[gianca]

I think I came up with a simple way to understand how to manage render nodes to get a target SL: every node on the farm on a cooperative render adds +1 SL to the merged final SL, as long as every node contribute the same amount of SL.
So for instance if I want to reach SL 15 and I got 5 nodes o the farm I can assume each node has to reach SL 10 to obtain the desired final SL result.
That also means that if I got 15 nodes I only need to reach SL 1 on each node to reach a final SL of 15. I can dream, can I?

In top of that I'm also assuming that which each additional SL we get half of the noise: I observed that by comparing a merged mxi with one that I merged in Photoshop, and results seems to confirm this assumption.

I'm posting here to hear your thoughts, confirmation of my observation or rebuttal. Either way is welcome.

Cheers,
Gianca

[gianca]

Hearing crickets here... I was hoping to be either endorsed or rebutted, but I got neither...
Maybe it's such an obvious thing?
(wasn't to me)
Are you big guns caring to comment?

Do not want to spread misinformation...

Cheers,
G

[Fernando Tella]

I think you can only make time calculations. If you have 5 nodes of the same power you can say it's going to take 5 times less to get to the same SL than with just one node. The relation between noise level and SL depends greatly on the scene and materials involved. Some scenes will be clean at SL 14 and other won't be at SL 25...

[gianca]

I think you can only make time calculations. If you have 5 nodes of the same power you can say it's going to take 5 times less to get to the same SL than with just one node

Interesting... I didn't felt the time calculation was that linear. I can never predict how long it takes from a SL to another as it varies greatly from scene to scene, so wasn't sure ie 5 nodes = 1/5 of the time (I know it should).
My observation came from the way deadline calculates how to distributes cooperative renders to reach a desired SL, although it doesn't seem to be pete gently accurate.
Need to do some more testing when I get back from my current travel.

The relation between noise level and SL depends greatly on the scene and materials involved. Some scenes will be clean at SL 14 and other won't be at SL 25...

I was actually purposely trying to be "clean noise" agnostic on this topic: I know every scene is different, I'm trying to determine an easy way to calculate how many nodes vs render time it takes to go from an arbitrary SL to another regardless of scene or amount if noise.

G

[Fernando Tella]

I think you'll enjoy the render calculator which is in tools menu in Maxwell Render window. Based on a scene data or mxi it can guess SL reached within an amount of time or time needed to get to an specific SL.

[egmehl]

I was looking through the plugin scripts for Deadline render manager a while ago and I came across this line:

Code: Select all

- LOCAL is the Sampling Level we need to pass for each co-op job
- SL_MERGED is the desired Sampling Level after merging
- NODES is the number of co-op jobs
double LOCAL = ( log( ( (exp( SL_MERGED * log (1.5) ) - 1.0 ) / (double)nNodes ) + 1.0 )) / log(1.5);

The note in the source file says it comes from Next Limit, but not sure if it still applies, or is really official, etc. So I guess use at your own risk!? In my experience though it seems to come pretty close.

[Fernando Tella]

I think you can only make time calculations.

Lately, every time I open my mouth somebody comes to prove me wrong. I meant you cannot make easy finger calculations about SL, obviously those can be done as Maxwell shows them on screen...

Good finding Eric!

[gianca]

I was looking through the plugin scripts for Deadline render manager a while ago and I came across this line:

Code: Select all

- LOCAL is the Sampling Level we need to pass for each co-op job
- SL_MERGED is the desired Sampling Level after merging
- NODES is the number of co-op jobs
double LOCAL = ( log( ( (exp( SL_MERGED * log (1.5) ) - 1.0 ) / (double)nNodes ) + 1.0 )) / log(1.5);

The note in the source file says it comes from Next Limit, but not sure if it still applies, or is really official, etc. So I guess use at your own risk!? In my experience though it seems to come pretty close.

Great find, that's what I was looking for, as I found too that while deadline cannot get an exact SL result it does get pretty close.
So based on the deadline script my own math was oversimplified.
Or is it?

G

[egmehl]

Haha, to answer your last question gianca, you're going to have to find someone with a better memory of their high school math than me