0

I'm using a m1 pro macbook, and I installed pytoch directly on my machine using pip3 install torch torchvision torchaudio this command. Also I created a virtualized env(phthon 3.10) with conda and using the same install command.

When I run a very small training set the result differed significantly. The non-virtualized version resulted in 20s and virtualized version resulted in 40s.

First I thought maybe the Apple SIP has some effect on the execution of the conda env, so I disabled it. But it turned out that it made no difference.

So, what is causing training process in conda env so much slower?

Liang
  • 1,127
  • 2
  • 11
  • 22
  • Not enough info. What "installed...directly" means is unclear - macOS does not come with Python installed. Please provide a minimal example that others can recreate. Also, please indicate whether your Conda installation is **osx-arm64** (native) or is **osx-64** (emulated). Possibly related: https://stackoverflow.com/q/70240506/570918 – merv Jul 07 '23 at 15:28

1 Answers1

0

Turns out I'm using the intel version of anaconda, after reinstalling with the m1 version, the performance issue goes away.

Liang
  • 1,127
  • 2
  • 11
  • 22