some interesting parts from the stratechery midjourney pod:
⁃ MJ uses more than 10k GPUs
- 90% for inference
- david: the GPU cloud need to scale with 1000x
⁃ cloud providers regularly reach max GPU capacity
⁃ v4 took 9 months
⁃ 60 sec is a sweet spot for lat/quality
Nov 11, 2022 · 9:58 AM UTC