the utilization of cpu and gpu is low, how to increase them?

Question

yan gao 2021-9-24

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1460134-the-utilization-of-cpu-and-gpu-is-low-how-to-increase-them

while i am using matlab to train a alexnet from the scratch on window, the utilization ratio of cpu and gpu of my computer is low, and I wonder how to increase them.

here is my code:

trainImagesetPath = 'E:\deep_learning_dataset\tiny-imagenet-200\train';
valImagesetPath = 'E:\deep_learning_dataset\tiny-imagenet-200\val';
testImagesetPath = 'E:\deep_learning_dataset\tiny-imagenet-200\test';
miniBatchSize = 960;
imdsTrain = imageDatastore(trainImagesetPath, 'IncludeSubfolders', true, ...
    'LabelSource', 'foldernames', 'FileExtensions',{'.jpg','.JPG', '.JPEG'}, 'ReadSize', miniBatchSize);
imdsValidation = imageDatastore(valImagesetPath, 'IncludeSubfolders', true, ...
    'LabelSource', 'foldernames', 'FileExtensions',{'.jpg','.JPG', '.JPEG'}, 'ReadSize', miniBatchSize);
imdsTest = imageDatastore(testImagesetPath, 'IncludeSubfolders', true);
layers = [imageInputLayer([224 224 3])
    convolution2dLayer(11, 96, 'Stride', [4, 4], 'Padding', [0 0 0 0])
    reluLayer
    batchNormalizationLayer
    maxPooling2dLayer(3, 'Stride', [2, 2], 'Padding', [0 0 0 0])
    
    groupedConvolution2dLayer(5, 128, 2, 'Stride', [1, 1], 'Padding', [2 2 2 2])
    reluLayer
    batchNormalizationLayer
    maxPooling2dLayer(3, 'Stride', [2, 2], 'Padding', [0 0 0 0])
    
    convolution2dLayer(3, 384, 'Stride', [1, 1], 'Padding', [1 1 1 1])
    reluLayer
    
    groupedConvolution2dLayer(3, 192, 2, 'Stride', [1, 1], 'Padding', [1 1 1 1])
    reluLayer
    
    groupedConvolution2dLayer(3, 128, 2, 'Stride', [1, 1], 'Padding', [1 1 1 1])
    reluLayer
    maxPooling2dLayer(3, 'Stride', [2, 2], 'Padding', [0 0 0 0])
    
    fullyConnectedLayer(4096)
    reluLayer
    dropoutLayer(0.5)
    
    fullyConnectedLayer(4096)
    reluLayer
    dropoutLayer(0.5)
    
    fullyConnectedLayer(200)
    softmaxLayer
    classificationLayer
    ];
% analyzeNetwork(layers);
inputSize = [224, 224, 3];
augimdsTrain = augmentedImageDatastore(inputSize, imdsTrain, 'ColorPreprocessing', 'gray2rgb', 'DispatchInBackground', true);
augimdsValidation = augmentedImageDatastore(inputSize, imdsValidation, 'ColorPreprocessing', 'gray2rgb', 'DispatchInBackground', true);
augimdsTest = augmentedImageDatastore(inputSize, imdsTest, 'ColorPreprocessing', 'gray2rgb');
options = trainingOptions('adam', ...
    'MiniBatchSize', miniBatchSize, ...
    'MaxEpochs',120, ...
    'InitialLearnRate',1e-4, ...
    'LearnRateSchedule', 'piecewise', ...
    'LearnRateDropFactor', 0.25, ...
    'LearnRateDropPeriod', 5, ...
    'DispatchInBackground', true, ...
    'Shuffle','every-epoch', ...
    'ValidationData', augimdsValidation, ...
    'ValidationFrequency', 20, ...
    'Verbose',true, ...
    'Plots','training-progress', ...
    'ExecutionEnvironment', 'auto');
tic
alexNetModel = trainNetwork(augimdsTrain,layers,options);
fprintf('training process time cost: ');
toc
[YPred,scores] = classify(alexNetModel,augimdsTest);
YTest = imdsTest.Labels;
accuracy = mean(YPred == YTest);
fprintf('test acc: %f\n', accuracy);
figure
confusionchart(YTest, YPred)