AI 集群 技术知识库

Appearance

ConnectX-7 网卡固件烧录 ​

1. 检查当前固件版本 ​

bash
ibstat | grep -E "CA|Firmware"
cat /sys/class/infiniband/mlx5_*/fw_ver
bash
mst start
for dev in /dev/mst/mt4129_pciconf*; do
    echo "=== $dev ==="
    sudo flint -d $dev q
done

结果：

tex
=== /dev/mst/mt4129_pciconf0 ===
Image type:            FS4
FW Version:            28.42.1000
FW Release Date:       8.8.2024
Product Version:       28.42.1000
Rom Info:              type=UEFI version=14.35.15 cpu=AMD64,AARCH64
                       type=PXE version=3.7.500 cpu=AMD64
Description:           UID                GuidsNumber
Base GUID:             a088c2030059c440        8
Base MAC:              a088c259c440            8
Image VSD:             N/A
Device VSD:            N/A
PSID:                  MT_0000000838
  • PSID：MT_0000000838 在选固件的时候选这个版本的，首先选择网卡型号，再选择PSID相同的版本固件

2. 准备固件文件 ​

bash
# 官方链接
https://network.nvidia.com/support/firmware/connectx7/

# 解压官方固件包
unzip fw-ConnectX7-rel-28_43_3608-MCX75310AAS-NEA_Ax-UEFI-14.37.50-FlexBoot-3.7.500.signed.bin.zip

# 确认 .bin 文件存在
ls -la *.bin

3. 启动 MST 服务 ​

bash
sudo mst start
sudo mst status

预期输出包含 /dev/mst/mt4129_pciconf0 ~ /dev/mst/mt4129_pciconf8。

4. 烧录固件 ​

4.1 单张网卡测试烧录 ​

bash
sudo flint -d /dev/mst/mt4129_pciconf0 -i fw-ConnectX7-rel-28_43_3608-MCX75310AAS-NEA_Ax-UEFI-14.37.50-FlexBoot-3.7.500.signed.bin burn

4.2 批量烧录所有网卡 ​

bash
for dev in /dev/mst/mt4129_pciconf{0..8}; do
    echo "=== Burning $dev ==="
    sudo flint -d $dev -i fw-ConnectX7-rel-28_43_3608-MCX75310AAS-NEA_Ax-UEFI-14.37.50-FlexBoot-3.7.500.signed.bin burn
done

5. 烧录后验证 ​

bash
for dev in /dev/mst/mt4129_pciconf*; do
    echo -n "$dev: "
    sudo flint -d $dev q | grep "FW Version" | awk '{print $3}'
done

所有输出必须为 28.43.3608。

6. 重启服务器 ​

bash
# 推荐断电重启
sudo shutdown -h now
# 手动开机

# 或软重启
sudo reboot