This week we installed a new server and migrated waifu2x onto it. This server has 2 GPU's instead of one, and a single card is around 30% faster then the old server. So in turn the server should be 2.6 times as fast as our old server. Queues should be mostly eliminated unless its at absolute peak traffic. I measured the power usage and at max load it draws 720 watts of power... that's quite something.
Server is paid for and hosted out of my own pockets, the patreon money is still used fully to pay for programmer cost. As soon as we're done with our current project we're going to try and improve this server as its not very efficiently coded at the moment. Every time a request is made for resize the backend program is booted and the model is loaded into GPU memory and then removed again. A more efficient route would be to keep it into memory while it waits for requests.
EDIT: New server is broke, always lucky! Gonna try to fix it, put live site back on the old server.
EDIT2: Should be fixed now.
Turner Coria
2024-04-21 20:32:48 +0000 UTCymjxuji
2020-03-03 06:12:00 +0000 UTCBrav0
2019-12-13 22:30:44 +0000 UTCLadripper
2019-11-29 22:43:02 +0000 UTCBill Lemmond
2019-09-28 16:32:02 +0000 UTCRagnar Oktopod
2019-09-20 21:29:48 +0000 UTC