search

OverLordGoldDragon / ssqueezepy

Synchrosqueezing, wavelet transforms, and time-frequency analysis in Python
MIT License
599 stars 92 forks source link

Using Multiple GPU #72

Closed tiavila closed 2 years ago

tiavila commented 2 years ago

Is there an environ setting to enable which GPU to use and the ability to use more than one for CUPY and Torch?

I am on a machine with 4 GPUs but it only uses GPU #1 which is in use by other processes.

tiavila commented 2 years ago

Or a way to provide a list of input values and get a list of Wx back.

OverLordGoldDragon commented 2 years ago

No built-in support for multi-GPU, but maybe torch/cupy can be told to use a particular device.

list of input values and get a list of Wx back.

Batched computation is supported, with separate inputs stacked along dim0:

x_batched = np.vstack([x0, x1])   # x0.ndim == x1.ndim == 1