D
@txt_file In my case, each chunk is: 2-60 minutes (80% < 15minutes), 1 cpu core, <8GB RAM. Random crashes are fine (any ongoing chunks of work would have to be restarted).
I think many SaaS companies have such tasks (if only they looked). E.g. cassandra compactions could be done this way (with a bit of patching, though).
Does anyone know of a cloud provider who offers very cheap rates (i.e. lower than Hetzner) in exchange for extremely low uptime, e.g. <50%?
In times of solar it seems it would make sense to just keep old(ish) servers connected and only power them up when electricity is free (or negative even).
I have a compute workload which I expect to complete in ~150days(!); if I could get cheap compute anywhen before that, I'd pay for it. [And I expect to have future workloads which can be delayed 2 days without much trouble.]
@clock @MLE_online Our kitchen's audio setup runs on a similarly "de-batteried" smartphone. Except no voltage reduction at all, I just put the USB 5V directly on the battery terminal and let the on-board voltage stabilizer deal with it.
@BartoszMilewski I'm morbidly curious about how an AI-genned video for some of the more advanced subjects would turn out. 
I would be more accepting of such compiler pedantry if cppreference.com wouldn't side with _my_ interpretation of the standard.
error: unknown type name 'nullptr_t'; did you mean 'std::nullptr_t'?
No, compiler. Of course I meant my own, very special nullptr_t. I want to scream, not stare into whatever lies at its end...
Because I was reminded about it when discussing code-reviews recently, I want to recommend https://dokumen.pub/introduction-to-the-personal-software-processsm-0201548097-9780201548099.html . I found the book both incredibly boring (the writing style) and very well-researched and eye-opening (the information).
