Blackmagic Forum

Hi there. Hope someone with experience can advise.

I'm getting multi-layered EXRs out of Nuke as check renders that I'm color processing with matte layer nodes in R.15v2.

As my Nuke guy is in L.A. and I'm in Toronto we're using dropbox to go back and forth, so ideally I'd like to keep the 4k files as small as possible. Resolve 15 (not tried it with 14) doesn't read DWAA compressed EXRs so we used PIZ as the first trial. Resolve does fine with them but goesn't seem to use the 2 GPUs much to decode them. Nuke, apparently likes to encode 1 line ZIP more than PIZ.

My question is does Resolve work more efficiently with ZIP or PIZ (which in theory would seem to require more CPU power to decode than PIZ)

EXRs are not decoded on GPUs, none of the compression schemes is heavy enough to gain anything from it. I don't have any links to back it up with, but I'd say that decompression is probably a non-issue. With exr-s, getting full speed reads from your disks/network with 4K frames is usually bigger problem. You can make test sequences and see if it makes any difference ofcourse.

Nuke likes single scanline zip, but for reading. Because the cache system can then pull/decode only the lines that are actually needed at current zoom/proxy level and region of interest.

Thanks, Hendrik. Flattened 4k EXRs have no trouble at 24fps on my system. but multi layered not so much There's no way to optimize them with layers so we'll try one-line zip with the next test

Struggling multilayers are probably disk IO related problem, it is faster to read only RGB than to jump around in file from one layer to another when reading masks for example. I'm not 100% sure but I think Resolve decodes full frame from exr, so there should be no technical advantage (besides computational workload difference) in reading PIZ or ZIP in Resolve. How decoding is multithreaded and how different compressions affect that is yet another story. So in the end testing is the best way to find good balance.

Hendrik Proosa wrote:Struggling multilayers are probably disk IO related problem,

In this case they're coming off a 2TB Samsung M.2 Pro so it's more a problem of decoding the layers.

I did try one thing in an attempt to speed up playback: I cached rendered a node of the entire EXR before creating a bunch of layer nodes. But strangely, even before adding the layer nodes, playback speed was unaffected compared to the original EXR. This is in 15b2.