| 知乎专栏 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)
输出结果是 cpu 表示当前环境使用的的是 cpu 而不是 GPU
D:\workspace\netkiller\.venv\Scripts\python.exe D:\workspace\netkiller\test\duda.py cpu
使用 nvidia-smi 命令检测显卡，是否支持 CUDA
PS C:\Users\neo> nvidia-smi Sun Nov 17 12:29:15 2024 +---------------------------------------------------------------------------------------+ | NVIDIA-SMI 546.67 Driver Version: 546.67 CUDA Version: 12.3 | |-----------------------------------------+----------------------+----------------------+ | GPU Name TCC/WDDM | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+======================+======================| | 0 NVIDIA GeForce RTX 3050 ... WDDM | 00000000:02:00.0 Off | N/A | | N/A 54C P0 8W / 50W | 0MiB / 6144MiB | 0% Default | | | | N/A | +-----------------------------------------+----------------------+----------------------+ +---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| | No running processes found | +---------------------------------------------------------------------------------------+ PS C:\Users\neo>
这里可以看到当前显卡是支持 CUDA Version: 12.3
下载 cuda 开发包 https://developer.nvidia.com/
release-notes https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html
检查 CUDA 版本
PS C:\Users\neo> nvcc -V nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2024 NVIDIA Corporation Built on Thu_Sep_12_02:55:00_Pacific_Daylight_Time_2024 Cuda compilation tools, release 12.6, V12.6.77 Build cuda_12.6.r12.6/compiler.34841621_0
查询设备
PS C:\Users\neo> & 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\extras\demo_suite\deviceQuery.exe'
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\extras\demo_suite\deviceQuery.exe Starting...
CUDA Device Query (Runtime API) version (CUDART static linking)
Detected 1 CUDA Capable device(s)
Device 0: "NVIDIA GeForce RTX 3050 6GB Laptop GPU"
CUDA Driver Version / Runtime Version 12.6 / 12.6
CUDA Capability Major/Minor version number: 8.6
Total amount of global memory: 6144 MBytes (6441926656 bytes)
(20) Multiprocessors, (128) CUDA Cores/MP: 2560 CUDA Cores
GPU Max Clock rate: 990 MHz (0.99 GHz)
Memory Clock rate: 5501 Mhz
Memory Bus Width: 96-bit
L2 Cache Size: 1572864 bytes
Maximum Texture Dimension Size (x,y,z) 1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
Maximum Layered 1D Texture Size, (num) layers 1D=(32768), 2048 layers
Maximum Layered 2D Texture Size, (num) layers 2D=(32768, 32768), 2048 layers
Total amount of constant memory: zu bytes
Total amount of shared memory per block: zu bytes
Total number of registers available per block: 65536
Warp size: 32
Maximum number of threads per multiprocessor: 1536
Maximum number of threads per block: 1024
Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
Max dimension size of a grid size (x,y,z): (2147483647, 65535, 65535)
Maximum memory pitch: zu bytes
Texture alignment: zu bytes
Concurrent copy and kernel execution: Yes with 1 copy engine(s)
Run time limit on kernels: Yes
Integrated GPU sharing Host Memory: No
Support host page-locked memory mapping: Yes
Alignment requirement for Surfaces: Yes
Device has ECC support: Disabled
CUDA Device Driver Mode (TCC or WDDM): WDDM (Windows Display Driver Model)
Device supports Unified Addressing (UVA): Yes
Device supports Compute Preemption: Yes
Supports Cooperative Kernel Launch: Yes
Supports MultiDevice Co-op Kernel Launch: No
Device PCI Domain ID / Bus ID / location ID: 0 / 2 / 0
Compute Mode:
< Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >
deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 12.6, CUDA Runtime Version = 12.6, NumDevs = 1, Device0 = NVIDIA GeForce RTX 3050 6GB Laptop GPU
Result = PASS
PS C:\Users\neo>
带宽测试
PS C:\Users\neo> & 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\extras\demo_suite\bandwidthTest.exe' [CUDA Bandwidth Test] - Starting... Running on... Device 0: NVIDIA GeForce RTX 3050 6GB Laptop GPU Quick Mode Host to Device Bandwidth, 1 Device(s) PINNED Memory Transfers Transfer Size (Bytes) Bandwidth(MB/s) 33554432 5971.0 Device to Host Bandwidth, 1 Device(s) PINNED Memory Transfers Transfer Size (Bytes) Bandwidth(MB/s) 33554432 6413.3 Device to Device Bandwidth, 1 Device(s) PINNED Memory Transfers Transfer Size (Bytes) Bandwidth(MB/s) 33554432 101343.0 Result = PASS NOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled. PS C:\Users\neo>
CUDA 开发包安装正确，但是 pytorch 代码 torch.cuda.is_available() 始终返回 False 无法使用 CUDA
import torch print(torch.__version__) print(torch.cuda.is_available())
输出结果
D:\workspace\netkiller\.venv\Scripts\python.exe D:\workspace\netkiller\test\duda.py 2.5.1+cpu False
问出出在，你安装 pytorch 的时候没有检测到 cuda 系统便默认给你安装 cpu 版本，解决方法是重新安装 cuda 版本的 pytorch
进入这个页面 https://pytorch.org/get-started/locally/ 选择你 CUDA 版本的 pytorch
(.venv) PS D:\workspace\netkiller> pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124
Looking in indexes: https://download.pytorch.org/whl/cu124
Requirement already satisfied: torch in d:\workspace\netkiller\.venv\lib\site-packages (2.5.1)
Requirement already satisfied: torchvision in d:\workspace\netkiller\.venv\lib\site-packages (0.20.1)
Collecting torchaudio
Downloading https://download.pytorch.org/whl/cu124/torchaudio-2.5.1%2Bcu124-cp312-cp312-win_amd64.whl (4.1 MB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 4.1/4.1 MB 7.3 MB/s eta 0:00:00
Requirement already satisfied: filelock in d:\workspace\netkiller\.venv\lib\site-packages (from torch) (3.16.1)
Requirement already satisfied: typing-extensions>=4.8.0 in d:\workspace\netkiller\.venv\lib\site-packages (from torch) (4.12.2)
Requirement already satisfied: networkx in d:\workspace\netkiller\.venv\lib\site-packages (from torch) (3.4.2)
Requirement already satisfied: jinja2 in d:\workspace\netkiller\.venv\lib\site-packages (from torch) (3.1.4)
Requirement already satisfied: fsspec in d:\workspace\netkiller\.venv\lib\site-packages (from torch) (2024.10.0)
Requirement already satisfied: setuptools in d:\workspace\netkiller\.venv\lib\site-packages (from torch) (75.3.0)
Requirement already satisfied: sympy==1.13.1 in d:\workspace\netkiller\.venv\lib\site-packages (from torch) (1.13.1)
Requirement already satisfied: mpmath<1.4,>=1.1.0 in d:\workspace\netkiller\.venv\lib\site-packages (from sympy==1.13.1->torch) (1.3.0)
Requirement already satisfied: numpy in d:\workspace\netkiller\.venv\lib\site-packages (from torchvision) (2.0.2)
Requirement already satisfied: pillow!=8.3.*,>=5.3.0 in d:\workspace\netkiller\.venv\lib\site-packages (from torchvision) (11.0.0)
Collecting torch
Downloading https://download.pytorch.org/whl/cu124/torch-2.5.1%2Bcu124-cp312-cp312-win_amd64.whl (2510.7 MB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 2.5/2.5 GB 13.6 MB/s eta 0:00:00
Requirement already satisfied: MarkupSafe>=2.0 in d:\workspace\netkiller\.venv\lib\site-packages (from jinja2->torch) (3.0.2)
Installing collected packages: torch, torchaudio
Attempting uninstall: torch
Found existing installation: torch 2.5.1
Uninstalling torch-2.5.1:
Successfully uninstalled torch-2.5.1
Successfully installed torch-2.5.1+cu124 torchaudio-2.5.1+cu124
再次验证
import torch print(torch.__version__) print(torch.cuda.is_available()) D:\workspace\netkiller\.venv\Scripts\python.exe D:\workspace\netkiller\test\duda.py 2.5.1+cu124 True
https://developer.nvidia.cn/cudnn
下载解压
将 C:\Users\neo\Downloads\cudnn-windows-x86_64-9.5.1.17_cuda12-archive.zip\cudnn-windows-x86_64-9.5.1.17_cuda12-archive 下面的三个目录，复制到 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6