What is CUDA?

"ML in a Minute" is our conversational series on answering machine learning questions. Have questions you want answered? Tweet at us.

What is CUDA (in 60 Seconds or Fewer)?

CUDA is NVIDIA's framework for using GPUs – graphical processing units – to do general purpose operations. Oftentimes, these are the same sorts of linear algebra ops that we would use for 3D graphics, but you can also use them for things like machine learning. And so you're taking these GPUs – which are traditionally used for games – and using that for high performance computing. cuDNN, a library optimized for CUDA containing GPU implementations, is often used with CUDA.

...And Are There Any Pain Points You've Run Into Using CUDA?

Yeah, this is such a new field, and it's evolving so rapidly that there's a whole bunch of different CUDA versions that don't necessarily play nicely with each other. So you're really going to have to pay attention to which specific point number of CUDA are the models that you're using and what what do they require or you're gonna have to use a service like Roboflow that abstracts that away. Perhaps create something like a Docker that has everything pre packaged for you.

Liked this? Be sure to also check out the computer vision glossary.