[Maxer]

When you say dice the image into n-pieces is this done by creating multiple cameras with very slim image aspect ratios?

[sidenimjay]

i cant speak for any other software than houdini.

however in houdini , you can do crop region rendering . . . this is not simply a houdini benefit it is actually a maxwell function . . .cant say if all plugins have implemented it, or ease of use . .

the crop region in houdini is a setting on the camera

in houdini , crop goes 0-1 left to right top to bottom . . . .so setting this to a value:

crop_right 0
crop_left .5
crop_top 0
crop_bottom .5

should yield, given a 360x243, the top right corner to halfway across, 180 pixels, and bring the bottom up to 122 pixels. the resulting render is still 360x243, but only the upper right quadrant is rendered, the rest is black. this is great because it allows a straight over operation in the composite, without having to do any math to figure which crop goes where

then you assign a name : myrender_imagetile_NTILENUM.frame.ext
and composite them back together . . .

this is very easy to script in houdini, such that for each tile the offset on the crop region is adjusted and the rendername is set appropriately then the mxs is written.

[Mihai]

There is also the bitmap render region feature in the Max plugin, but keep in mind it doesn't work for caustics. The whole images calculations must be taken into account for caustics so splitting up an image no matter what method you use would actually take as long to get clean as if you had done the entire render.

[Maxer]

Mihai, where is this bitmap render region feature?

[Neil Evans]

So what your saying is that NL should not give us 4 Procs per lic but 10!! and maybe chuck in a couple of rendernodes as well.

I am speaking for myself here but I could not afford to buy 10 licences just to be able to do one 60 hr render in 6hrs. I don't think the quality is that amazing, I would perfer to use Mental ray or Turtle and get the job out.

[pluMmet]

I also think that M~R is ahead of it's time. It makes me happy and sad at the same time. I hoping they will make M~R work with clearspeed Then 2 nodes should be enough for a small outfit like me

You've got 3.....I've got 30!

Hey Maxer, How do you fit 30 licenses on a:
Opteron 248
3 Gigs Ram
7200 rpm SATA Drive
Invida Quadro 1100

???

[Mihai]

It's in the system rollout, right above the mxcl command line options box.

You just load a b/w bitmap in the slot. It can have any shape but keep in mind not to have any antialiased borders between the black and white parts or you get a strange effect on those borders.

It can speed up rendering even if some caustics are involved I think. In any case it's worth trying more than network rendering atm.

[Maxer]

I've got a render farm of 115 machines so the plan is to eventually have a Maxwell farm once it can be run as a service in windows.

Thanks Mihai I'll have to give that a try!

[sidenimjay]

i will have to do more testing with this as i only implemented a few weeks ago . . .

mihai, if you are reffering to bitmap region rendering, i cant speak, but if you say that caustics wont work with a cropped render, i must disagree, and have a few renderings as proof

initial tests have shown me that it is also incorrect to say that the image takes as long to clean up when rendering smaller tiles of the original size.

if you render an image at 360x243 then render that same image at 160x122 it takes roughly 1/4th the amount of time to render, regardless of what you have in the scene. true??

this is the theory behind crop region rendering. the enitre scene is voxelized and one quarter of the pixels are rendered.....therefore for the tile in question, instead of calculating for the 87480 pixels in a 360x243 rendering you only calculate for 21870 pixels . . . . .so you are effectively rendering 65610 fewer pixels . . . . .now i would hope that tile would not take 60 hours to complete as it would when rendering the entire image considering again it is rendering 65610 fewer pixels

with enough machines to render every tile concurrently you are limited only by the longest tile

we used this theory and a similar method to render the fire for Reign of Fire. initially the frames were taking 20-35 hours each at film rez. since we were rendering with jig, we had a voxelized data set similar to maxwell , and we tile rendered on a 500 proc farm. this did not reduce the 20-35 hour calculated render times, but allowed us to turn around a bad frame in a couple of hours as opposed to days . . .

[Maxer]

sidenimjay, I see how this could be a very useful tool but if you have 500 machines I'd guess that they are not all the same speed. How do you compensate for faster machines that reach a higher sample level than slower ones, do you just place a limit on the sample levels and the time that the images can render for?

[Mihai]

mihai, if you are reffering to bitmap region rendering, i cant speak, but if you say that caustics wont work with a cropped render, i must disagree, and have a few renderings as proof

Please post them, it would be great if it worked but I don't think so with caustics. I have done tests myself using the bitmap render region and the caustics didn't clear up any faster.

You have to keep in mind Maxwell being unbiased has to take into account all light interactions, throughout the scene, so with caustics I don't think it matters if you crop or not the render.

Try testing a scene where you have largely a diffuse simple surface, and only a small part where you have clear caustics effect, for example an emitter shining through a piece of glass.

In fact let me dig up the scene I tested with....I posted it here a while ago...

[sidenimjay]

i think we are not talking the same thing :


there is a bitmap region render function : this uses a bitmap to mask out areas of the render

there is also a cropregion function that actually sets the camera to only render pixels x1-x2 and y1-y2

i am talking about the second . . . .

in the scene i was testing i have a mirror, which is an eigth inch thick glass and 1mm chrome backing, and a glass sphere . . .

i am not at home right now, but i will find the pix i used in my testing and post them . . .

we must also admit that we are not using the same plugin, i for one(the only one possibly) am using the houdini variant . . . .

[sidenimjay]

maxer,

forgot to send an earlier reply:

yes it is a must to set a target sampling level when tile rendering and when rendering animations

i would go a step further to say that the farm should be homogenous in that it should be all amd or intel or apple but dont mix them as they all handle floating point differently . . .not sure if that matters to maxwell or not . . .

[sidenimjay]

mihai and all,
i have decided not to post a pic, as anyone can simply say, oh those black areas were added post . . . .


dont worry however, i want to post the mxs file so all who care to see my point can download . . . .

alas, i too make like the sloth when looking through 1300+ returned posts from a search using the terms "how upload mxs" . . . .

anyone give me a good place to upload an mxs file? i have no ftp of my own

it is a mere 9kb file so , i can email it out , but that would get tedious . . .

tho i must note , while testing again i am having some troubles with certain values of crop regions , ones that i could swear were working when i implemented the code.

wierd , thread terminated while locking mutex . . .

perhaps nl could use the houdini plugin to generate the same result i get
lisux?

[Mihai]

Ok I see what you mean now, I wonder how this can be set up in Max though....I think Max supports the "regular" region render with Maxwell but I don't know maxscript at all to automate the cropping..

You can use rapidshare.de for uploading the file, no registering or login necessary.