Re:Mark Shuttleworth sees increased demand for enterprise Ub...
by opportune on Tuesday April 30, 2019 @05:04PM
(#19791915)
That also works, I guess I just mean that I don't think people people should be relying on training on local GPUs. Since yeah, a lot of use cases require multi-GPU training running on the order of days, seems very amateurish to really need a beefy local GPU to get your work done.
I guess I rarely run into ubuntu vs macos compatibility issues while I do notice the lack of QOL features on ubuntu. I would rather stay on macos
What we'll often do is use a small dataset locally on our desktop to confirm that something works, and then ship it to a multi-GPU setup in our cluster to train things for real-world uses. It's much easier to do this when you're developing on the same OS that is running in the cluster, hence Ubuntu desktop.