DEV Community

Super Kai (Kazuya Ito)
Super Kai (Kazuya Ito)

Posted on • Edited on

Check PyTorch version, CPU and GPU(CUDA) in PyTorch

#python #pytorch #version #device

Buy Me a Coffee☕

*My post explains how to create and acceess a tensor.

__version__ can check PyTorch version as shown below. *__version__ can be used with torch but not with a tensor:

import torch torch.__version__ # 2.2.1+cu121
Enter fullscreen mode Exit fullscreen mode

cpu.is_available(), cpu.device_count() or cpu.current_device() can check if CPU is available, getting a scalar as shown below:

*Memos:

  • cpu.is_available(), cpu.device_count() or cpu.current_device() can be used with torch but not with a tensor.
  • cpu.device_count() can get the number of CPUs. *It always gets 1:
  • cpu.current_device() can get the index of a currently selected CPU. *It always gets cpu: 
import torch torch.cpu.is_available() # True  torch.cpu.device_count() # 1  torch.cpu.current_device() # cpu
Enter fullscreen mode Exit fullscreen mode

cuda.is_available() or cuda.device_count() can check if GPU(CUDA) is available, getting a scalar as shown below:

*Memos:

  • cuda.is_available() or cuda.device_count() can be used with torch but not with a tensor.
  • cuda.device_count() can get the number of GPUs. 
import torch torch.cuda.is_available() # True  torch.cuda.device_count() # 1
Enter fullscreen mode Exit fullscreen mode

In addition, you can use cuda.current_device(), cuda.get_device_name() or cuda.get_device_properties(), getting a scalar as shown below:

*Memos:

  • cuda.current_device(), cuda.get_device_name() or cuda.get_device_properties() can be used with torch but not with a tensor.
  • cuda.current_device() can get the index of a currently selected GPU.
  • cuda.get_device_name() can get the name of a GPU. *Memos:
    • The 1st argument with torch is device(Optional-Default:None-Type:(str, int or device().
    • If it's None, cuda.current_device() is used.
    • The number must be zero or positive.
    • Only cuda can be set to device.
    • My post explains device().
  • cuda.get_device_properties() can get the properties of a GPU. *Memos:
    • The 1st argument with torch is device(Required-Type:str, int or device()).
    • The number must be zero or positive.
    • Only cuda can be set to device.
    • My post explains device(). 
torch.cuda.current_device() # 0  torch.cuda.get_device_name() torch.cuda.get_device_name(device='cuda:0') torch.cuda.get_device_name(device='cuda') torch.cuda.get_device_name(device=0) torch.cuda.get_device_name(device=torch.device(device='cuda:0')) torch.cuda.get_device_name(device=torch.device(device='cuda')) torch.cuda.get_device_name(device=torch.device(device=0)) torch.cuda.get_device_name(device=torch.device(type='cuda')) torch.cuda.get_device_name(device=torch.device(type='cuda', index=0)) # Tesla T4  torch.cuda.get_device_properties(device='cuda:0') torch.cuda.get_device_properties(device='cuda') torch.cuda.get_device_properties(device=0) torch.cuda.get_device_properties(device=torch.device(device='cuda:0')) torch.cuda.get_device_properties(device=torch.device(device='cuda')) torch.cuda.get_device_properties(device=torch.device(device=0)) torch.cuda.get_device_name(device=torch.device(type='cuda')) torch.cuda.get_device_name(device=torch.device(type='cuda', index=0)) # _CudaDeviceProperties(name='Tesla T4', major=7, minor=5, # total_memory=15102MB, multi_processor_count=40)
Enter fullscreen mode Exit fullscreen mode

!nvidia-smi can get the information about GPUs as shown below:

!nvidia-smi Wed May 15 13:18:15 2024 +---------------------------------------------------------------------------------------+ | NVIDIA-SMI 535.104.05 Driver Version: 535.104.05 CUDA Version: 12.2 | |-----------------------------------------+----------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+======================+======================| | 0 Tesla T4 Off | 00000000:00:04.0 Off | 0 | | N/A 56C P0 28W / 70W | 105MiB / 15360MiB | 0% Default | | | | N/A | +-----------------------------------------+----------------------+----------------------+ +---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| +---------------------------------------------------------------------------------------+
Enter fullscreen mode Exit fullscreen mode

Top comments (0)