question regarding the lightcache computing using DR...

[n6]

Hi,
maybe this question has been allready answered but I cant find the answer ...

When I use DR, I do it on 2 PCs.
My main PC is a dual Xeon @ 3.6Ghz with 2Gb RAM.
The second one is a single P4 3Ghz with only 1Gb.
My lighting method is allways IR map + Lightcache.
You can easily anderstand that when I launch DR, my main PC (Xeon) ends the lightcache computation much more faster than the second PC and begins the IR map computing as the second one is still computing the lightcache.

I would like that ONLY my main PC compute the lightcache, and only when it would be done, the second one would start working on the IR map.

Hope you anderstand me

Is it possible?

[RErender]

Just render and save a LC on your main computer then render the rest in DR using the saved LC?

[n6]

Just render and save a LC on your main computer then render the rest in DR using the saved LC?

yes thats a solution but if this is not a problem to do that once before launching the final rendering, when you are trying to tweak your lighting, materials etc... you will have to do that 300 time per day.... an huge waste of time ...

Would be nice to do that automaticaly.

[RErender]

Yep, that's why I put the question mark. I wasn't sure if it was a workflow issue or a wishlist thing, I think I remember Vlado saying he would look at it.

[Clifton Santiago]

It seems to me that when doing IR/LC on DR, that the LC map is calculated only locally, and the distribution does not occur until the IR calcs start.

[n6]

It seems to me that when doing IR/LC on DR, that the LC map is calculated only locally, and the distribution does not occur until the IR calcs start.

just test it, you will see that all PC's have 100% CPU use during the LC computing.

[Clifton Santiago]

Okay, I hadn't noticed that. However, when I do DR on a file from different machines, I get different LC calc times.

My fastest machine is 3x faster than my slowest. When I do DR on the fastest machine, the LC is 3x faster than the same on the slowest. This is as one would expect if the LC were rendered locally first.

Vlado should know. Vlado?

[RErender]

Each machine renders it's own LC

[n6]

Okay, I hadn't noticed that. However, when I do DR on a file from different machines, I get different LC calc times.

My fastest machine is 3x faster than my slowest. When I do DR on the fastest machine, the LC is 3x faster than the same on the slowest. This is as one would expect if the LC were rendered locally first.

Vlado should know. Vlado?

yes, but as your fatest machine starts the IR, your other PC are still computing the lightcache. And so they would start to help the IR computing much later that your fatest did.

[Clifton Santiago]

I have 4 dedicated render nodes set up for DR. When do DR on a file, using my fast workstation and those render nodes, the LC takes 5 minutes to calculate. If I open the same file on an old workstation (3x slower than fast workstation), and do DR using the same 4 rendernodes, the LC takes 15 minutes. How is that possible if all the rendernodes are working on the LC calcs? The CPU usage on the render nodes is at 100%, but they don't actually seem to be contributing to the LC calcs.

[ene.xis]

if i recall it right they re all calculating the LC solution untill one finishes...when the first one is done with it it sends the data to the others and those just ditch their LC...
but i m not sure imo

[Dynedain]

Currently each machine calculates LC independently during DR.

http://www.chaosgroup.com/forum/phpB...ic.php?t=19275

[ene.xis]

yes, but they use one LC data? i mean they each doesn t use a diferent LC, right?

[Dynedain]

yes, but they use one LC data? i mean they each doesn t use a diferent LC, right?

Nope. Each machine uses the LC that was independently generated on that machine. That's why it can take so long for render nodes to show up in the next step (IRR or final image), especially if they are slower machines. They're still doing their own LC calcs.

My wishlist item that I linked to was to deploy the LC solution to the rendernodes when it is completed, which IRRmap already does.

[Clifton Santiago]

So if each machine is rendering their own LC map, but only one LC map is used for the rendering, how is the decision made as to which LC map to use?