[Frances]

As in the first levels a big part of the screen is covered with black spots, and that for some levels, the image is changing rapidly, I would tend to think that there is a minimum sl limit for the merging to work...

The algorithm used by cooperative rendering would have to be different from that used by unconnected machines working parallel. Each cpu would be given certain pixels to work on. If it's not done that way, it certainly could be. It's only number crunching. It's like bucket or strip rendering, only at the pixel level.

[leoA4D]

Bottom line? NL has put you on notice to start saving for three quads or five duals or nine singles. If you are going to play in the professional league, this is the basic spec for MW. As time goes on, MW will bloat like apps normally do.

Hey EBIII, it sounds like a good time for a Kool Ade, eh?

[morbid angel]

I think the email can bounce on some accounts because of the image sizes that they have attached...

But honestly....whats the big deal, cooperative rendering is great and all, but 1 - you need license, 2 - you need nodes. Not to mention fantastic nl policy per cpu.
Bah this is redicules.

[neil hayes]

Sorry but I don't get it... in all honesty.

Well I could just as easily ask, "Whats the point in threads like this?" NextLimit are communicating which is a good thing, lets not discourage it. I welcomed the email, it suggests that co-op rendering is working, which is good news.

-neil

[pluMmet]

The algorithm used by cooperative rendering would have to be different from that used by unconnected machines working parallel. Each cpu would be given certain pixels to work on. If it's not done that way, it certainly could be. It's only number crunching. It's like bucket or strip rendering, only at the pixel level.

I agree...but this system clearly is not that as seen by the merging of images. If the pixel level bucket rendering you're talking about was in effect then there would be no merging.

We will see when this system becomes available but I'm very confident about the cpu/level sweet spot

[leoA4D]

Well I could just as easily ask, "Whats the point in threads like this?" NextLimit are communicating which is a good thing, lets not discourage it. I welcomed the email, it suggests that co-op rendering is working, which is good news.-neil

IMHO, what is being said here is:
1) Many of the folks here are rabid MW supporters but are small businesses/studios with limited hardware capability and limited funds. They bought into MW thinking they could stay small and do professional level renderings. NL is telling us optomization will be via 9 +/- nodes.
2) NL is telling us for the first time that it is our responsibility to reduce the time by buying more nodes and licenses because they do not see a software solution to reduce the time. It is the cul de sac again. Others have said it could not be done and maybe NL is agreeing with them.

[kingpin]

I've got the mail today... and here is my thoughts.

The mail really didn't answer or show the "almost-linear" improvements in speed with cooperative rendering. The sample scene is... a good/bad example to me(good for NextLimit to show but bad for user to judge).

For me, the scene was something that it really shouldn't require that much time to render. What I am interested in is... will it improves render speed (almost linearly) in scene where a lot of metallic reflections, caustics and out-of-focus areas... where picture requires very high sampling level? ... something like glebe digital has posted in "60's Honda" thread, sandykoufax's Subway in Pusan or jakegoat&Andronikos's candle scene.

And right now(even after seeing the mail), I don't know if the cooperative rendering is a real thing(ie works for all possible situation).

PS. Oh, and they should have posted the pict at 2.5 hours with 2 CPU so we can directly see the improvements with additional CPUs.

PS2. Then again, as if I can afford additional....

[neil hayes]

Well I could just as easily ask, "Whats the point in threads like this?" NextLimit are communicating which is a good thing, lets not discourage it. I welcomed the email, it suggests that co-op rendering is working, which is good news.-neil

IMHO, what is being said here is:
1) Many of the folks here are rabid MW supporters but are small businesses/studios with limited hardware capability and limited funds. They bought into MW thinking they could stay small and do professional level renderings. NL is telling us optomization will be via 9 +/- nodes.
2) NL is telling us for the first time that it is our responsibility to reduce the time by buying more nodes and licenses because they do not see a software solution to reduce the time. It is the cul de sac again. Others have said it could not be done and maybe NL is agreeing with them.

We are all aware of the speed/accuracy trade off than faces all renderers and NL are not forcing you to go out and buy licences. In terms of individual freelancers/small studios, as we're all embrancing the delights of multicore processing even with 2 dual core machines render times will be somewhat more palatable with co-op working.

-neil

[Kabe]

how would the program tell the data from the noise?

simple averaging.

The noise in the result get's decreased because it's normally distributed.

If you average noise into an image you are lowering the quality of the 'good' (non-noise, actual data) pixels. You are not refining the image, you are degrading it.

Sorry to correct you, but this is not true!

If you have random noise due to the sampling process, be it a CCD chip or
with M~R, then the "good" pixels are exactly the average of your "noise".

Every digicam samples the noise of the CCD chip for quite some time to
get the final image, and so does M~R, too. To achieve the "spectral" solution,
it has to shoot quite a couple of rays in different energy levels. This causes
the noise, because we simply have a sampling process here.

The average of the noise *is* the real picture so to speak.

Hope it helps to understand it

Kabe

[paxreid]

do you think the 'carpet' on the picture(s) is made with displacement ?

Hahaha...slow down Dennis

Haven't even seen normal maps work

[GuyR]

Just got the email.

The one positive is the image seems very noise free behind the glass windows?

Guy

[paxreid]

Just got the email.

The one positive is the image seems very noise free behind the glass windows?

Guy

Guy...don't assume there is glass there...I would guess no.

[Frances]

I have to say that image included in the email is one of the uglier renders I've seen done by Maxwell. It's dark and depressing. Any area of light or highlight is heavily aliased. Good to know we can have the sublimely mediocre in less than half the time for more than twice the cost.

[leoA4D]

We are all aware of the speed/accuracy trade off than faces all renderers and NL are not forcing you to go out and buy licences. In terms of individual freelancers/small studios, as we're all embrancing the delights of multicore processing even with 2 dual core machines render times will be somewhat more palatable with co-op working.

-neil

It is about expectations/sales hype before buying MW versus what we received to date and are being told now. It would be interesting to poll maxwellians to learn if they plan to upgrade hardware and to what number of CPUs they will buy to use MW in production. I am grappling with hardware upgrade now. And hey, if RC6 shows enough promise, I may buy another license, too. If I buy it within three weeks, it will be one year since I bought my first license.

[johann.dugge]

ernest I think you're not right. Regarding the average noise and your little example, noise is random but converges towards the correct solution. Maxwell will never "finish" an image, and basically, even if you cannot see it anymore, there is always some noise left (of very small amplitude so it is practically the correct solution).
Now when you got three renderings, with a different seed, the noise amplitude (which is so disturbing when it's high) will be the same in all. But they are not the same:

0.8 | 1.1 | 1 (notice that the last value "1" may be purely coincedental)

here the amplitude would be .2, and the average would be (.8+1.1+1)/3=.97
if you take more renderings into this, say another one with value 1.2 the result would be 1.025. I hope you can see what I'm trying to explain.
And actually, to clarify, the sampling level that we're given by maxwell is the amplitude of the noise, not as a linear value (as .2 would be), but something like e^(-SL).

Hope this was not too mathematical / explained it with a bit more reasoining than just some "feeling".