Google Colab become slower with the same code sometimes. What is the possible reasons?

user_11529019 0
Viewed 3135

I am training a CNN model with Google Colab's GPU through pytorch. My question is, even though running with the same code, it gets about three times slower sometimes(30s -> 90s in my case).

I've tried restart runtime(it clears all local variable but keep files), it doesn't work

I have seen this post, however, I've checked my GPU status, it works well with 11.5GB.

Sometimes it goes normal after disconnect it for a while though, I still want to figure out what can be the possible reason of it.

pytorch google-colaboratory machine-learning gpu
1 Answers

Guys I think I found the possible answer here

So it might be the limit of Google Colab itself. Due to their policy, sometimes you'll get fewer computation resources, which slower down the process even though no change in any code.

user_115290190
Related