Machines rendering differently?

[SnipeyX]

I'm looking for some troubleshooting advice here, we have three machines that are identical as far as I can tell (windows server 2003 R2 64bit, 4GB ram, intel Xeon dual proc quad core) and one of them renders frames differently than the other two. (sometimes shadows are different, sometimes a certain triangle of a mesh is shaded diffferently, it's always been minor, if noticeable at all on some scenes, but still causes flickering.)

What sort of things should I look at changing? The bad machine is running vray license manager, haven't had an opportunity to move that yet, but I'm not expecting that to fix it.

[werticus]

check that this machine has the same decimal place symbol either , or . as this can cause truncation of render values.

check it can read/ write to the nessasary directories for textures / stored ir maps etc.

ensure you have enough network bandwidth to deal with the demand when a scene loads, you could test this by bringing one render box online at a time. (even DR renders now stagger automatically when a box comes online).

thats all i can think of right now... im sure others will have other things to try.

i had a similar problem and it turned out that my router although rated as a 'gigabit' switch, could not handle the full speed on all ports at the same time and so textures got messed up. You'll want an enterprise level switch imo. But i've got more computers.

[voltron7]

I had a similar problem...it was fixed by re-installing latest max+sp/updates on both machines. Also same up to date vray versions... Now works fine, except only if I have the maps/proxies pathed to and on on the slave machine. Odd because the slave machine has full rights/access to the other. I can copy, edit, delete, create files between them just fine, but at render time, the slave cannot access the maps...maybe it's an xp 32 vs. xp64 issue? Tried it with all the fierwall/anti virus stuff disabled and made no difference.

[DIGITAL_FACTORY]

Now works fine, except only if I have the maps/proxies pathed to and on on the slave machine. Odd because the slave machine has full rights/access to the other. I can copy, edit, delete, ...

Voltron7,

This sounds like a DNS problem within your domain or workgroup...

[dlparisi]

I'm looking for some troubleshooting advice here, we have three machines that are identical as far as I can tell (windows server 2003 R2 64bit, 4GB ram, intel Xeon dual proc quad core) and one of them renders frames differently than the other two. (sometimes shadows are different, sometimes a certain triangle of a mesh is shaded diffferently, it's always been minor, if noticeable at all on some scenes, but still causes flickering.)

What sort of things should I look at changing? The bad machine is running vray license manager, haven't had an opportunity to move that yet, but I'm not expecting that to fix it.

Snipey: Did you ever find a solution for this? I'm experiencing similar quirks...

The attached images of from the benchmark illustrate the problem. All images were done using a save imap and LC and rendered via BB. All machines are winXP x64, max 9 64bit (SP2), Vray 1.5 SP1 (double checked and even reinstalled on the dual opteron). I checked the region setting as well for the ,/. issue and these are all the same.

First image is from a dual quad xeon 5345, second is from a dual opteron 270, third is from a core 2 duo (don't remember the proc #). In order to see the frame differences you have to flip between them but it's pretty obvious in the reflections and the refractions that something is going on.

I've also noticed a really annoying discrepancy between light caches produced on different machines. I understand the samples are random but even a high sampling rate produces really different results. The 4th and 5th images are computed using just light cache as primary and secondary. Fourth image is from a dual quad xeon, 5th is from a dual opteron 270. The strange thing is that all of the machines except the opteron produce almost identical images (like a couple of samples are off here or there-nothing major) but the opteron one is waaayyy off. The light levels are the same but it just looks a lot different.

Any help is much, much appreciated.

[dlparisi]

Small update...

I was able to fix the jittery reflections on the core 2 duo by running the Max sp2 update so it seems like that was the problem. (P.S.: is there any way to check the installed version, sp1-sp2-etc., without starting max. That's a PITA for rendernodes that don't have a license). I'm still having trouble with the Dual Opteron though -mismatched reflections and lightcache. Something just seems off with this machine. It's a relatively new install of x64-about 1.5 months- with a fresh install of max (max is fresh install because I've been experiencing these problems so it's not because of the new install).

Anyone Else having problem on a mixed farm, AMD and Intel?

David

[dlparisi]

Another update...

I ran a series of tests as follows to try and deduce where the problem is occuring:

SCANLINE (converted scene to grey materials and glass with a few omni lights thrown in):
All machines render the frame exactly the same.

MR (same as above but with final gather on - set to "LOW"):
All machines render the frame exactly the same.

VRAY (grey vray material everywhere except for teapot which is the original vray glass and scene to no GI):
The Dual Opteron stil renders out funky (see attched first image). All the other achines render out the exact same frame (second attached image). It appears that the sampling is different on opterons vs. the intels which is leading to the different patterning. I still have no idea what's causing it though.

[vlado]

Looks to me like the normals on the teapot are slightly different; I don't have the benchmark scene here - does this happen if you convert the teapot to an editable mesh?

Best regards,
Vlado

[dlparisi]

Just tried it - same result.

It might not be obvious from the images posted above but it's more than just the reflections that are different. The overall sampling on the lights is also different. To better illustrate it I've lowered all of the vray light subdivs to 1 and lowered the min. samples to 2. The attached images are what I get: the first is from a Dual Quad Xeon 5420, the second is a Dual quad Xeon 5345, the last is from the Dual Opteron 270. The first and second image match exactly ("difference" compared in photoshop) while the last image is radically different. Any ideas?

edit-> I also turned off the laser lights in this set of images just in case that was causing any problems.

[dlparisi]

Looking further into a possible normals problem, I rendered out a normals pass just to make sure everything was matching up. In general, it's the same except for a few discrepancies at the edges of some of the surfaces on the back wall - weird.

[Morne]

so Vlado when did you say you guys going to rewrite DR from scratch?

[dlparisi]

so Vlado when did you say you guys going to rewrite DR from scratch?

This isn't a DR issue, all renders are done via BB.

[vlado]

We don't plan on re-writing the DR anytime soon although there are a few extensions planned in that regards, like support for the incremental add to current map mode for the irradiance map.

I would not be surprised to see different noise pattern on different machines (and even between 32- and 64-bit versions), but the shifting of the refractions is a weird one, which looks to me like different normals. Since the difference looks to be fairly small, it will probably not be visible in a RGB normal pass. I can't say anything more without being able to reproduce this here though.

Best regards,
Vlado

[dlparisi]

OK, this is becoming a real PITA for me. I'm trying to deduce just where the error is occurring so I've run the following tests...

On a spare partition on the Opteron machine, I installed xp x64 then installed 64 bit AND 32bit Max 2008 (I couldn't get max 9 to activate so I was "stuck" moving up to 2008 ). On two of the Xeon renderslaves I similarly installed max 2008 32 and 64bit (other than Max 9 64 bit there's almost nothing else on them). Followed up by installing the appropriate version of VRay 1.5 SP1 on each. So basically I had 3 clean installs of x64 with the 32bit and 64bit versions of Max 2008 and Vray.

Here's what I found out. Rendering out the same scene under 32bit 2008 I get exactly the same image from all machines - no glitches at all. Each machine computed it's own Imap and LC as well. I did this test first and was at least encouraged by the results thinking that there was some problem with my installation of Max 9. Then I moved on to the 64bit version....

Well let's start out by saying they don't match. First off, the light levels in all the images are a little different from the 32bit versions but I wasn't too surprised by that, remembering the warnings about rendering from 32bit and 64bit renderfarms. Secondly, while the two images from the intel machines matched perfectly, the opteron machine is again off - in terms of reflections and lighting quality.

Arrgghh!!!! I just don't get it... Is my opteron machine 'off' somehow (if so why do the 32bit renders match)? Why is it just the 64 bit renders that don't match? Is it the vray 64bit version or is it max 64bit that's messing it up? Has it always been like this? Is it x64 in general that's messing it up. I've only been on 64 bit now for a few months but have just now noticed these problems.

I've attached the scene (max 2008 only) and images below so hopefully someone can compare them to their machines and let me know if they are having similar issues. The first two images are the 32 bit versions (which match) the next three are the 64bit ones (included all three to show that the Xeons match and the Opteron does not).

http://dpict3d.com/vray/VrayRC3_Max8...1-max_2008.zip
http://dpict3d.com/vray/bmark_max200...mx64_XEONS.jpg
http://dpict3d.com/vray/bmark_max200...64_OPTERON.jpg
http://dpict3d.com/vray/bmark_max200...mx64_XEON1.jpg
http://dpict3d.com/vray/bmark_max200...mx64_XEON2.jpg
http://dpict3d.com/vray/bmark_max200...64_OPTERON.jpg

Please help. Much thanks.
David

[jujubee]

Try distributed rendering from a saved LC or IRmap (from a single main machine) and see if that fixes it.

Sometimes on my network not all the computers pick up the texture maps right away and cause shading issues. Also, DR calcing a LC seems to give different results on different machines.