[Mihai]

The mistake in your calculation is that you say it took 10 hours to reach SL 15, but then suddenly it will take 20 hours just to go from 15 to 16.

14 to 15 didn't take 10 hours, those 10 hours you counted are ALL the SL, from 1 to 15.

Perhaps from 14 to 15 it took 2 hours, and even so usually the time doesn't necessarily double, so from 15 to 16 you would perhaps need 3 hours.

[Ernesto]

I do not know it that is correct, I would like you to tell me which is the relationship between time and sampling levels in each step.

Ernesto

[Ernesto]

The point is the following:

Exponential relationships, or geometrical relationships are tricky.
They are much more dangerous that we use to think.
This is my warning!

Of course I would like a mathematic calculation from NL.

Ernesto

[Mihai]

From what I've noticed it depends on the scene.....most often I haven't seen a doubling of time from one SL to the next.

[Ernesto]

Ok, for You that want Real Numbers

I have made some tests and took note of the time needed to increase one SL. My calculations says that to increase quality to the next SL you will need 1.5 of the time needed to reach the previous SL. This is so predictable that Maxwell tells you the exact moment in which the next SL will be reached, and this prediction is perfect.
This means that it is not an estimation, Maxwell really knows how long it needs to render to reach to certain sample level.

This is not Scene Dependant, this seems to be a constant. What is not a constant is the acceptability of certain SL. It could be great for certain scene a SL of 10, but others may need 22.

The other variable is the power of your processor. Yes it could be faster or slower deppending on the kind of processor.

But what I have just discovered is that One Valuable information that Maxwell Render knows, and is not telling us, this is: How long you need to render to reach to SL 20 for example.
If I would have this info, I could be on time to change the resolution, or telling the client that it is not possible.
I hope NL team will add this indicator in the Maxwell Render Progress window.

Let,s suppose that we reached to SL 1 in 150 seconds
Then, all the following steps are perfectly calculables.
next SL 2 will be reached at 375 seconds
SL 3 will be reached at 712 seconds
SL 4 will be reached at 1217 seconds
SL 5 will be reached at 1975 seconds
SL 6 will be reached at 3112 seconds
SL 7 will be reached at 4818 seconds
SL 8 will be reached at 7377 seconds
SL 9 will be reached at 11215 seconds
SL 10 will be reached at 16972 seconds
SL 11 will be reached at 25608 seconds
SL 12 will be reached at 38563 seconds
SL 13 will be reached at 57996 seconds
SL 13 will be reached at 87145 seconds
SL 14 will be reached at 130869 seconds
SL 15 will be reached at 196455 seconds
SL 16 will be reached at 294835 seconds
SL 17 will be reached at 442405 seconds
SL 18 will be reached at 663760 seconds
SL 19 will be reached at 995792 seconds
SL 20 will be reached at 1493840 seconds (414 Hours)

Now let´s relate this to the amount of machines needed to increase 1 SL:
As for the previous calculation we would need to double the processing power to increase 2 SL

So the same render could be done by 2 machines in 207 hours
In the same way 4 machines could do it in 103 hours
then 8 machines will do it in 51 hours
16 machines will take 26 hours
32 machines will do it in 13 hours

You see: my cals were not so bad.
I knew it although I was totally wrong with the times, but I knew that that error would be insignificant in relationship to the Exponential component of the calculation.

Now I would like to ask NL to add more information in the progress rendering window:
i.e. if we ask for a desired SL Maxwell could tell us how long we have to wait for it.
Knowing the sad answer we will be on time to change resolution, or just tell the client to give us more time!

Ernesto

[Maxer]

Ernesto,
I wish you would have seen my previous posts, I actually discovered this weeks ago but my calculations weren't as elegant as yours. I think this partially explains the free licenses we will be getting.

Go about halfway down the first page.
http://www.maxwellrender.com/forum/view ... +rendering

[Hybaj]

So what you guys are basicaly saying is that when you have a picture that's noisy even on higher SL's, you just won't get it clear with any reasonable computer power?? Correct??

[Ernesto]

So what you guys are basicaly saying is that when you have a picture that's noisy even on higher SL's, you just won't get it clear with any reasonable computer power?? Correct??

Sorry Maxer, It seems that we were thinking the same way, through independent ways!

Hybaj, what I want to say, is that what looks close may be impossible.
Maxwell says something like this: "it will ever reach to the right solution, with enough procesing time" The problem is that "enough processing time" could be more than your lifetime!
If we think in Cooperative rendering, we could say: "You can ever reach to the righ solution in your desired time, deppending on how many machines you have" The problem is that you may need 1000.000 machines or more!
Both phrases are true, but tricky, as soon as you are fully concious of the exponential component in the calculation, we can understand that we cannot live forever to see the final image, nor can invest millions in licencing and equipment.

It is as tricky as this problem:

Let's suppose you are in a race, 2 meters from the finish line.
You are allowed to make a first 1 m step towards your goal, but the following step must be 1/2 of the length of the previous step, and so.
How long would it take to reach to the goal?

You see!

Ernesto

[Mihai]

The results I have after the render:
In the maxwell folder of every server clients I have a cooperative.mxi.
In the maxwell folder of the manager I have a cooperative.mxi and a merge.mxi.
In the network folder I have a *.tga and a *.mxi file.

The tga and the mxi file in the network folder are not as good as the four cooperative.mxi merged manually.

And what about the merge.mxi from the Maxwell folder of the manager? What does it look like?

[b-kandor]

.....mxi file in the network folder are not as good as the four cooperative.mxi merged manually.

Hi Fritz, when you load the job into the viewer how are you specifying the path to save the merged mxi? I use the format \\machine_name/myfolder/myfile.mxi and it works well for me.

Although if your nodes are reporting 'success' that indicates that the merging has occured.... hmmmm..


Kandor

[Mihai]

Yes, in the network folder, I was wondering what the merge.mxi from the maxwell folder of the manager looks like.

[Ernesto]

Ernesto,
I wish you would have seen my previous posts, I actually discovered this weeks ago but my calculations weren't as elegant as yours. I think this partially explains the free licenses we will be getting.

Go about halfway down the first page.
http://www.maxwellrender.com/forum/view ... +rendering

Maxer:

I have just read your post. I agree 100% with your point. We are talking about the same thing. I was insisting because I know that these mathematic problems are hard to understand in an intuitive way. Most people do not expects it, because the problem seems so easy.

Regarding the licences... al least the pricelist should be proportional to the amount of help in the Cooperative rendering process:
This mean, that if the first licence is $1000, the second should be $500, the third $250, the 4th $125 the 5th $62 the 6st $31, the 7th $15, the 8th $7.5, the 9th $3.7- the 10th $1.8- the 11th $0.90 the 12nd $0.45, the 13th $0.22
the 14th 0.11 the 15th $0.05 and so....

Hahaha!

Ernesto

[Ernesto]

What I propose, as a solution, is to change the information in the dislay window in Maxwell Render.

The user could fill the resolution box, ie: 2048 x 2048
Then there will be a Desired SL: ie: 22
And Maxwell will tell the calculated time for them: xxxxxx Hs

In that way we could accept and send to process, or change any parameter.

This will avoid false expectatives, after 20 or 40 processing hours.
Does it sounds reasonable?

Ernesto

[Ernesto]

I was expecting to use Maxwell for Architectural use, and feelt bad because I couldn't get to a decent SL in order to get rid of the noise...
I thought that Cooperative Rendering was the awaited solution.
Now I understand that Cooperative Rendering will not help me.
Perhaps maxwell will be limted to low resolution images (in the Architectural market) or to (simplier) microscopic or jewelery scenes.
I work in low res at 2048 x 1536 , and final presentation images at 4096 x 3072. But could never do it in Maxwell with my modest equipment.

My only hope now, is that NL could invent a way to make it faster.

Perhaps a filter that can fill the noise, made of black dots, by interpolating the clear dots.

Perhaps a way to increase the size of light dots, so that there will not be any black background.

Perhaps if we could control the width of lightrays in a way the processed dots could be 30 cm in diameter, so that the image will appear faster...

Anything have to be done... since like it is right now is severely limited.

Sincerely

Ernesto

[b-kandor]

Ernesto,

It's true that MR need longer rendertimes per sl. It's not a failure of co-op rendering. If I run my 4 cores in co-op for 10 hours I am still achieving an image equal to running my 1 core for approx. 40 hours so I have saved 30 hours. Whatever sl level I've reached isn't really the point for me. As I see it I can get farther (whatever sl that means) in less time and that's a good thing!

Kandor