Networking ConnectX-8 Firmware Release Notes

Changes and New Features

Security Hardening Enhancements: This release contains important reliability improvements and security hardening enhancements. NVIDIA recommends upgrading your devices' firmware to this release to improve the devices’ firmware security and reliability.

To generate PLDM packages for firmware updates, users must install and use the MFT version that corresponds with the respective firmware release.

Feature/Change

Description

40.51.1002

Live Firmware Patch (LFWP) Support

Added Live Firmware Patch (LFWP) support for ConnectX-8. When USER_PROGRAMMABLE_CC is set to false, PROGCC code updates applied via LFWP will not take effect until a full chip reset is performed.

Live Firmware Patch (LFWP) on LTS Branches

Starting with LTS 2026, supported devices can apply LTS patches via live update within the same branch. This uses a level 0 ("live-patch") reset via mlxfwreset, so the driver, PCI link, and network link all remain up. A level 3 reset or level 4 warm reboot is not required.

Enforce CC Hints (AppAware) Capability Check in Firmware

CC hint capability checking was moved from the DPA (pcc_send_set_nv_conf) to the firmware. The logic itself is unchanged. As part of the move, is_cc_hint_general_object_cap now includes the limitations that were previously part of pcc_send_set_nv_conf.

CC Hint Status in QUERY_CC_HINT

QUERY_CC_HINT now returns a 64-byte status block written by the Spectrum-X congestion-control application. The workload can use it to see whether a hint was accepted and applied. The size is reported in cc_hint_status_size_max. A value of 0 means this status is not available.

Transceiver Firmware Update over PLDM

Added support for out-of-band firmware updates for supported CMIS transceivers through a BMC using PLDM. The NIC functions as a Firmware Device Proxy, enabling transceiver discovery, inventory reporting, firmware download, and activation.

Key capabilities include:

  • Updating an individual transceiver or sequentially updating all matching, update-capable modules

  • Reporting update progress, cancellation, retries, timeout handling, and failures

  • Supporting self-contained firmware activation and reporting the running firmware version after activation

This capability requires PLDM downstream update support to be enabled and transceivers that support CMIS CDB firmware download.

DOCA Telemetry PHY Statistics API Support (VDM)

DOCA can now retrieve PHY telemetry through the Port Extended Module Info Register (PEMI), using the same mechanism as mlxlink. This enhancement also extends PEMI VDM telemetry support beyond SNR readings to include all CMIS-defined VDM telemetry, such as PAM4, pre-FEC BER, and FERC.

SPDM Attestation Notification on Live Update (ISSU)

Added a user notification when a live update (ISSU) occurs, since it may break SPDM attestation. A new NV configuration option to disable live update is available for users who want to prevent this.

Multiplane RTT probe failover

Improved congestion control behavior on multiplane configurations when one plane fails. Previously, when a plane failed, a QP's RTT probes were dropped when traffic moved to healthy planes, causing congestion control to reduce the QP's rate to the minimum. RTT probes now follow the traffic to healthy planes, preserving the QP's rate. This applies to in-order QPs (UD, UC, and in-order RC). Out-of-order QPs are not affected as they run congestion control separately on each plane.

Flex parser support in multi-host setups

Added support for parse graph nodes in multi-host setups.

Multiplane vport speed reporting

Added explicit speed reporting for two aggregated speed notions per vport: cap_tx_speed (the maximum speed the vport could reach under any circumstances, reflecting the internal connection topology) and effective_tx_speed (the maximum speed currently available to the vport, accounting only for link-up ports).

PCIe event log: critical events

Added support for a new PCIE_EVENTS segment in resource dump to collect PCIe events. The index1 (LOG_ID) parameter selects the log source: 0 for a non-volatile log containing only critical events, and 1 for a volatile operational log containing all events.

PCIe event log: volatile operational log

Added a volatile operational log for all PCIe events, including critical and non-critical events. The log is accessible through MGNLE using the new log_id field, or through resource dump using the PCIE_EVENTS segment with index1 (LOG_ID) = 1. The log is 2 MB by default on PCIe switch-only devices; on other devices it can be enabled using the OPER_LOG_BUFFER_SIZE mlxconfig parameter.

DOCA Flow CT support in VNF mode

Added support for local eswitch per VF in VNF mode, extending DOCA Flow Connection Tracking (CT) functionality to virtual network function deployments.

SVPC hostless DPA encapsulation

Added a new internal DPA application that provides east-west VPC isolation using VXLAN encapsulation in a steering domain that is not accessible to the eswitch manager. The feature is disabled by default and can be enabled using an NV config parameter.

Bug Fixes

See Bug Fixes in this Firmware Version section.

Last updated: