Maxwell Render | Official public forum

If I'm running a render to, say, SL 16 and wanted to run a second render on a different machine that I could later merge in to increase the SL of the final image to a higher one, say SL17, how do you calculate the required SL for the second render?

This comes up for me when I run a coooperative render and find that the final result is still not high enough SL. I have tried resuming the merged result of the co-operative render, but it created some weird artifacts in the file and trashed the MXI. From now on I think I'm better off to just render another pass and merge, but need to figure out how to hit my target SL this way.

Is there a simple way to calculate this?

Thanks in advance
/b

I use as a general guideline: each new S.L. requires 1.5x the amount of time it took to reach the current S.L.

So if a render reached S.L. 16 in 30 minutes, S.L. 17 will take an additional 45 minutes.

Someone correct me if I'm wrong on this. I haven't tested this for a few versions now.

Thanks Brian. Not sure if that holds - always seemed more exponential to me - but what I'm hoping to find is what SL to set to, since the time is not directly relevant for me in this case (it would be a result of a co-op render, so the time to reach SL16 would not be meaningful for the single node render)

Is there a SL based formula of some kind?

b

simmsimaging wrote:Thanks Brian. Not sure if that holds - always seemed more exponential to me - but what I'm hoping to find is what SL to set to, since the time is not directly relevant for me in this case (it would be a result of a co-op render, so the time to reach SL16 would not be meaningful for the single node render)

Is there a SL based formula of some kind?

b

Brian's formula actually is exponential: t = a * 1.5 ^ (x - 1) where a is time to reach SL1 and x is a desired SL. Although there is no single formula as there are too many factors that influence the calculation.

ababak wrote:

simmsimaging wrote:Thanks Brian. Not sure if that holds - always seemed more exponential to me - but what I'm hoping to find is what SL to set to, since the time is not directly relevant for me in this case (it would be a result of a co-op render, so the time to reach SL16 would not be meaningful for the single node render)

Is there a SL based formula of some kind?

b

Brian's formula actually is exponential: t = a * 1.5 ^ (x - 1) where a is time to reach SL1 and x is a desired SL. Although there is no single formula as there are too many factors that influence the calculation.

Sorry, should have been more precise in my wording: I just meant that it seemed more like the time was doubled or more each SL, but thanks for clarifying

So what you are saying though, is that you cannot know in advance for sure what SL I would need to merge into a pre-existing render to raise it by 1 SL, or just that it's not possible to figure out how long it would take based on simple calculations?

simmsimaging wrote:So what you are saying though, is that you cannot know in advance for sure what SL I would need to merge into a pre-existing render to raise it by 1 SL, or just that it's not possible to figure out how long it would take based on simple calculations?

In fact, SL is just some arbitrary number, one render may look fine at 12 SL, one may be noisy at 24. When you merge, you make the final image better, you don't need to match some precise SL number. Roughly, rendering say an hour on a second computer and then merging two images is like rendering an hour more on a single computer — that's my observations.

I've done tests on this recently myself and I can confirm Maxwell SL timing is exponential. In my test I got an R^2 value of 0.9912, which got closer to 1 over time. I'm not excellent at maths but this was my workflow:
I rendered up to SL16 and noted down the time in seconds required to reach each SL, then plotted the numbers and divided the total time taken into equal sections to calculate the required SL to reach on each machine. Each machine should take roughly the same amount of time to render, assuming:
the machines are the same spec
when Maxwell is rendering, the machines are used the same amount

2 machines to SL 16: Approx SL14
3 machines to SL 16: Approx SL13.2
4 machines to SL16: Approx SL12.5
5 machines to SL16: Approx SL12

It's worth noting that this was with low priority mode.

In the near future I'll probably do a test to a higher SL, 20 or something. In theory it doesn't require a long scene to work out the times, but the more time the incremental updates take up the less accurate the results will be.

Thanks for taking the time to help Andrey.

I understand your point that SL is somewhat arbitrary, but for any given image it can still represent a meaningful difference from one SL to the next, and I feel like it is more useful as a measure than one hour, or two hours etc. An hour could make a huge difference, or none at all depending on the res and benchmark etc.

I would still like to know if I can figure out what SL mxi I need to merge into another to raise it by one SL.

/b

JamesColeman wrote:I've done tests on this recently myself and I can confirm Maxwell SL timing is exponential. In my test I got an R^2 value of 0.9912, which got closer to 1 over time. I'm not excellent at maths but this was my workflow:
I rendered up to SL16 and noted down the time in seconds required to reach each SL, then plotted the numbers and divided the total time taken into equal sections to calculate the required SL to reach on each machine. Each machine should take roughly the same amount of time to render, assuming:
the machines are the same spec
when Maxwell is rendering, the machines are used the same amount

2 machines to SL 16: Approx SL14
3 machines to SL 16: Approx SL13.2
4 machines to SL16: Approx SL12.5
5 machines to SL16: Approx SL12

Thanks James -
I'm probably just dense, but I'm not really sure how to use this info? How would you project it forward to answer the root question?

simmsimaging wrote: Thanks James -
I'm probably just dense, but I'm not really sure how to use this info? How would you project it forward to answer the root question?

I don't think I can answer the root question properly, it's plenty of maths which I don't fully understand myself but in my experience this applies:

To increase an SL by 2 requires double the total render time.

So for this problem it's easier to think in multiples of 2. E.g. if you have an SL16 image that needs to go to SL18, you need to render a new .mxi up to SL16, then merge and it should be approx SL18.

There should be a similar way to calculate an increase of 1 SL, possibly the current SL minus 2 e.g. SL16 + SL14 = SL17, only tests will tell.

JamesColeman wrote: ...

There should be a similar way to calculate an increase of 1 SL, possibly the current SL minus 2 e.g. SL16 + SL14 = SL17, only tests will tell.

Or NL could tell tell.... Was it too much to hope to save a lot of hours of testing to answer something I thought they probably already knew? Thanks for your help with it in any case.

For the record, just done some quick tests:

SL10+SL8.5=approxSL11 (actually SL11.04)

So I think this should apply to all SLs i.e. SLx + SL(x-1.5) = SL(x+1)

That's great - thanks James! When my current batch of renders are done I will give it test as well.

Thanks for all the assistance with it.

b