如何在美國GPU服務(wù)器Ubuntu 20.04 Linux系統(tǒng)上測試CPU和GPU性能

創(chuàng)新互聯(lián)公司專業(yè)為企業(yè)提供博山網(wǎng)站建設(shè)、博山做網(wǎng)站、博山網(wǎng)站設(shè)計、博山網(wǎng)站制作等企業(yè)網(wǎng)站建設(shè)、網(wǎng)頁設(shè)計與制作、博山企業(yè)網(wǎng)站模板建站服務(wù)，10年博山做網(wǎng)站經(jīng)驗，不只是建網(wǎng)站，更提供有價值的思路和整體網(wǎng)絡(luò)服務(wù)。

1、安裝必要的軟件包

打開終端并登錄到Ubuntu 20.04服務(wù)器。

運(yùn)行以下命令以更新軟件包列表：

“`

sudo apt update

“`

安裝以下軟件包，用于測試CPU和GPU性能：

“`

sudo apt install buildessential cmake nvcc

“`

2、下載并編譯CUDA工具包

訪問NVIDIA官方網(wǎng)站（https://developer.nvidia.com/cudadownloads）下載適用于您的GPU的CUDA工具包。

解壓下載的文件到適當(dāng)?shù)哪夸浿小?/p>

進(jìn)入解壓后的目錄，并運(yùn)行以下命令以編譯CUDA工具包：

“`

sudo make j$(nproc)

sudo sudo make install

“`

3、安裝其他必要的軟件包

運(yùn)行以下命令以安裝其他必要的軟件包：

“`

sudo apt install g++ libopenblasdev liblapackdev libx11dev libxmudev libglib2.0dev libgtk3dev libboostalldev libeigen3dev libcudnn7=7.6.5.321+cuda10.1 libcudnn7dev=7.6.5.321+cuda10.1 libcufft7=7.6.5.321+cuda10.1 libcufft7dev=7.6.5.321+cuda10.1 libcurand7=7.6.5.321+cuda10.1 libcurand7dev=7.6.5.321+cuda10.1 libnccl2=2.8.31+cuda10.1 libnccl2dev=2.8.31+cuda10.1 libnpp=7.6.5.321+cuda10.1 libnppdev=7.6.5.321+cuda10.1 libnppc=7.6.5.321+cuda10.1 libnppcdev=7.6.5.321+cuda10.1 libnumba=0.49.0 dfsg5+cuda10.1 libnumbadev=0.49.0 dfsg5+cuda10.1 libomp5 libomp5dev libopencvcore3.4 libopencvhighgui3.4 libopencvimgcodecs3.4 libopencvimgproc3.4 libopencvvideoio3.4 libopencvfeatures2d3.4 libopencvcalib3d3.4 libopencvml3.4 libopencvobjdetect3.4 libopencvcontrib3.4 libopencvflann3.4 libopencvstitching3.4 libopencvsuperres3.4 libopencvvideo4.4 libopencvvideoio4.4 libopencvtracking4.4 libopencvtext3.4 libopencvdnn3 python3numpy python3scipy python3matplotlib python3pandas python3sklearn python3tensorflow python3pytorch python3jupyter python3jupyterlab python3sympy python3cython python3h5py python3mpi4py python3pyyaml python3networkx python3nlopt python3pygame python3pyqt5 python3pyqtgraph python3pyqtgraph_notebook python3pyqtgraph_examples python3pyqtgraph_widgets python3pyqtgraph_viewer python3pyqtgraph_mpl_interaction python3pyqtgraph_multiplottool python3pyqtgraph_console python3pyqtgraph_customize python3pyqtgraph_parallel pybind11 tbb cmake qtbase5dev qttools5dev qttools5devtools qtmultimedia5dev qtdeclarative5dev qtquickcontrols25 qmlscene qmltest qtwebengine5 qtxcb qtscript5

4、編寫一個簡單的程序來測試CPU和GPU性能

在終端中創(chuàng)建一個新的Python文件（例如test_performance.py），并輸入以下代碼：

“`python

import numpy as np

import timeit

# CPU性能測試函數(shù)

def test_cpu():

return np.sum(np.random.rand(1000, 1000))

# GPU性能測試函數(shù)

def test_gpu():

import torch

a = torch.randn(1000, 1000).cuda()

b = torch.randn(1000, 1000).cuda()

return (a + b).sum().item()

# 測試CPU性能并輸出結(jié)果

start_time = timeit.default_timer()

result_cpu = test_cpu()

end_time = timeit.default_timer()

print("CPU性能測試結(jié)果：", result_cpu, "時間：", end_time start_time)

# 測試GPU性能并輸出結(jié)果

start_time = timeit.default_timer()

result_gpu = test_gpu()

end_time = timeit.default_timer()

print("GPU性能測試結(jié)果：", result_gpu, "時間：", end_time start_time)

“`

保存文件后，在終端中運(yùn)行以下命令執(zhí)行測試程序：

“`

python test_performance.py

“`

程序?qū)⑤敵鯟PU和GPU性能測試的結(jié)果以及所需的時間。

網(wǎng)頁題目：如何在美國GPU服務(wù)器ubuntu20.04Linux系統(tǒng)上測試CPUGPU性能
轉(zhuǎn)載源于：http://uogjgqi.cn/article/cdcggsc.html