itch.io

Thx for the reply, but i already do that, i always run DAIN alone to give it as much of my measly 6GB of VRAM from my 2060. The issue is really the CPU RAM, not the VRAM, which has already been confirmed to be a bug in PyTorch that for some reason is causing all the rendered frames to be stored in RAM and virtual memory till both are full and the program halts. The VRAM usage never went above 4GB since it was a 540p video, and rendering it in chunks works just fine. Its just that DAIN 1.0 has this bug and idk when its getting fixed... for now my fix was to just use the 0.48 version which works just fine.

https://github.com/BurguerJohn/Dain-App/issues/6
yeah the problem apparently comes from PyTorch and not DAIN according to this page on github

hm maybe ill give it a try with absurd amounts of swap

Huh, odd... Guess we wait until PyTorch gets fixed or they find a way to fix this in DAIN

I tried doing that, but it started loading all the frames from the start until it ran out of memory again

I have 16Gb of RAM and an RTX2060 with 6Gb of VRAM. Today I started interpolating a 30 minute video, 720x540 from 24 to 96 FPS. I let it run and when i came back DAIN had stopped around 60% completion and printed a message saying it had ran out of RAM. I thought it was a random error so i ran it again, and now its at 30% using 5Gb of RAM and it just keeps going up. Why is this happening? Virtual memory usage is at 25Gb too! DAIN dumps all the frames to the disk as soon as it interpolates them, so why does RAM usage increase as the videos are being processed?