99% of final pass - Stuck

Started by SuddenPlanet, December 20, 2021, 11:58:35 AM

Previous topic - Next topic

SuddenPlanet

Anyone ever do a large render and have it get stuck on the final filtering pass?

My render is a 16,000 x 16,000 perspective render from above (top view but using perspective, not ortho).

I'm using the Mitchell-Netravali filter, and it's so slow (not multi-threaded).

Are any of the filtering methods multi-threaded?

Thanks,

Derek

WAS

The final filtering pass are technically layers, overlayed on your final image. That being said, if you're using more than just GISD, I'm pretty sure there are 16,000x16,000 rasterization being generated ontop of the final render tif, and then applied to your image. This is where people that don't have enough RAM often crash when their render is already filling RAM. Additionally, some of the filters like Bloom/Atmo bloom are doing calculations to a degree over a basic image effect like Starburst.

SuddenPlanet

OK, I'll just wait and see if it finishes.  I'm only using 75GB of my 128GB RAM, and I've done 16k renders before without issues.

The render itself only too 37 minutes, but this filtering pass has been going 2 hours so far.

I'll try the other filters to see if any are faster.  I assume that Box is the fastest/lowest quality, and they are in order of quality but not sure.

Thanks,

Derek

WAS

That does seem a bit extreme, but I'm not sure what the file is like to know. I know sometimes they can hang, like I once did a starburst that was too strong, and it made the render window go unresponsive, and when it finally applied I just had a white blown out image xD

Nala1977

it happens to me a lot of times even on smaller resolutions.

WAS

Yeah it can definitely hang even at low resolutions. If there is a lot of shadow I think there is a lot more GISD applied for example.

I'm not really a fan of TG Bloom (though atmo bloom is wonderful), so I often just do bloom in post with levels, desaturation, Gaussian blur and than layer style as screen.

SuddenPlanet

Are you talking about Post Process Effects, or Pixel Filtering (Box, Tent, Cubic etc...)?

It's still stuck, so it looks like I'll have to kill it.  I'll let it go one more hour just in case.

Derek

WAS

Post Processing. Pixel filtering should be done at render time, not post. You see the result when the bucket is finished.

SuddenPlanet

Quote from: WAS on December 20, 2021, 03:22:57 PMPost Processing. Pixel filtering should be done at render time, not post. You see the result when the bucket is finished.
Ah ok then I guess my issue had nothing to do with the pixel filtering.

Not sure why it got stuck, but I had to kill it.  I didn't turn on any post effects.  Anti-aliasing bloom is on by default, so I'll try turning that off.

Thanks,

Derek

WAS


SuddenPlanet


WAS

Hmm. But you do mention 75gb of 128. If a GISD pass at same resolution is being applied it could be around the same size in memory, which would be about 150gb so the pass could be fragmented with scratch disk (which could become errored). I am not positive and someone more technical with the code would know better than me.

Dune

At least the rendertime itself is extremely short, so you could have another go. Perhaps try two crops...

SuddenPlanet

I'm trying some more things.  After doing some more renders it seems like the render itself finishes, but it croaks on the saving out part.
I'm going to try turning off the auto save so it doesn't automatically save an exr of my render to see if that makes a difference.

WAS

Maybe try 32bit TIF output instead of EXR? There isn't much benefit to a 32bit EXR as it doesn't contain extra data the format is used for.

Oh, that's right. TG still just does half-float. >.<