我要评分
获取效率
正确性
完整性
易理解

Environment Requirements

Before the performance test, ensure that the environment meets the hardware and software requirements. The hardware configuration includes CPUs. The software configuration includes the OS and applications.

Hardware requirements

Table 1 lists the hardware requirement.

Table 1 Hardware requirement

Hardware

Description

Number of Cores

CPU

New Kunpeng 920 processor model

160 cores + 160 cores (two processors with hyper-threading enabled)

OS requirements

Table 2 lists the OS requirement.

Table 2 Verified OS

System

Version

Kernel

Description

How to Obtain

openEuler

22.03 LTS SP3

5.10.0

When installing an OS, choose Minimal Install and select Development Tools to minimize manual operations.

Link

Software Requirements

The software requirements for model training and inference are different, as shown in Table 3 and Table 4.

Table 3 Software requirements for model training

Item

Version

Description

Build Mode

TensorFlow

1.15.5

Training models from ModelZoo

Build it using source code.

Python

3.7

Dependency for TensorFlow 1.15 installation

Obtain it from Conda.

CMake

3.22.0 or later

Cross-platform build system generator with cross-compiling support.

Install it using a Yum repository.

GCC/G++

10.3.1/12.3.1

GNU Compiler Collection (GCC), a compiler system developed by the GNU project, supporting multiple programming languages.

Install it using a Yum repository.

Bazel

0.26.0

Build tool for TensorFlow 1.15.5 installation

Build it using source code.

Git

No version requirement

Code management tool

Install it using a Yum repository.

Table 4 describes the software requirements in the model inference phase.

Table 4 Software requirements for model inference

Item

Version

Description

Build Mode

TensorFlow Serving

2.15

Server for model deployment and inference request processing in the benchmarking system

Build it using source code.

Python

3.9 or later

Python execution environment

Obtain it from Conda.

CMake

3.22.0 or later

Cross-platform build system generator with cross-compiling support.

Install it using a Yum repository.

GCC/G++

12.3.1

GNU Compiler Collection (GCC), a compiler system developed by the GNU project, supporting multiple programming languages.

Install it using a Yum repository.

Bazel

6.5.0

Fast and scalable build tool, required for compiling TensorFlow Serving 2.15.

Build it using source code.

TritonServer (Triton server)

24.05

Performance test tool

Pull an image from Docker.

Git

No version requirement

Code management tool

Install it using a Yum repository.