Ethical/opinion: Free GPU instances

Deep Learning
#1

I see some thread on the free lunch theorem (?) so this might be timely…my mind has been meowing a bit over those free GPU instances available on Colab and Kaggle kernels. While I’m not rolling in dough, I can afford to build my own DL box, or pay for non-free services like Paperspace. Now, my cheap/selfish self prefers to go for the free ones, even with the time limits and other restrictions. But that got me thinking, if anyone is familiar enough with these companies to chime in on this:

To what extent are folks like me (can afford to pay, but don’t) hogging up the available free instances, thus shutting others out? I recognize there are people who really cannot afford to pay, so I would like to make sure that they get a chance to use these great resources. If there are zillions of slots available, then no problem, but if us “free lunchers” as a group really do create problems for those who are less fortunate, or if we’re trending towards that scenario in the near future, then I’d rather go another route. Thoughts?

1 Like
#2

Well I found a crucial clue in a thread on the Beginner 2018 forum, that led to this winner:

So it seems that either by lottery or possibly some form of blacklisting, there’s a chance you may end up with a severely crippled GPU (95% used) on Collab. That tells me the resources are likely stretched too thin already, and who knows if Kaggle will soon have the same issues. Though I fault no one for having different inklings, I’m personally more inclined to go by “if you can afford to pay, then do.”

(Fun, totally irrelevant fact: I just had to look up whether inkling and inclination come from the same origin, and it turns out they do not. The former is from Middle English, the latter is from Middle French.)