Download Download NVIDIA 900-9D3B4-00SV-EA DPU Firmware 32.39.2048 for OS Independent Free

Firmware is added at the time of manufacture to run user programs on the device and can be thought of as the software that allows the hardware to run. Embedded firmware is used to control the functionality of various hardware devices and systems, much like a computer’s operating system (OS) controls the functionality of software applications. Firmware can be written to read-only memory (ROM), erasable programmable read-only memory (EPROM), or flash memory.

NVIDIA BlueField-3 DPU delivers innovative acceleration, security and efficiency to every host. BlueField-3 data center infrastructure combines the power of NVIDIA ConnectX-6 Dx with programmable Arm cores and hardware offload for software-defined storage, networking, security and management workloads.

NVIDIA BlueField-3 also delivers superior performance, security and lower total cost of ownership for cloud computing platforms, enabling organizations to efficiently build and operate virtualized, containerized and bare metal infrastructure at scale.

Variety:

– Changed default FEC configuration for “Protocol Aware” and “Active DME Modules” (ETH cable). See the table below for a list of cable identifiers.
– Added support for two pass-through channels on dual-port adapter cards.
– Added caching mechanism to improve expansion ROM performance and avoid any slow startup when loading expansion ROM drivers.
– Added support for image sizes above 4GB when performing live migration by splitting the image into chunks.
– Extended role-based authentication to cover all encryption algorithms. Now it’s TLS. IP security. MAC seconds. GCM, mem2mem and NISP work when nv_crypto_conf.crypto_policy = CRYPTO_POLICY_FIPS_LEVEL_2, which means that all crypto engines also work in wrapped mode, not just plaintext mode.
– Added the ability to configure the DSCP (priority) of ACK packets using the ROCE_ACCL access register.
– Added support for forcing large MTU for loopback QP to improve performance (using aes_xts_tweak_inc_64 parameter). This feature is enabled by the mlxconfig LARGE_MTU_TWEAK_64 parameter.
– When there is DDR poisoning (uncorrectable ECC error), the firmware reports health syndrome ICM_FETCH_PCI_DATA_POISONED_ERR (0x14) and triggers FLR on the function that caused this error. Due to this bug, the DDR data is mostly corrupted, so the firmware blocks other operations on this feature.
– Added support for live firmware patching.
– Added new support for preserving mkey index ranges. When enabled, a series of mkey indexes are maintained for mkey by name.
– Added support for managing queues in virtio device objects.
– When using enhanced NIC mode, enable the GGA module for all operating modes except RXP.

Bug fixes:

– Fixed an issue on special systems with separate power supplies that caused the host to hang and RDMA to fail in the virtio-net-controller when performing the following steps: hotplug 31 vnet device, host power off, unplug 31 vnet device , hot-swappable 31 vnet device or host power on
– A BlueField Arm core acting as a PCIe root port for a PCIe endpoint device (such as an NVMe SSD) connected to the BlueField PCIe interface may receive MSI-X (used by the device to indicate the event) before the PCIe CQE is written, causing the driver to interrupt The handler attempts to retrieve data that is in an inconsistent state.
– Fixed cr_space in port configuration to prevent timestamp errors for cqes.
– Fixed an issue that prevented the PCC flow context database from being cleared when starting a new DOCA PCC application, which was used to avoid “legacy state” active applications from affecting the behavior of the new application.
– Update HW ETS (QETCR RL) default to per host port instead of per physical port to prevent bandwidth degradation.
– Fixed an issue that caused TX to hang and create a “TX timeout” error in dmesg after force-unplugging the device during a server warm restart.
– Fixed an issue that caused the server to fail to start (after a reboot) when there were 31 hot-swappable devices on a BlueField-3 DPU custom server with an independent power supply.
– Fixed an issue that caused the reset to fail when uninstalling the network driver on an external host, and the sync1 reset was still reported as “supported” even though it was not.
– Increased the number of flex buffer lock attempts to prevent rare Tx issues during Gen1.
– Fixed an issue in LAG mode that caused RoCE traffic to have lower throughput when Congestion Control (CC) mode was enabled than when CC mode was disabled.
– Added support for hairpin drop counter in QUERY_VNIC_ENV command.
– Fixed an issue that caused corruption of migration data when running parallel save_vhca_state/load_vhca_state commands on the same PF.
– Updated OOB counter behavior.
– The DPC mechanism is currently not supported.

Known issues:

– Supported DDR5 link speed is 5200 MT/s.
– Occasionally, the device may hang when performing a hotplug from an unknown direction.
– Some pre-OS environments may fail when a hotplug operation is detected during the boot phase.
– The I2C clock fall time is below the 12ns minimum defined in the I2C bus specification.
– When connecting to a high-speed traffic generator at 400G speeds, link time may take up to 3 minutes.
– External flash access (such as reading flash using MFT tools) will fail if there is a pending image on the flash.
– On BlueField-3 devices, from DOCA 2.2.0 to 32.37.1306 (or lower), host crashes when performing partial Arm resets (e.g. Arm reboot, BFB push, mlxfwreset).
– When uninstalling a network driver on an external host, sync1 reset may still be reported as “supported” even though this is not the case. Therefore, initiating the reset process may cause the reset to fail after a few minutes.
– When connecting to a Spirent switch at 400G speed, the connection time may be up to 3 minutes.
– PCIe PML1 is disabled.
– Unexpected system behavior may occur if drivers are loaded during reset.
– When performing a PCIe link auxiliary bus reset, disable/enable, or mlxfwreset on an AMD-based Genoa system, the device takes longer than expected to connect due to a misconfigured PCIe receiver terminal.
– If the QPN field is configured in mkey (as it only allows a given QP to use this Mkey), firmware rollback of the signed retransmission flow fails because the firmware rollback flow relies on the internal QP using mkey.
– Socket-Direct is not currently supported.

About the network adapter driver:

Once connected, the operating system typically installs a universal driver that helps the computer recognize the newly connected device.

However, if you want to take advantage of all the features offered by your network adapter, you must apply the appropriate software. This task also allows the computer to correctly identify all device characteristics such as manufacturer, chipset, technology, etc.

Updating your adapter’s driver and utility versions may improve overall performance and stability, increase transfer speeds, fix different compatibility issues and some network-related bugs, and bring various other changes.

To install this version, simply grab the package, unzip it if necessary, run the installer, and follow the on-screen instructions. Once completed, don’t forget to reboot your system and reconnect your network adapter to ensure all changes take effect correctly.

Without further ado, if you plan to apply this versi on, click on the download button and install the package. Also, please visit our website as often as possible so you never miss a new release.