Deploying NVIDIA Converged Accelerator
It is recommended to upgrade your BlueField product to the latest software and firmware versions available to benefit from new features and latest bug fixes.
This section assumes that you have installed the BlueField OS BFB on your NVIDIA® Converged Accelerator using any of the following guides:
NVIDIA® CUDA® (GPU driver) must be installed in order to use the GPU. For information on how to install CUDA on your Converged Accelerator, refer to NVIDIA CUDA Installation Guide for Linux.
After installing the BFB, you may now select the mode you want your NVIDIA Converged Accelerator to operate in.
Standard (default) – the NVIDIA® BlueField® DPU and the GPU operate separately (GPU is owned by the host)
BlueField-X – the GPU is exposed to the DPU and is no longer visible on the host (GPU is owned by the DPU)
It is important to know your device name (e.g., mt41686_pciconf0).
MST tool is necessary for this purpose which is installed by default on the DPU.
Run:
mst status -v
Example output:
MST modules:
------------
MST PCI module is not loaded
MST PCI configuration module loaded
PCI devices:
------------
DEVICE_TYPE MST PCI RDMA NET NUMA
BlueField2(rev:1) /dev/mst/mt41686_pciconf0.1 3b:00.1 mlx5_1 net-ens1f1 0
BlueField2(rev:1) /dev/mst/mt41686_pciconf0 3b:00.0 mlx5_0 net-ens1f0 0
BlueField-X Mode
Run the following command from the host:
mlxconfig -d /dev/mst/<device-name> s PCI_DOWNSTREAM_PORT_OWNER[4]=0xF
P erform a BlueField system-level reset for the mlxconfig settings to take effect.
Standard Mode
To return the DPU from BlueField-X mode to Standard mode:
Run the following command from the host:
mlxconfig -d /dev/mst/<device-name> s PCI_DOWNSTREAM_PORT_OWNER[4]=0x0
P erform a BlueField system-level reset for the mlxconfig settings to take effect.
Use the following command from the host or BlueField:
$ sudo mlxconfig -d /dev/mst/<device-name> q PCI_DOWNSTREAM_PORT_OWNER[4]
Example of Standard mode output:
Device #1: ---------- [...] Configurations: Next Boot PCI_DOWNSTREAM_PORT_OWNER[4] DEVICE_DEFAULT(0)
Example of BlueField-X mode output:
Device #1: ---------- [...] Configurations: Next Boot PCI_DOWNSTREAM_PORT_OWNER[4] EMBEDDED_CPU(15)
The following are example outputs for when the DPU is configured to BlueField-X mode.
The GPU is no longer visible from the host:
root@host:~# lspci | grep -i nv
None
The GPU is now visible from the DPU:
ubuntu@dpu:~$ lspci | grep -i nv
06:00.0 3D controller: NVIDIA Corporation GA20B8 (rev a1)
Get GPU Firmware
smbpbi: (See SMBPBI spec)
root@dpu:~# i2cset -y 3 0x4f 0x5c 0x05 0x08 0x00 0x80 s
root@dpu:~# i2cget -y 3 0x4f 0x5c ip 5
5: 0x04 0x05 0x08 0x00 0x5f
root@dpu:~# i2cget -y 3 0x4f 0x5d ip 5
5: 0x04 0x39 0x32 0x2e 0x30
root@dpu:~#
root@dpu:~#
root@dpu:~# i2cset -y 3 0x4f 0x5c 0x05 0x08 0x01 0x80 s
root@dpu:~# i2cget -y 3 0x4f 0x5c ip 5
5: 0x04 0x05 0x08 0x01 0x5f
root@dpu:~# i2cget -y 3 0x4f 0x5d ip 5
5: 0x04 0x30 0x2e 0x36 0x42
root@dpu:~# i2cset -y 3 0x4f 0x5c 0x05 0x08 0x02 0x80 s
root@dpu:~# i2cget -y 3 0x4f 0x5c ip 5
5: 0x04 0x05 0x08 0x02 0x5f
root@dpu:~# i2cget -y 3 0x4f 0x5d ip 5
5: 0x04 0x2e 0x30 0x30 0x2e
root@dpu:~# i2cset -y 3 0x4f 0x5c 0x05 0x08 0x03 0x80 s
root@dpu:~# i2cget -y 3 0x4f 0x5c ip 5
5: 0x04 0x05 0x08 0x03 0x5f
root@dpu:~# i2cget -y 3 0x4f 0x5d ip 5
5: 0x04 0x30 0x31 0x00 0x00
root@dpu:~#
39 32 2e 30 30 2e 36 42 2e 30 30 2e 30 31 00 00 → 92.00.6B.00.01
Updating GPU Firmware
root@dpu:~# scp root@10.23.201.227:/<path-to-fw-bin>/1004_0230_891__92006B0001-dbg-ota.bin /tmp/gpu_images/
root@10.23.201.227's password:
1004_0230_891__92006B0001-dbg-ota.bin 100% 384KB 384.4KB/s 00:01
root@dpu:~# cat /tmp/gpu_images/progress.txt
TaskState="Running"
TaskStatus="OK"
TaskProgress="50"
root@dpu:~# cat /tmp/gpu_images/progress.txt
TaskState="Running"
TaskStatus="OK"
TaskProgress="50"
root@dpu:~# cat /tmp/gpu_images/progress.txt
TaskState=Frimware update succeeded.
TaskStatus=OK
TaskProgress=100