Rate This Document
Findability
Accuracy
Completeness
Readability

Environment Requirements

This document uses openEuler 24.03 LTS SP3 + TensorFlow 2.21.0 as an example to describe how to install and verify the GPU. In addition, this section describes other environments supported by TensorFlow installation on Kunpeng servers.

Supported Versions

Table 1 Supported versions

OS

Python

TensorFlow

openEuler 24.03 LTS SP3

Python 3.11.6

TensorFlow 2.21.0

openEuler 22.03 LTS SP4

Python 3.9.9

Not applicable to TensorFlow 2.21.0, which requires a later Python version.

The official pip GPU installation path of TensorFlow 2.21.0 uses the NVIDIA pip package of the CUDA 12 series. When building the GPU version from source code, use the CUDA Toolkit, cuDNN, and NVIDIA driver that match the build configuration. In the verification environment, the NVIDIA driver version is 580.159.04, and CUDA 13.0 is displayed on the driver side. The driver is compatible with the CUDA 12 user-mode library.

Dependent Component Requirements

Table 2 Dependent components and versions

Component

Version

python3

3.11.6-20.oe2403sp3

pip

26.1.1

TensorFlow

2.21.0

CUDA user-mode dependency

NVIDIA pip package of CUDA 12 series

NVIDIA Driver

580.159.04

GPU

NVIDIA A100-PCIE-40GB