为何PyTorch 2.0比TensorFlow 2.0在CUDA环境下运行慢3倍？

我正在测试TensorFlow 2.0与PyTorch 2.0的运行速度（刚接触PyTorch），在相同模型架构、批次大小、优化器且均启用CUDA的前提下，发现PyTorch的训练耗时约为TensorFlow的3倍（TF耗时1分钟，PT耗时3分钟），且验证集精度更低（TF为83%，PT为78%）。

同时观察到：TensorFlow占用约60%的CUDA使用率及全部专用GPU显存，而PyTorch的CUDA利用率在0%-30%间波动，显存占用极低。已排除CUDA_LAUNCH_BLOCKING参数导致的问题（代码中未设置该参数）。

TensorFlow 代码 #

(train_images, train_labels), (test_images, test_labels) = tf.keras.datasets.fashion_mnist.load_data()
model = tf.keras.Sequential([
    tf.keras.layers.Flatten(input_shape=(28,28)),
    tf.keras.layers.Dense(512, activation="relu"),
    tf.keras.layers.Dense(512, activation="relu"),
    tf.keras.layers.Dense(10, activation="softmax")
])
model.compile(optimizer=tf.keras.optimizers.SGD(1e-3), loss="sparse_categorical_crossentropy", metrics=["accuracy"])
model.fit(train_images, train_labels, epochs=20, batch_size=64, validation_data=(test_images, test_labels))

PyTorch 代码 #

import torch
from torch import nn
from torch.utils.data import DataLoader
from torchvision import datasets
from torchvision.transforms import ToTensor

device = torch.device("cuda")

training_data = datasets.FashionMNIST(
    root="data",
    train=True,
    download=True,
    transform=ToTensor()
)

test_data = datasets.FashionMNIST(
    root="data",
    train=False,
    download=True,
    transform=ToTensor()
)

train_dataloader = DataLoader(training_data, batch_size=64)
test_dataloader = DataLoader(test_data, batch_size=64)

class NeuralNetwork(nn.Module):
    def __init__(self):
        super().__init__()
        self.flatten = nn.Flatten()
        self.linear_relu_stack = nn.Sequential(
            nn.Linear(28*28, 512),
            nn.ReLU(),
            nn.Linear(512, 512),
            nn.ReLU(),
            nn.Linear(512, 10),
        )

    def forward(self, x):
        x = self.flatten(x)
        logits = self.linear_relu_stack(x)
        return logits

model = NeuralNetwork()
model.to(device)

learning_rate = 1e-3
batch_size = 64
epochs = 5
loss_fn = nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(model.parameters(), lr=learning_rate)

def train_loop(dataloader, model, loss_fn, optimizer):
    size = len(dataloader.dataset)

    model.train()
    for batch, (X, y) in enumerate(dataloader):
        X, y = X.to(device), y.to(device)
        pred = model(X)
        loss = loss_fn(pred, y)

        loss.backward()
        optimizer.step()
        optimizer.zero_grad()

        if batch % 100 == 0:
            loss, current = loss.item(), (batch + 1) * len(X)
            print(f"loss: {loss:>7f}  [{current:>5d}/{size:>5d}]")


def test_loop(dataloader, model, loss_fn):
    model.eval()
    size = len(dataloader.dataset)
    num_batches = len(dataloader)
    test_loss, correct = 0, 0

    with torch.no_grad():
        for X, y in dataloader:
            X, y = X.to(device), y.to(device)
            pred = model(X)
            test_loss += loss_fn(pred, y).item()
            correct += (pred.argmax(1) == y).type(torch.float).sum().item()

    test_loss /= num_batches
    correct /= size
    print(f"Test Error: \n Accuracy: {(100*correct):>0.1f}%, Avg loss: {test_loss:>8f} \n")

loss_fn = nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(model.parameters(), lr=learning_rate)

epochs = 20
for t in range(epochs):
    print(f"Epoch {t+1}\n-------------------------------")
    train_loop(train_dataloader, model, loss_fn, optimizer)
    test_loop(test_dataloader, model, loss_fn)
print("Done!")

硬件配置 #

• RTX 3060
• Intel i7-10700（超频至~4.2GHz）
• 64GB内存

已尝试的优化措施 #

• 增加DataLoader的workers数、固定内存，仅为20轮训练节省约15秒
• 设置torch.backends.cudnn.benchmark = True，无任何效果

现寻求该现象的原因及优化方案。

内容的提问来源于stack exchange，提问作者Addicted2Cookies