Multiscatter Xeon Issues

Post Reply
figment
Posts: 3
Joined: Sat Mar 14, 2015 9:50 pm

Multiscatter Xeon Issues

Post by figment »

Hi all,

We are experiencing some unusual problems with Multiscatter. We are currently attempting to render a scene on our render farm which has multiple Multiscatter nodes in the scene to create grass, bushes and trees etc.

Our render farm has a combination of 6-core Sandy Bridge and Ivy Bridge Intel i7 and dual 12-core Xeon E5 V3 processors with between 24-64 BG RAM. We are running Windows 8.1 64-bit, 3Ds Max 2015 with EXT1, VRay 3.00.07 and Multiscatter 1.3.2.3 on all machines. These don't look like issues being caused by conflicting software versions.

Two odd things are happening:
1. The Xeon machines (Dual 12-core Xeon E5 V3s with 64GB RAM) take almost twice as long to render the scene as the single 6-core i7 machines. The Xeon machines are 3x more powerful than the i7 machines so this is very odd. We've tried rendering the same scene without Multiscatter and the frames render as fast as we'd expect, so it's definitely an issue with Multiscatter. We've also tried different GI options and Light Cache seems especially slow.

2. The positions of some blades of grass shifts on the Xeon machines but not on the i7 machines What is really odd here is that 90% of the blades of grass are in the same place on all frames, but a few seem to move to a different position on all the Xeon machine's renders. All machines have access to the distribution maps, texture maps, Vray Proxies etc. so it is 100% not a missing assets issue.

There seems to be a huge discrepancy between what we're seeing on the Xeon machines and the i7s. We are gradually switching our rendering to Xeon based machines so if Multiscatter is going to perform this badly then we might have to look else ware for this kind of plugin unless it can be resolved.

If you are able to help with these issues that would be appreciated.

Many thanks

Ollie
anvar
Posts: 578
Joined: Sun Jul 11, 2010 12:34 pm

Re: Multiscatter Xeon Issues

Post by anvar »

MultiScatter (or any other scattering plug-ins) is only creating lots of geometries just before rendering
so it doesn't really care what proccesor you have. and not involved in actual rendering process.

from the rendering point of view than more cores you have than more memory you may need
e.g 6 cores with 24gb will have 4gb per core

so 24 cores with 64 gb will have 2.6gb per core
and the rendering may slow down (depending on how much geometries you generated)

and especially with Light Cache you will need tons of memory
figment
Posts: 3
Joined: Sat Mar 14, 2015 9:50 pm

Re: Multiscatter Xeon Issues

Post by figment »

Thanks for the response Anvar.

The memory issue could explain the issue we're having with speed, but it doesn't explain why we're getting objects appearing in different places. Could that somehow be related to not having sufficient memory for the scene?
anvar
Posts: 578
Joined: Sun Jul 11, 2010 12:34 pm

Re: Multiscatter Xeon Issues

Post by anvar »

please send us some screenshots
may be some glitches with LC or irradiance map
figment
Posts: 3
Joined: Sat Mar 14, 2015 9:50 pm

Re: Multiscatter Xeon Issues

Post by figment »

I think I've found the cause of the problem and it seems to be completely unrelated to Multiscatter itself!

I haven't had a chance to test this with Multiscatter yet but we noticed that we were getting less than expected performance from our dual-Xeon machines in general. I looked into this some more and it seems as though 'Bitmap Paging' in 3Ds Max was the issue. This feature is turned on by default in 3Ds Max and turning if off has sped up rendering by about 5x!

For those who don't know; Bitmap Paging is 3Ds Max trying to be helpful by caching large bitmaps to the HDD in order to save RAM. This might not have a big impact on single core machines but has a huge performance hit on dual-CPU systems. You can turn this setting off under the Common tab of Render Settings.

As I said; I haven't specifically tried this with the scene in which we were experiencing problems with Multiscatter but I'm confident that that was the culprit. We didn't ever seem to be maxing out the RAM in the scene in question but we were using fairly high resolution maps for distributing grass, trees etc. If those maps were being cached rather than loaded into RAM then that would explain the slowdown.

We're quite busy at the moment but if I get the chance to re-render that scene I will report back to confirm that this was the problem.
Post Reply