Post

Strata is amazing. I love running Qwen Flash Next on 32GB of VRAM and barely even touching system RAM. there's almost no penalty for streaming experts from SSD compared to resident in RAM (for my system anyway)

Doing same, but 5080/16 and 128gb Awesome Gonna test ds and xolotl later, need to finish some work first

What kind of speeds are you getting? Might we worth trying it on my 5090/16gb vram + 64gb ram

I should fully wake up before posting... I actually have a RX9070XT and the strata repo lists numbers for my card: about 44 out/1420 prefill with the coding model