Multipass crippling time penalty?

[Rhodesy]

Im trying to debug a very slow rendering scene. Its an interior that has run for 15 hours and only reached 55 passes on a 16core xeon PC (CB 22,000). Its predicted time is 53 hours to get to 250passes. There is definitely something wrong here and I think I have tracked it down to the multipass. If I switch it off the predicted time drops to 6hrs which I would be very happy with. My initial thought was to switch off filtering for all the multipasses but that only takes it down to 46 hours predicted time which is still far too slow. I have 14 multipasses including a few light select layers, so quite a lot and I could probably drop a few if needed. RAM levels are fine.

Im using the C4D latest daily build and corona VFB but I'm assuming its the same system in the MAX version. I was wondering if anyone had any experience with this sort of extreme slow down with passes? Thanks

[Rhodesy]

I've made a bit of progress and managed to get the predicted time down to 26 hours by switching the bloom and glare down to 0. Also the denoiser but as far as I know that only kicks in once the render has stopped anyway. I did think it was odd that the bloom and glare was being calculated from pass one but I suppose it has to be for the interactive use. Perhaps there could be a global switch for on and off in the VFB.

[Fluss]

+1 an "apply effects on render end" checkbox would drastically decrease render times

[Ludvik Koutny]

This sounds more like running out of RAM honestly. Render passes create memory overhead, especially in larger resolutions. And bloom+glare are just another set of passes that needs to be generated.

[Jpjapers]

I had the same problem at work. had 16gb RAM and tried to render a 15000px wide image. said it would take 40+ hours and frequently crashed out.
We upped the machine to 128GB and it finished within 6.

[Rhodesy]

I dont think its RAM in this case. I've got 64GB and windows says its taking 24GB for the scene with the CPU constantly around the 95% mark. I do have some IES lights in the scene if that makes any difference. My previous tests were with mat override which clocked the 6hr estimate with no MP but was still 53hrs when I had bloom and glare on. re testing without MP it said 16hrs for the fully textured scene with no MP and 26 with MP. Its better but still quite slow. Im hoping I can get away with 100 passes and some denoising to get in in the 12-15hr overnight render slot.

[Jpjapers]

Hmmm youre probably right.  What passes do you have set up?

[Ludvik Koutny]

Glare and bloom may indeed increase the time if resolution is too large. Switch to apply it once at the render render would really make sense.

[Rhodesy]

Would you say 50% increase in render times with 14 passes is about right? Does filtering put a big strain on the overall time? My testing isn't really conclusive and shows only marginal difference.

[Jpjapers]

How big is your render?
Ive been doing 4k renders for the past week and glare/bloom usually only takes about 15-20 seconds to render and about double that if i make changes in the post panel.

[Rhodesy]

6500x4500. dual Xeon 2640v3 2.6GHZ (16 core). Bloom and glare are quite slow to adjust but that is during rendering and the VFB is trying to process it in real time every time it refreshes so ends up in a cycle of trying to render and process the bloom constantly. My bloom is set at 5 and glare at 1.4 for this scene but I think as long as I remember to switch it off before rendering then just add it at the end then that should be better. Still 26 hours predicted is quite slow for 250 passes I think.

[Jpjapers]

I suppose depending on the complexity of your scene and your material graphs that might be about right. Can you get away with less passes with denoising?