[Maxer]

Ok I decided this morning I was going to try V1.5 out again since NL released the new plug-in for Max. I hadn’t tried cooperative rendering yet and had seen some threads saying that it wasn't working properly so I decided to give it a try. My main workstation is a dual processor quad core Intel system with 4 Gigs of ram and on it I ran one of the tests while I was also using the system for e-mail and Photoshop works so it wasn't 100% dedicated to rendering. Collectively all 8 of it's processors are running at a speed of 21.36 GHz. For cooperative rendering I used 11 dual core machines all Intel with 4 Gigs of ram, collectively the are running at a speed of 52.32 GHz. I let the scene run on both machines for about 1 hour and 18 minutes and according to the speed of my machines I should have gotten roughly 2.5 times the speed out of the cooperative rendering machines when compared to my work station. Here are the results:

1 Dual processor quad core machine: SL 15.26


11 dual core machines: SL 14.68


I don't have to say it but I'm going to, cooperative rendering isn't working correctly. 10 out of the 11 networked machines reached a SL of 12.11 the 11th machine was at 11.71. If I allowed all of these machines to render all night long from this data it appears they wouldn’t reach a higher SL than my single workstation would. Someone from NL please take a look at this and tell me what is going on, how can Cooperative rendering not be reaching a higher SL than my single machine when it has almost 2.5 times the rendering power?

[b-kandor]

I'm having the same problem -

It seems that while coop is working in theory - it's merged output is actually lower than the sl of a single machine. I just finished this test:

4 machines stopped at:

machine 1: 13.4
machine 2: 11.8
machine 3: 11.62
machine 4: 14.17

The merged mxi file shows an sl of 13.47 when I open it. So it seems official - coop merging is broken.

********Update:

If I look at the mxi files left behind in the temp folders of each machine I find something interesting.

Each machine has either one or two mxi files left in the temp directory named 'cooperative.mxi' or 'cooperative_aux.mxi'

Here is a list of what I found with the corresponding sl of each file:

SL coop.mxi coop_aux.mxi
machine 1: 13.40 13.40 n/a
machine 2: 11.88 13.47 11.65
machine 3: 11.65 11.65 n/a
machine 4: 14.17 15.52 13.40

So, it's seems clear that coop is MAYBE working - it's just that the final merged mxi is ending up in one of the machines temp directories, in this case, machine 4 had a coop.mxi with an sl of 15.52 which seems about right for the group effort.



Kandor

[lllab]

i had that too once!
i thought it was my fault, but if i read this now this was exactly as here.

thanks for bringing this up, as coop rendering is important.
cheers
stefan

p.s i remember it did work in the instant preview, the merged image there was much better than the final merged one. could this be because of a faulty setup? or is it more a real merge bug?

[b-kandor]

I updated my post above - short version- the actual merged mxi is not being written to the specified location but instead is ending up in the temp directory of one of the machines.

[Maxer]

How do you know where the final MXI is winding up, if you’re using 10 or 20 computers to do cooperative rendering it's going to be a big pain to try and find it. I've tried loading both the cooperative.mxi and merged.mxi files out of various render nodes and I continually get and error saying that it can't open the file so I'm unable to verify your findings.

NL any comment?

[b-kandor]

Hi Maxer,

Where are you looking? I find them in the %username%/document and settings/local settings/temp folder

But I have none called 'merged' only cooperative.mxi and cooperative_aux.mxi.

They all open for me as well. Looking for the right one is an act of desperation and for sure should be addressed with a patch asap. It's easier for me since I'm dealing with only four systems.

If you hit windows + break on your keyboard to open the system properties, then click the advanced tab, then enviroment variables you can read what folder is assigned to the temp variable - it will probably read as above, but may be different for you.

Kandor

[lllab]

ok, then it basicly works , but it just saves it into wrong directory.

so this can be solved quite easily by a smalll update i guess.

please NL for 1.5 sp1....
thanks,
cheers
stefan

[Maxer]

I know where the files are but I'm unable to open them, when I try and load them I get the message

"//Boxx-800066/C/Program Files (x86)/Next Limit/Maxwell/cooperative.mxi cannot be read"

Are you sure it's working because in my test I did get a successful cooperative merger and the SL was higher than any single machine was able to reach on it's own it's just that my 22 CPU's were producing a lower SL than my single machine with 8 CPU's. This seems to be different from what you described b-kandor. And let me stress that without cooperative rendering Maxwell is absolutely and completely useless to me. In fact the reason I hadn’t switched to 1.5 yet was because I had an issue with it not wanting to render animations, that turned out to be a naming problem but still I had to revert back to 1.1 which takes quite a bit of time when you dealing with over 100 machines!!!

[b-kandor]

I'm sure the coop.mxi is not in the maxwell directory. It's in the %username%/document and settings/local settings/temp folder. Just have a look there to make sure.

Otherwise I'm clueless as to a) why they don't open and b) why we are having different results?

[Maxer]

Ok I was looking in the wrong place, yes I did find them and I went through all 11 machines and looked at each file. The biggest SL I found was 14.68 which is what the final SL for the cooperative merger was. It's apparent that the final SL is being written to the network drive I've specified so at least for me that is working properly. The issue is that final SL isn't as high as it should be for the number of machines used.

[b-kandor]

Well that's weird - what sl was reached by the 11 machines?

In my case the merged mxi which was written to the location I specified had a very low sl compared to highest 'temp' file I found.

So, now what?

[Maxer]

In every case except one all the MXI's were lower than the final, one machine had the final MXI on it and that was it.

[Peter Cernak]

hi,

glad you are bringing this up (even if it's in the wrong section - i've been impatiently waiting for a response in the network section), because this problem is a big threat to my deadlines.

As i have found out the cause may be in writing two final mxis by two different nodes on one job (one overwrites the other), which is definitely weird.

see this:

http://www.maxwellrender.com/forum/view ... hp?t=23846

or this:

http://www.maxwellrender.com/forum/view ... 7&start=15

in the network section

[Maxer]

I'm shocked that NL hasn’t done anything about this yet since there obviously aware of the problem, or are they trying to say its user error?

Can someone from NL at least confirm if what we’re seeing is a bug or not?

[b-kandor]

Yes, I would appreciate confirmation as well, I have a job to run over the weekend and I'm not sure how to run it.

I guess I should do manual merging - it seems safest.