Search code examples
luagputorch

Set GPU usage in Torch Machine Learning Framework

currently I'm using Theano for machine learning, now I wanted to try out Torch.

In Theano there is an option to set flags for GPU Memory usage:

os.environ["THEANO_FLAGS"] = "mode=FAST_RUN, device=gpu,floatX=float32,lib.cnmem=0.9"

So Theano uses the set capacity of the GPU, here 90%. In Torch however a similar network uses around 30% GPU load.

Is there any way to set a higher GPU load in Torch similar to Theano?

Solution

  • Torch will use as much GPU memory as it needs based on it's standard allocator.

    The amount of memory torch uses does not need to be pre-specified like the example you do in Theano.