UOS V2500 昇腾 CANN 异构平台部署手册

部署环境

项目 内容
服务器 鲲鹏 920 服务器(aarch64)
NPU 卡 昇腾 310P
操作系统 UOS V2500
CANN 版本 8.3.RC2

如果使用其他版本 NPU 卡或 CANN 版本,请从昇腾官网下载对应的驱动和套件安装包。

准备安装运行用户

本文档示例步骤将以 root 安装驱动、固件和 CANN,以 HwHiAiUser 为驱动固件运行用户和运行用户属组。执行如下命令创建 HwHiAiUser 用户和用户属组:

groupadd HwHiAiUser
useradd -g HwHiAiUser -d /home/HwHiAiUser -m HwHiAiUser -s /bin/bash

安装 NPU 驱动

步骤 1:检查源是否可用

yum update
yum install python3-pip
pip3 install --upgrade pip -i https://pypi.tuna.tsinghua.edu.cn/simple
ln -s /usr/bin/python3 /usr/bin/python
ln -s /usr/bin/pip3 /usr/bin/pip

如果命令执行报错或等待时间过长,则检查网络是否连接或者修改 /etc/yum.repos.d/UnionTechOS-aarch64.repo 文件配置。更新完成后重启系统。

步骤 2:修改执行权限并校验软件包

chmod +x Ascend-hdk-310p-npu-driver_25.3.rc1_linux-aarch64.run
chmod +x Ascend-hdk-310p-npu-firmware_7.8.0.2.212.run
./Ascend-hdk-310p-npu-driver_24.1.rc2_linux-aarch64.run --check
./Ascend-hdk-310p-npu-firmware_7.3.0.1.231.run --check

出现如下回显信息,表示软件包检验成功:

Verifying archive integrity...  100%   SHA256 checksums are OK. All good.

步骤 3:安装驱动

驱动默认安装路径为 /usr/local/Ascend

./Ascend-hdk-310p-npu-driver_25.3.rc1_linux-aarch64.run --full --install-for-all

若出现如下信息,则表示驱动安装成功:

Verifying archive integrity...  100%   SHA256 checksums are OK. All good.
Uncompressing ASCEND DRIVER RUN PACKAGE  100%
[Driver] [2025-02-13 15:57:39] [INFO]Start time: 2025-02-13 15:57:39
[Driver] [2025-02-13 15:57:39] [INFO]LogFile: /var/log/ascend_seclog/ascend_install.log
[Driver] [2025-02-13 15:57:39] [INFO]OperationLogFile: /var/log/ascend_seclog/operation.log
[Driver] [2025-02-13 15:57:39] [INFO]base version is none.
[Driver] [2025-02-13 15:57:39] [WARNING]Do not power off or restart the system during the installation/upgrade
[Driver] [2025-02-13 15:57:39] [INFO]set username and usergroup, HwHiAiUser:HwHiAiUser
[Driver] [2025-02-13 15:57:51] [INFO]driver install type: Rebuild
[Driver] [2025-02-13 15:57:51] [INFO]Rebuilding by kernel path(/lib/modules/4.19.90-2403.3.0.0270.95.uel20.aarch64/build), detail in /var/log/ascend_seclog/ascend_rebuild.log
[Driver] [2025-02-13 15:58:10] [INFO]upgradePercentage:10%
[Driver] [2025-02-13 15:58:14] [INFO]upgradePercentage:30%
[Driver] [2025-02-13 15:58:15] [INFO]upgradePercentage:40%
[Driver] [2025-02-13 15:58:21] [INFO]upgradePercentage:90%
[Driver] [2025-02-13 15:58:23] [INFO]Waiting for device startup...
[Driver] [2025-02-13 15:58:23] [INFO]Device startup success
[Driver] [2025-02-13 15:58:33] [INFO]upgradePercentage:100%
[Driver] [2025-02-13 15:58:35] [INFO]Driver package installed successfully! Reboot needed for installation/upgrade to take effect!

步骤 4:安装固件

./Ascend-hdk-310p-npu-firmware_7.8.0.2.212.run --full

若出现如下信息,则表示固件安装成功:

Verifying archive integrity...  100%   SHA256 checksums are OK. All good.
Uncompressing ASCEND-HDK-310P-NPU FIRMWARE RUN PACKAGE  100%
[Firmware] [2025-02-13 16:02:57] [INFO]Start time: 2025-02-13 16:02:57
[Firmware] [2025-02-13 16:02:57] [INFO]LogFile: /var/log/ascend_seclog/ascend_install.log
[Firmware] [2025-02-13 16:02:57] [INFO]OperationLogFile: /var/log/ascend_seclog/operation.log
[Firmware] [2025-02-13 16:02:58] [WARNING]Do not power off or restart the system during the installation/upgrade
[Firmware] [2025-02-13 16:03:01] [INFO]upgradePercentage: 79%
[Firmware] [2025-02-13 16:03:07] [INFO]upgradePercentage: 100%
[Firmware] [2025-02-13 16:03:07] [INFO]The firmware of [1] chips are successfully upgraded.
[Firmware] [2025-02-13 16:03:12] [INFO]Firmware package installed successfully! Reboot now or after driver installation for the installation/upgrade to take effect.

步骤 5:重启系统

reboot

步骤 6:验证驱动加载

npu-smi info

若返回驱动相关信息说明加载成功;否则说明加载失败。

安装 CANN 软件包

安装 Toolkit 开发套件

安装 Toolkit 软件前请确保安装目录可用空间大于 7GB,如不满足请清理空间或更换安装目录。

步骤 1:修改执行权限并校验软件包:

chmod +x Ascend-cann-toolkit_8.3.RC2_linux-aarch64.run
./Ascend-cann-toolkit_8.3.RC2_linux-aarch64.run --check

出现如下信息,表示软件包检验成功:

Verifying archive integrity...  100%   SHA256 checksums are OK. All good.

步骤 2:安装开发套件:

yum install python3-pip
./Ascend-cann-toolkit_8.3.RC2_linux-aarch64.run --install

执行安装命令过程中会显示"华为企业业务最终用户许可协议"的条款内容,输入 Y 同意协议后继续安装。安装完成后显示如下信息,则说明软件安装成功:

Driver:   Installed in /usr/local/Ascend/driver.
Toolkit:  Ascend-cann-toolkit_8.3.RC2_linux-aarch64 install success, installed in /usr/local/Ascend.

步骤 3:配置环境变量:

# 临时设置环境变量
source /usr/local/Ascend/ascend-toolkit/set_env.sh

# 永久设置环境变量
echo 'source /usr/local/Ascend/ascend-toolkit/set_env.sh' >> ~/.bashrc

# 环境变量立即生效
source ~/.bashrc

步骤 4:安装后检查:

cat /usr/local/Ascend/ascend-toolkit/latest/aarch64-linux/ascend_toolkit_install.info

文件内信息如下:

package_name=Ascend-cann-toolkit
version=8.3.RC2
innerversion=V100R001C23SPC002B210
compatible_version=[V100R001C15],[V100R001C18],[V100R001C19],[V100R001C20],[V100R001C21],[V100R001C23]
arch=aarch64
os=linux
path=/usr/local/Ascend/ascend-toolkit/8.3.RC2/aarch64-linux

安装 Kernels 算子包

步骤 1:修改执行权限并校验软件包:

chmod +x Ascend-cann-kernels-310p_8.3.RC2_linux-aarch64.run
./Ascend-cann-kernels-310p_8.3.RC2_linux-aarch64.run --check

出现如下信息,表示软件包检验成功:

Verifying archive integrity...  100%   SHA256 checksums are OK. All good.

步骤 2:安装除静态库之外的其他文件:

./Ascend-cann-kernels-310p_8.3.RC2_linux-aarch64.run --devel

执行安装命令过程中会显示"华为企业业务最终用户许可协议"的条款内容,输入 Y 同意协议后继续安装。安装完成后显示如下信息,则说明软件安装成功:

Ascend-cann-kernels-310p_8.3.RC2_linux-aarch64.run install success. The installation path is /usr/local/Ascend/ascend-toolkit.

步骤 3:安装完成后检查版本信息:

cat /usr/local/Ascend/ascend-toolkit/latest/opp_kernel/version.info

文件内容信息如下:

Version=8.3.0.2.220
version_dir=8.3.RC2
timestamp=20251117_000024591
ops_version=8.3.0.2.220
adk_version=8.3.0.2.220
required_package_amct_acl_version=">=7.6, <=8.3" required_package_aoe_version=">=7.6, <=8.3" required_package_compiler_version=">=7.6, <=8.3" required_package_fwkplugin_version=">=7.6, <=8.3" required_package_hccl_version=">=7.6, <=8.3" required_package_nca_version=">=7.6, <=8.3" required_package_ncs_version=">=7.6, <=8.3" required_package_opp_version=">=7.6, <=8.3" required_package_runtime_version=">=7.6, <=8.3" required_package_toolkit_version=">=7.6, <=8.3"

安装 NNAL 神经网络加速库

NNAL 神经网络加速库提供了 ATB(Ascend Transformer Boost)加速库和 SiP(AscendSiPBoost)信号处理加速库。安装加速库前需已安装同一版本的 Toolkit 并配置环境变量。

步骤 1:增加软件包可执行权限:

chmod +x Ascend-cann-nnal_8.3.RC2_linux-aarch64.run

步骤 2:安装软件包:

./Ascend-cann-nnal_8.3.RC2_linux-aarch64.run --install

步骤 3:配置环境变量:

当前以非 root 用户安装后的默认路径为例,请根据加速库对应的 set_env.sh 的实际路径进行替换。

# 临时设置 ATB 加速库环境变量
source /usr/local/Ascend/nnal/atb/set_env.sh

# 永久设置环境变量
echo 'source /usr/local/Ascend/nnal/atb/set_env.sh' >> ~/.bashrc

# 环境变量立即生效
source ~/.bashrc

出现以下内容表示安装成功:

[NNAL] [20251225-16:01:19] [INFO] Ascend-cann-atb_8.3.RC2_linux-aarch64.run --install --install-path=/usr/local/Ascend/nnal --install-for-all --quiet --nox11 start
[NNAL] [20251225-16:01:31] [INFO] Ascend-cann-atb_8.3.RC2_linux-aarch64.run --install --install-path=/usr/local/Ascend/nnal --install-for-all --quiet --nox11 install success
[NNAL] [20251225-16:01:31] [INFO] Ascend-cann-SIP_8.3.RC2_linux-aarch64.run --install --install-path=/usr/local/Ascend/nnal --install-for-all --quiet --nox11 start
[NNAL] [20251225-16:01:31] [INFO] Ascend-cann-SIP_8.3.RC2_linux-aarch64.run --install --install-path=/usr/local/Ascend/nnal --install-for-all --quiet --nox11 install success
[NNAL] [20251225-16:01:31] [INFO] Ascend-cann-nnal_8.3.RC2_linux-aarch64.run install success

注意:如果同时设置 atb 和 asdsip 的环境变量,会产生意外后果。请根据使用场景导入对应的环境变量:

  • atb 模块(大模型场景):source /usr/local/Ascend/nnal/atb/set_env.sh
  • asdsip 模块(嵌入式场景):source /usr/local/Ascend/nnal/asdsip/set_env.sh

常见问题解决

错误 1:Drv_dkms_env_check failed

问题描述:

Drv_dkms_env_check failed, details in :/var/log/ascend_seclog/ascend_install.log

解决方法:

yum install kernel-devel kernel-headers -y

错误 2:Dkms install failed

问题描述:

[ERROR]Dkms install failed, details in : var/log/ascend_seclog/ascend_install.log.
[ERROR]Driver_ko_install failed, details in : /var/log/ascend_seclog/ascend_install.log.

解决方法:

cd /var/lib/dkms
rm -rf davinci_ascend

错误 3:缺少工具 lspci 和 ifconfig

问题描述:

[ERROR]The list of missing tools: lspci,ifconfig

解决方法:

yum install -y net-tools pciutils
Logo

作为“人工智能6S店”的官方数字引擎,为AI开发者与企业提供一个覆盖软硬件全栈、一站式门户。

更多推荐