-rc.1

This release includes a host agent for GPU telemetry collection and attestation.

Available Packages

• DEB packages: amd64 and arm64 for Debian/Ubuntu systems
• RPM packages: x86_64 and aarch64 for RHEL/CentOS/Fedora systems
• Binary archives: Standalone binaries for both architectures

Installation

Download the appropriate package for your system from the assets below. For installation instructions, see the project README.

Changelog

New Features

• 37f954e: feat: add MALLOC_ARENA_MAX to limit the dcgm mem (#182) (@ambermingxin)
• 17425be: feat: add agent enroll time in inventory loop (#188) (@ambermingxin)
• f1e5740: feat: add backend client and inventory attestation call (#176) (@jingxiang-z)
• a601987: feat: add loop and retry for dcgm detection (#180) (@ambermingxin)
• b5bebe6: feat: add support for ada lovelace and ampere GPU architecture (#197) (@ambermingxin)
• 8c48f2d: feat: create nvml component (#194) (@ambermingxin)

Bug Fixes

• b4c188f: fix: change the helm default node selector to be gpu.deploy.dcgm (#181) (@jingxiang-z)
• 685e639: fix: use canonical reflect pointer kind (#186) (@jingxiang-z)

Refactors

• 18af139: refactor: simplify inventory and attestation loops (#184) (@jingxiang-z)

Other Changes

• e2cc962: Clarify the documentation on depdencies. (@ebalduf)
• a8c71cf: Fix RPM release docs too. (@ebalduf)
• 0a93e84: light wording change to Hopper support. (@ebalduf)
• 24e2b5c: security: parameterize metrics component filter to prevent SQL injection (#178) (@rsampaio)

Verification

All release artifacts are checksummed. Verify downloads using:

sha256sum -c checksums.txt

Support

For issues and questions, please visit: https://github.com/NVIDIA/fleet-intelligence-agent/issues