Results from running on p3.16xlarge

[paulbalaji]

I will probably do some short runs on the submitted versions with a bigger instance (e.g. a p3.16xlarge, at $27 an hour...), just out of interest, but they will not be assessed.

Just curious when this will happen - super curious to see some graphs for the final code :)

[m8pple]

I'm still trying to diagnose the machine-freeze bug from #47, as otherwise I can't do a full run - certainly not on a $27 an hour version, as when it freezes it sometimes takes 10 minutes (~$5) just to get it to terminate. I've found a reliable freezing implementation with a fair number of commits, so am currently doing bisection to try to work out what is happening. Still not sure if it is GPU related or memory related. Once that is done and any silly crashing bugs are fixed, I'll do a pretty long run on the g2.2xlarge. Then a run on a bigger instance, but only for about 4 hours (for financial reasons), so about 1 min/puzzle. I'm quite interested to see the spread too, as we have a much more diverse class in terms of course/dept. this year.

…

On 28/11/2017 04:09, Paul Balaji wrote: I will probably do some short runs on the submitted versions with a bigger instance (e.g. a p3.16xlarge, at $27 an hour...), just out of interest, but they will not be assessed. Just curious when this will happen - just super curious to see some graphs for the final code :) — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#66>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AC4nQa6EwUzwcXgiOLKIuOh8mtYxJrSSks5s64eCgaJpZM4Qsqac>.