@corwii

Hello again! Got a little observation I wanted to share about this slowdown issue I've had for a while. But not to worry I seemed to have found a temporary fix that might help you see an issue if there is one. Idk if it's just a 16gb card issue. Oddly at times when it loads Minimax ref model and it decides it's going slow, I notice it acts like there's no learned runs to give an ETA. I notice that if I start a generation, and it's saying "Calculating ETA", it's going to be a slow run. How I seem to fix it is switching to the 4 step Minimax model, setting it to 5 seconds for speed, generating the whole thing first. Once it finishes, I can switch back to the normal Minimax pruned model, and suddenly it's telling me the ETA and how many learned runs it knows. And then it goes my normal expected speeds. So far it hasn't failed me even on the latest dev branch. Still loving the app and everything you've done! It literally runs 24/7 on my system. My Settings if it matters: Auto-tune turned off, VRAM and RAM allocators set to pytorch, no smart memory pinning or read ahead, and video profile set to Profile 2 which is what Auto-tune recommends for my 5080 16gb. Everything else default.

Just wanted to share that is seems to be mostly fixed or at least midigated on the dev branch. Been using it for a few days now. It seems to be back to normal speeds (Anywhere from 30 minutes to an hour depending on number of seconds), mind you I'm not using any optimizations or turbo stuff for the fullest quality. So it's taking roughly 40 seconds to 2 minutes per step which is my normal base speed. But I notice that every so often even on the same settings that some scenes end up going back to being slow. Like I just generated a 13.7 second clip in around 40 minutes, yet the next one I did took 1 hour and 45 minutes for 12.3 seconds when technically it should have been the same or slightly faster. But it only happened a few times, not sure if you have a lingering bug still going about. But for the most part my speeds are normal again from what I can tell. I'll let you know if I discover anything else.

Seem to be having a huge slowdown since the recent updates. Had to roll back to version 2.7.0 to gain my speeds again with Minimax H3. Gemini wrote this issue report. I can confirm that rolling back to 2.7.0 and disabling all the MMGP stuff fixed my issue and I'm back to normal speeds. Disabling MMGP on version 3.1 didn't work, was still 3x slower. Gemini says it's a bug with MMGP on 16gb cards. Performance Regression on 16GB GPUs (RTX 5080) Issue: Updating to v3.0 / v3.1 causes generation times on the MiniMax H3 model to drop significantly (3x slower), going from ~1 hour per generation to over 4 hours (step times ballooning from ~3 mins to 6+ mins). Hardware: • GPU: Nvidia RTX 5080 (16GB VRAM) • RAM: 64GB System RAM Details: Disabling MMGP or switching backend profiles back to raw PyTorch in v3.1 does not resolve the slowdown. Rolling the codebase back entirely bypasses the issue and restores normal generation tracking speeds, indicating that the new MMGP memory planning scheduler / weight-shuttling logic is over-allocating or aggressively paging out VRAM on 16GB cards, severely degrading rendering performance.