First Published: August 24, 2025

Last Updated: May 7, 2026

Revision History

Revision

Date

Description

D0

May 07, 2026

Created release notes for 2.0.2.260101 for the following servers:

  • Cisco UCS C845A M8 Rack Server

C0

March 11, 2026

Created release notes for 2.0.2.260062 for the following servers:

  • Cisco UCS C845A M8 Rack Server

B0

November 03, 2025

Created release notes for 2.0.1.250127 for the following servers:

  • Cisco UCS C845A M8 Rack Server

A0

August 24, 2025

Created release notes for 2.0.1.250096 for the following servers:

  • Cisco UCS C845A M8 Rack Server

Overview of the Cisco UCS C845A M8 Rack Server

The Cisco UCS C845A M8 Rack Server is a scalable and flexible AI system based on the NVIDIA MGX reference design for accelerated computing. This server supports high-performance AI workloads, including enterprise inferencing, agentic AI, and generative AI fine-tuning. The Cisco UCS C845A M8 Rack Server is a customizable AI system designed for accelerated computing environments.

  • Supports two to eight NVIDIA or AMD PCIe GPUs.

  • Integrates with Cisco Intersight for unified infrastructure management.

  • Aligns with NVIDIA AI Data Platform reference designs to support agentic AI and RAG applications.

  • Optimized for enterprise AI inferencing and agentic AI.

  • Incorporates distributed security through Cisco AI Defense, Hybrid Mesh Firewall, Isovalent, and Hypershield.

The platform offers high-performance compute resources optimized for data-intensive AI applications. Key technical specifications include:

  • Memory: Up to 8TB of DDR5-6400 system memory.

  • Storage: Gen5 E1.S NVMe storage supporting up to 15.3TB per drive.

  • Processing: 5th Gen AMD EPYC Turin CPUs and NVIDIA BlueField-3 DPUs.

Supported GPU options include:

  • NVIDIA: RTX PRO 4500, RTX PRO 6000D, RTX PRO 6000 Blackwell, H200 NVL, and L40S.

  • AMD: Instinct MI210.

Provides a scalable platform for AI initiatives by integrating compute, storage, and networking into a unified management interface. This server supports enterprise-grade AI workloads through high-performance hardware and integrated security features.

The platform utilizes advanced hardware and software components to maintain security and processing efficiency:

  • NVIDIA BlueField-3 DPUs: Offload security policy enforcement from the CPU and GPU to preserve cycles for AI processing.

  • Observability: Hardens the stack through integration with Splunk Enterprise Security and Splunk Observability Cloud.

  • Deployment Acceleration: Supports NVIDIA NIM microservices, NeMo, and Blueprints to shorten the path from pilot to production.


Note

The system leverages NVIDIA BlueField-3 DPUs with crypto acceleration to offload security policy enforcement, preserving CPU and GPU cycles for AI processing.

Overview of the Server Software

The Cisco Baseboard Management Controller 2.0 (Cisco BMC 2.0) is the management service for the Cisco UCS C845A M8 servers. Cisco BMC 2.0 runs within the server.


Note

The Cisco BMC 2.0 management service is used only when the server is operating in Standalone Mode. If your server is integrated into a UCS system, you must manage it using Cisco Intersight. You may see Cisco Intersight Help Center to get more information on how to manage the server.

You can use a web-based GUI, an SSH-based CLI, or a REST API to access, configure, administer, and monitor the server. Each interface offers different capabilities, and the tasks supported by each interface are described in their respective configuration guides. You cannot do the following:

  • Use Cisco BMC 2.0 GUI to invoke Cisco BMC 2.0 CLI.

  • View a command that has been invoked through Cisco BMC 2.0 CLI in Cisco BMC 2.0 GUI.

  • Generate Cisco BMC 2.0 CLI output from Cisco BMC 2.0 GUI.

What's New

Release 2.0.2.260062

  • Remote Management: Added local upload for Slot 1 and remote upload for Slots 2 and 3, with Slot 3 designated as high-priority.

  • FlexMMC: Added remote download support, enabling the upload of remote files into the FlexMMC.

  • MSTOR-RAID: Added functionality to create and delete RAID 1 virtual drives.

  • User Management: Added support for creating users with various combinations of SNMP and BMC access.

  • Shared NIC: Added NCSI support to enable shared_NIC configuration via the eth1 interface.

  • Date and Time: Added support for configuring DHCP and static NTP servers independently.

  • SNMP v3: Added support for comprehensive SNMP v3 trap configurations, including authentication and destination settings.

  • Fan Policies: Added support for toggling fan duty cycles between Manual and Auto modes.

Release 2.0.1.250127

  • Release 2.0.1.250127 enables PSU firmware upgrade using host upgrade utility.

Supported Platforms

The following servers are supported in 2.0.1.250096 and later releases:

  • Cisco UCS C845A M8 Rack Server

Browser Requirements

To access the Cisco BMC 2.0 GUI, Cisco recommends using the most recent version of one of the following supported browsers for Windows, Linux RHEL, and MacOS:

  • Microsoft Edge

  • Mozilla Firefox

  • Google Chrome

  • Apple Safari

For detailed information about supported Operating System, see the interactive UCS Hardware and Software Compatibility matrix.

Default Ports

Following is a list of server ports and their default port numbers:

Port Name Port Number

HTTPS

443

SSH

22

IPMI

623

Firmware Files


Note

Always upgrade all the components BIOS, BMC, and FPGA. Do not upgrade individual components (only BIOS or only BMC), since this could lead to unexpected behavior. If the BIOS and the BMC versions are from different container releases, it could result in unexpected behavior.

For release specific ISO versions, see Cisco Baseboard Management Controller 2.0 Firmware Files, Release 2.0

Firmware Files in Release 2.0.2.260101

The 2.0.2.260101 software release includes the following software files:

Table 1. 2.0.2.260101 Cisco Firmware Bundle

CCO Software Type

File Name

Unified Computing System (UCS) Server Firmware

ucsc-845a-m8-huu-2.0.2.260101.iso

Cisco UCS Server Diagnostics Utility

ucs-diag-2.0.1.250007.iso

Firmware Files in Release 2.0.2.260062

The 2.0.2.260062 software release includes the following software files:

Table 2. 2.0.2.260062 Cisco Firmware Bundle

CCO Software Type

File Name

Unified Computing System (UCS) Server Firmware

ucsc-845a-m8-huu-2.0.2.260062.iso

Cisco UCS Server Diagnostics Utility

ucs-diag-2.0.1.250007.iso

Firmware Files in Release 2.0.1.250127

The 2.0.1.250127 software release includes the following software files:

Table 3. 2.0.1.250127 Cisco Firmware Bundle

CCO Software Type

File Name

Unified Computing System (UCS) Server Firmware

ucsc-845a-m8-huu-2.0.1.250127.iso

Cisco UCS Server Diagnostics Utility

ucs-diag-2.0.1.250007.iso

Firmware Files in Release 2.0.1.250096

The 2.0.1.250096 software release includes the following software files:

Table 4. 2.0.1.250096 Cisco Firmware Bundle

CCO Software Type

File Name

Unified Computing System (UCS) Server Firmware

ucsc-845a-m8-huu-2.0.1.250096.iso

Cisco UCS Server Diagnostics Utility

ucs-diag-2.0.1.250007.iso

Resolved Caveats

Resolved Caveats in Release 2.0.2.260101

The following defects are resolved in Release 2.0.2.260101:

Defect ID

Symptom

First Affected Release

Resolved in Release

CSCwt66661

The HTTPS certificate replace operation fails when the existing certificate is expired and not visible in the BMC after a reboot. The expired certificate is removed from the BMC certificate list upon reboot, causing the replace operation to fail because it depends on the presence of the existing certificate.

This issue is resolved.

2.0.2.260062

 2.0.2.260101

CSCwu15165

The firmware upgrade, using Redfish interface, intermittently fails on fully populated Cisco UCS C845A M8 rack server, especially when all components are claimed by Cisco Intersight. This issue occurs during host-based component firmware updates. The issue is observed by triggering the upgrade using Redfish API automation scripts and monitoring task status, with failures occurring intermittently across multiple upgrade attempts.

This issue is resolved.

2.0.2.260062

 2.0.2.260101

Resolved Caveats in Release 2.0.2.260062

The following defects are resolved in Release 2.0.2.260062:

Defect ID

Symptom

First Affected Release

Resolved in Release

CSCwp22429

The FirmwarePackageVersion property is empty for the FHHL network adapter on the Cisco UCS C845A M8 Rack Server. This causes the firmware version and the MAC address of the PCI device in the FHHL slot to be missing when queried using the Redfish API. This issue occurs during firmware version retrieval for the PCI device.

This issue is resolved.

2.0.1.250096

2.0.2.260062

CSCws42716

System logs a sensor detection failure alert for PSU1 during normal operations. PSU1 does not appear in the firmware inventory endpoint, causing its health status to show as unavailable.

This issue is resolved.

2.0.1.250127

2.0.2.260062

CSCwq50085

When an Operator user launches a KVM session, the Session Management page incorrectly shows the privilege level as Administrator instead of Operator.

This issue is resolved.

2.0.1.250096

2.0.2.260062

CSCwt34598

Power supply unit (PSU) firmware upgrade may fail. The upgrade process reports an Update.1.1.1.ApplyFailed error, and the affected PSU is not discovered in the Cisco BMC UI or Redfish Firmware Inventory after the upgrade attempt.

This issue is resolved.

2.0.1.250096

2.0.2.260062

Resolved Caveats in Release 2.0.1.250127

The following defects are resolved in Release 2.0.1.250127:

Defect ID

Symptom

First Affected Release

Resolved in Release

CSCwp08112

The Cisco UCS C845A M8 Rack Server experiences a delay in displaying the host video output signal after system power cycle. Although the host is on, the KVM display remains blank for an extended time, causing confusion.

This issue is resolved.

2.0.1.250096

2.0.1.250127

CSCwp37991

The FHHL_10 DPU link status and speed do not display in the Redfish API output. The link state and current link speed fields return null despite the port being enabled and signal detected. This issue occurs with the PCI device connected to the FHHL 10 slot.

This issue is resolved.

2.0.1.250096

2.0.1.250127

CSCwp39909

After a BMC reboot completes and the system reaches BIOS Post Complete, BMC Boot Complete, and BMC Inventory Ready states, the GPU firmware version reported in PCIeDevices is trimmed by a few characters. This results in a temporary mismatch between the server and Intersight firmware version displays.

This issue is resolved.

2.0.1.250096

2.0.1.250127

CSCwp91773

During ESXi installation on the Cisco UCS C845A M8 Rack Server, when accessed remotely from Mac operating systems, the F11 keystroke cannot be sent through KVM, blocking installation progress.

This issue is resolved.

2.0.1.250096

2.0.1.250127

CSCwq91165

The Cisco UCS C845A M8 Rack Server experiences a firmware update issue where updates initiated with the "on-next boot" option remain pending and fail after the host OS reboots. The firmware update does not start when the host reboots from the OS environment, and the Redfish task reports an internal error. Additionally, events are missed following the host OS reboot.

This issue is resolved.

2.0.1.250096

2.0.1.250127

CSCwq09352

In Cisco UCS C845A M8 Rack Server, after a BMC update or reboot, the Remote Management page takes several minutes to load. During this time, the KVM launch option remains disabled. This delay affects immediate access to remote management features but resolves once the page fully loads.

This issue is resolved.

2.0.1.250096

2.0.1.250127

CSCwr75062

In Cisco UCS C845A M8 Rack Server with fully populated GPUs, spurious GPU temperature critical SEL assert and de-assert logs occur continuously during host reboot.

This issue is resolved.

2.0.1.250096

2.0.1.250127

Open Caveats

Open Caveats in Release 2.0.2.260101

The following caveats are open in release 2.0.2.260101:

Defect ID

Symptom

Workaround

First Affected Release

CSCwr18300

OS installation succeeds only when the SCU ISO is mapped to Slot 3 and the OS ISO is mapped to Slot 2. Mapping ISO to either Slot 0, Slot 1, or using the KVM Map blocks the installation process. The system fails to boot to the SCU ISO if the SCU image is mapped on virtual media slots 0, 1, or 3.

Map the SCU ISO to vMedia slot 3 and OS image to Slot 2.

2.0.2.260101

CSCwt85211

After downgrading firmware from release 2.0.2.260101 to 2.0.1.250127 on B3140H and B3220 BlueField-3 cards using HSU, the firmware version does not update to the downgraded version. The BMC UI and Redfish interface continue to show the previous version. The firmware is not properly activated or reflected across interfaces after reboot. This issue affects firmware downgrade visibility and activation on these cards.

Perform a two step downgrade:

  1. First downgrade from release 2.0.2.260101 to 2.0.2.260062

  2. Then downgrade from release 2.0.2.260062 to release 2.0.1.250127.

2.0.2.260101

Open Caveats in Release 2.0.2.260062

The following caveats are open in release 2.0.2.260062:

Defect ID

Symptom

Workaround

First Affected Release

CSCwt50090

During the HSU upgrade, the KVM screen goes blank and does not display video redirection after power cycles or reboots. The issue occurs during or after the HSU update and persists through OS boot.

Disable and then again enable the KVM from GUI.

2.0.2.260062

Open Caveats in Release 2.0.1.250127

The following caveats are open in release 2.0.1.250127:

Defect ID

Symptom

Workaround

First Affected Release

CSCws42716

System logs a sensor detection failure alert for PSU1 during normal operations. PSU1 does not appear in the firmware inventory endpoint, causing its health status to show as unavailable.

This issue does not have impact on system functionality or performance.

2.0.1.250127

Open Caveats in Release 2.0.1.250096

The following caveats are open in release 2.0.1.250096:

Defect ID

Symptom

Workaround

First Affected Release

CSCwo53738

The Cisco UCS C845A M8 Rack Server shows an issue where multiple characters print for a single key press when typing at the Ubuntu 22.04 or 24.04 login prompt using the KVM console. The issue affects keystroke input on the host console during login.

Disable DF-CState.

For information on how to disable DF-CState, see Workaround for Defect - CSCwo53738

2.0.1.250096

CSCwp22429

The FirmwarePackageVersion property is empty for the FHHL network adapter on the Cisco UCS C845A M8 Rack Server. This causes the firmware version and the MAC address of the PCI device in the FHHL slot to be missing when queried using the Redfish API. This issue occurs during firmware version retrieval for the PCI device.

Reboot the host.

2.0.1.250096

CSCwp37991

The FHHL_10 DPU link status and speed do not display in the Redfish API output. The link state and current link speed fields return null despite the port being enabled and signal detected. This issue occurs with the PCI device connected to the FHHL 10 slot.

There is no known workaround.

2.0.1.250096

CSCwp91773

During ESXi installation on the Cisco UCS C845A M8 Rack Server, when accessed remotely from Mac operating systems, the F11 keystroke cannot be sent through KVM, blocking installation progress.

Enable function keys on the remote operating system as they are disabled by default. This allows the F11 keystroke to be sent through KVM during ESXi installation. Adjust the remote system settings to activate function keys for proper remote key mapping.

2.0.1.250096

CSCwq50085

When an Operator user launches a KVM session, the Session Management page incorrectly shows the privilege level as Administrator instead of Operator.

This issue has no functional impact.

2.0.1.250096

CSCwq55206

The ipmitool command on Ubuntu 24.04.1 with version 1.8.19 may encounter Device or resource busy errors when attempting to access sensor data.

You can perform one of the following:

  • Utilize the Remote IPMI Interface with the lanplus option for remote management.

  • Alternatively, use ipmitool version 1.8.18 on Ubuntu 24.04 or Ubuntu 22.04.

  • Or, operate on any other supported operating system that is compatible with ipmitool.

2.0.1.250096

CSCwp08112

The Cisco UCS C845A M8 Rack Server experiences a delay in displaying the host video output signal after system power cycle. Although the host is on, the KVM display remains blank for an extended time, causing confusion.

Relaunch KVM or wait for few minutes and let the KVM display the correct status.

2.0.1.250096

CSCwq91165

The Cisco UCS C845A M8 Rack Server experiences a firmware update issue where updates initiated with the "on-next boot" option remain pending and fail after the host OS reboots. The firmware update does not start when the host reboots from the OS environment, and the Redfish task reports an internal error. Additionally, events are missed following the host OS reboot.

You can reboot the host using the options available in the Web GUI, through KVM, or by using the ComputerSystem.Reset Redfish API.

2.0.1.250096

CSCwr73079

The BMC NC-SI driver forces the x710 NIC Port 0 to operate at 100 Mbps on the host side, while Port 1 runs at 1G/10G speeds. This behavior causes the link on Port 0 to drop when the switch attempts higher-speed auto-negotiation. The driver overrides network settings, limiting Port 0 speed to 100 Mbps instead of allowing 1G or 10G operation.

Update BMC firmware version to release 2.0.1.250127.

2.0.1.250096

Workaround for Defect - CSCwo53738

From BIOS Setup

  1. Power on the system and enter the BIOS setup by pressing the ESC or DEL key.

  2. Click

    Advanced > AMD CBS > NBIO Common Options > SMU Common Options > DF CStates

  3. Set the DF CStates option to Disabled.

  4. Save the changes and exit the BIOS setup.

Using Redfish Interface

  1. Boot the host into the operating system.

  2. Run the following curl command to update the DF CStates setting via Redfish API:

    curl -ksu Username:Password https://10.10.10.10/redfish/v1/Systems/system/Bios/
Settings -H 'content-type:application/json' -H 'If-Match:' -X PATCH -d '
{"Attributes":{"CbsCmnGnbSmuDfCstates":"Disabled"}}' | jq

  3. Reboot the host.


    Note

    The server reboots twice during this process.

  4. After the automatic reboot, enter the BIOS setup menu and verify that the DF CStates value is set to Disabled.

  5. Exit the BIOS setup without making changes and allow the system to boot into the OS or UEFI shell.

  6. Confirm that the Redfish data reflects the updated DF CStates setting after boot.

Known Behaviors and Limitations

Known Behaviors and Limitations in Release 2.0.2.260101

The following caveat is a known limitation in release 2.0.2.260101:

Defect ID

Symptom

Workaround

First Affected Release

CSCwp27054

During a Graceful Shutdown initiated from the BMC and KVM, when the system boots to the HUU UI, the Power Off options do not function.

Use Power Down and Power On to manage system power.

2.0.1.250096

Known Behaviors and Limitations in Release 2.0.2.260062

The following caveats are known limitations in release 2.0.2.260062:

Defect ID

Symptom

Workaround

First Affected Release

CSCwr18364

After installing RHEL 10, the desktop environment may freeze or only show the RHEL 10 logo when opening any application, icon, or folder. This prevents normal user interaction with the OS. The issue occurs post-installation and reboot.

If the monitor uses the built-in ASPEED graphics controller, select Install Red Hat Enterprise Linux in basic graphics mode during installation to prevent conflicts with ATI cards.

When using a virtual display through the management port, ATI cards must be detached from the default seat (seat0) and reassigned to seat1 using the specified loginctl attach commands. First, switch to the multi-user target with systemctl isolate multi-user.target, perform the device reassignment, then return to the graphical target with systemctl isolate graphical.target. After these steps, verify that you can log in and access the GNOME session normally.

2.0.2.260062

Known Behaviors and Limitations in Release 2.0.1.250096

The following caveats are known limitations in release 2.0.1.250096:

Defect ID

Symptom

Workaround

First Affected Release

CSCwq15755

The Cisco UCS C845A M8 Rack Server does not honor the configured one-time boot device after reboot. The server continues to boot using the default boot order, ignoring the selected boot option. This issue occurs when BIOS token and one-time boot device are configured together. The host reboots twice but fails to boot from the chosen device, requiring manual boot device selection to proceed.

Configure BIOS token and one time boot one at a time.

2.0.1.250096