Continuing the discussion from 500GB storage boost cannot be used in 2 repls
Paying for 500GB must cost a lot of money. Are you by chance using replit for non-hobby-related activities? Just interested in the diversity of Replit use cases.
Also, 256GB is a really special number 2^8, and this seems to be an infrastructure problem (GCP rejecting >256GB), so my guess is that replit would have to find a way to incrementally pay more money (e.g. for boosts >256GB you need to pay 1.2x to combat GCP rates) or make the storage “RAID-like”, spanning across multiple instances of GCP containers.
I’m using it to replace some monolithic servers that process business training video through a lot of picky, hard to setup tools. I have it working fine, now it’s just a matter of getting the 20+ hour processing times down to something reasonable, and adding more tooling… like translations and gpu acceleration.
The servers used to use a pair of 375GiB NVMe drives and had 20Gb networking. As they ran sporadically, the cost was comparable. The only issue was maintaining the image, dirty stateful code and keeping up with updates.
I may end-up having to learn how to run NixOS on kubernetes and do code-server instead, as GPU support is becoming a necessity for what we want to do in the months… and it’s 100x slower without acceleration.