PyTorch uses CuBLAS [1] under the hood, among other libraries, so basic linear a... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

		p1esk on Dec 7, 2018 \| parent \| context \| favorite \| on: PyTorch 1.0 is out PyTorch uses CuBLAS [1] under the hood, among other libraries, so basic linear algebra ops should be fast. You might also look at CuPy [2], especially if you like NumPy. [1] https://developer.nvidia.com/cublas [2] https://cupy.chainer.org/

amelius on Dec 8, 2018 [–]

Now I'm wondering if anybody wrote a DL library on top of CuPy, and if such library could be competitive with PyTorch in terms of performance.

smhx on Dec 8, 2018 | [–]

Yes, it is called Chainer and it's quite competitive.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact