What is the difference between using tensor.cuda() and tensor.to(torch.device(“cuda:0”))

distributed
1

Using PyTorch, what is the difference between the following two methods in sending a tensor to GPU:

Method 1:

X = np.array([[1, 3, 2, 3], [2, 3, 5, 6], [1, 2, 3, 4]])
X = torch.DoubleTensor(X).cuda()

Method 2:

X = np.array([[1, 3, 2, 3], [2, 3, 5, 6], [1, 2, 3, 4]])
X = torch.DoubleTensor(X)

device = torch.device("cuda:0")
X = X.to(device)

Similarly, is there any difference in the same two methods above when applied to sending a model to GPU:

Method A:

gpumodel = model.cuda()

Method B:

device = torch.device("cuda:0")
gpumodel = model.to(device)

Many thanks in advance!

2

there is no difference

3

There might be a difference, if you were resetting the default CUDA device via torch.cuda.set_device() as seen in this code snippet:

torch.cuda.set_device('cuda:1')
x = torch.randn(1).cuda()
print(x)
> tensor([0.9038], device='cuda:1') # uses the default device now

y = torch.randn(1).to('cuda:0')
print(y)
> tensor([-0.7296], device='cuda:0') # explicitly specify cuda:0
4

Ok thanks @iffiX for confirming they are both essentially doing the same thing.

5

Ah yes, thats important, I forgot this :slightly_smiling_face:

6

Ok many thanks @ptrblck for the more detailed answer where the 2nd method is specifying which GPU device to use and the 1st method is just using the default GPU device.

7

Their syntax varies slightly, but they are equivalent:

⠀ .to(name) .to(device) .cuda()
CPU to('cpu') to(torch.device('cpu')) cpu()
Current GPU to('cuda') to(torch.device('cuda')) cuda()
Specific GPU to('cuda:1') to(torch.device('cuda:1')) cuda(device=1)

Note: the current cuda device is 0 by default, but this can be set with torch.cuda.set_device().