Irfan Alibay

> the `c4.xlarege` seemed fine IMHO `= 2 failed, 855 passed, 40 skipped, 1 xfailed, 2 xpassed, 3594 warnings, 3 rerun in 4059.65s (1:07:39) =` the tests ran super fast...

> then it saves us money spending more on the instance that takes less time. @mikemhenry just to point out, I think the main win in the other PR is...

@mikemhenry looks like all my quota increases were in `us-east-2` not `us-east-1`, which would explain things.

The integration tests are really the ones that have any kind of GPU testing. In many ways we might want to _only_ run those.

> @IAlibay any thoughts on how often we want to run this? It costs us ~$1.06 per run I think we need to have a chat / re-organize things -...

@mikemhenry this is more a to-do for later 😅

@hannahbaumann you did the most charge changing simulations, do you see the same thing?

Experiment 1: enforcing cpu threads to 1 doesn't seem to have any effect, could even be detrimental. Experiment 2: removing coverage on long tests shaves off ~ 1/3 of the...

Ok current state gets us to ~ 50% less, which at least gets rid of more immediate worries about 5h long runners. Merge will still be a problem though :/

> I don't know if you are using `feflow`'s HTF, but it might be worth noting that an optimization was performed when creating the HTF object that reduces the time...