CUDA可用，共有 1 个GPU设备可用。

当前使用的GPU设备索引：0

当前使用的GPU设备名称：NVIDIA T1000

GPU显存总量：4.00 GB

已使用的GPU显存：0.00 GB

剩余GPU显存：4.00 GB

PyTorch版本：1.10.1+cu102

python
展开代码
import torch
# 检查CUDA是否可用
cuda_available = torch.cuda.is_available()
if cuda_available:
    # 获取GPU设备数量
    num_gpu = torch.cuda.device_count()
    # 获取当前使用的GPU索引
    current_gpu_index = torch.cuda.current_device()
    # 获取当前GPU的名称
    current_gpu_name = torch.cuda.get_device_name(current_gpu_index)
    # 获取GPU显存的总量和已使用量
    total_memory = torch.cuda.get_device_properties(current_gpu_index).total_memory / (1024 ** 3)  # 显存总量(GB)
    used_memory = torch.cuda.memory_allocated(current_gpu_index) / (1024 ** 3)  # 已使用显存(GB)
    free_memory = total_memory - used_memory  # 剩余显存(GB)
    print(f"CUDA可用，共有 {num_gpu} 个GPU设备可用。")
    print(f"当前使用的GPU设备索引：{current_gpu_index}")
    print(f"当前使用的GPU设备名称：{current_gpu_name}")
    print(f"GPU显存总量：{total_memory:.2f} GB")
    print(f"已使用的GPU显存：{used_memory:.2f} GB")
    print(f"剩余GPU显存：{free_memory:.2f} GB")
else:
    print("CUDA不可用。")
# 检查PyTorch版本
print(f"PyTorch版本：{torch.__version__}")
import torch
print(f"CUDA版本：{torch.version.cuda}")

windows先装显卡驱动，再装CUDA10.2，最后装了pytorch。

pip install torch==1.10.1+cu102 torchvision==0.13.1+cu102 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu102

python
展开代码
import torch
import os
import subprocess

def check_gpu_availability():
    print("==== Checking GPU Availability ====")
    if torch.cuda.is_available():
        print(f"GPU is available. Number of GPUs: {torch.cuda.device_count()}")
        for i in range(torch.cuda.device_count()):
            print(f"GPU {i}: {torch.cuda.get_device_name(i)}")
    else:
        print("No GPU available. Please check your hardware or drivers.")
        return False
    return True

def check_cuda_and_cudnn():
    print("\n==== Checking CUDA and cuDNN Availability ====")
    if torch.cuda.is_available():
        print(f"CUDA version: {torch.version.cuda}")
        print(f"cuDNN version: {torch.backends.cudnn.version()}")
        print(f"cuDNN enabled: {torch.backends.cudnn.enabled}")
    else:
        print("CUDA or cuDNN not available. Please check the installation.")

def test_gpu_computation():
    print("\n==== Testing Simple Computation on GPU ====")
    try:
        
        device = torch.device("cuda:2")
        a = torch.randn(50, 50, device=device)
        b = torch.randn(50, 50, device=device)
        c = torch.matmul(a, b)
        print("Matrix multiplication on GPU successful.")
    except Exception as e:
        print(f"Error during computation on GPU: {e}")

def check_gpu_memory():
    print("\n==== Checking GPU Memory Usage ====")
    try:
        gpu_memory = torch.cuda.memory_summary()
        print(gpu_memory)
    except Exception as e:
        print(f"Error retrieving GPU memory info: {e}")

def run_nvidia_smi():
    print("\n==== Running nvidia-smi Command ====")
    try:
        # Run the nvidia-smi command and capture the output
        result = subprocess.run(["nvidia-smi"], stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)
        if result.returncode == 0:
            print(result.stdout)
        else:
            print(f"Error running nvidia-smi: {result.stderr}")
    except Exception as e:
        print(f"Error running nvidia-smi command: {e}")

def main():
    print("==== Starting GPU Health Check ====")
    
    # Step 1: Check GPU availability
    if not check_gpu_availability():
        return
    
    # Step 2: Check CUDA and cuDNN versions
    check_cuda_and_cudnn()
    
    # Step 3: Test GPU computation
    test_gpu_computation()
    
    # Step 4: Check GPU memory usage
    check_gpu_memory()
    
    # Step 5: Run nvidia-smi to check GPU status
    run_nvidia_smi()
    
    print("\n==== GPU Health Check Completed ====")

if __name__ == "__main__":
    main()