Performance difference between torch.zeros((...), device=dev) and torch.zeros((...)).to(dev)?

[bananacode]

I’ve got two questions regarding performance.

1. Is there a performance difference between first creating a tensor, then sending it to the device with the “to” function, and with specifying the device directly during creation?

2. When having tensors on the GPU and switching data type (e.g from int to float), does it do this directly on the device or first moves it back to CPU before casting?

[SimonW (Simon Wang)]

@bananacode

Is there a performance difference between first creating a tensor, then sending it to the device with the “to” function, and with specifying the device directly during creation?

yes, one is directly creating on the particular device. and the other one creates on cpu and does a copy (if dev is not cpu)

@bananacode

When having tensors on the GPU and switching data type (e.g from int to float), does it do this directly on the device or first moves it back to CPU before casting?

directly on the device

[bananacode]

Thanks for the rapid response. So to confirm, torch.zeros((…), device=dev) is the faster way?

[SimonW (Simon Wang)]

yes you are correct indeed