Hans Kristian Rosbach

@KungFuJesus This needs a rebase

@KungFuJesus @ccawley2011 Any ideas on how we can avoid/fix this? --- @Un1q32 If you can, could you test compiling with CFLAGS="-DOPTIMAL_CMP=8" to see whether that fixes it? ex: `cmake ....

And I just realized you already said that it also fails on 2.1.7. I guess you didn't find any version where it does work then. Well, that neatly throws my...

Btw, this might actually coincide with times when the OSS-brikken loses contact with the mobile network over a longer period of time (Coverage is really bad in our basement, and...

Just to add a little context that I feel is missing here, zlib does not guarantee bit-identical compressed output, and that is not part of the spec. In fact zlib...

On my 7900XTX with 24GB vram, I get the same kinds of images after vram usage reaches ~17.5GB (so right around 75%). That is also the point where generating images...

7900XTX with 24GB of vram Windows 10, Radeon 23.4.3, DirectML 1.10.1 Sampler: `Euler a` Prompt: `house with impossible features`, no negative prompt Using SD v1.5 model downscaled to 16bit, resulting...

I did some quick followup tests with the DirectML.dll v 1.11.0 using parameters `--medvram --no-half --no-half-vae --precision=full`, and it reduces the rate of memory usage increase up to 16.6GB. So...

### Some more observations: I managed to trigger corrupted images *once*, but I did not manage to repeat it. Generating a dozen images and then increasing to `batch-size 2` worked...

> It is quite strange and interesting. If this problem occurs due to high vram occupation, limiting it might help. But unfortunately I can't test and reproduce it because I...