Maxwell 2.6 - anyone try it for Softimage yet?

[Kzin]

i did a test with the classroom scene and some basic shader with bump and glossy.
compared to a complete brute force method in mr with ibl and ip's its way slower.
scene setup for shading and lighting is comparable with mr, so the advantage of faster setup is no more.
the light distribution is better in maxwell, but you pay with higher rendertimes. if rendertime is not a problem i would give it a try for arch viz stuff.
at the moment i cant talk about other features like hair or dispmapping, had not the time to test it.
the fire preview is a nice addition for "fast" previews, makes things a bit easier. integration is not that bad, you have to learn a bit the shadersetup, but this take 5 minutes. ;)

[mihaiNL]

I hope I have time to do more renders....I have to finish the plugin manual, I'll let you know when it's done.

Kzin, can you give me a bit more details? Or if you can please send me the XSI scene to test? What were the render times and resolution for both? There is one thing you have to be careful in Maxwell with the materials, it's to not set a too white reflectance 0 color since this will introduce more noise than necessary and also a lack of contrast. A limit I recommend is around 220-225. If you set it at, or close to 255 that means this material will reflect back almost all light it receives so light just keeps bouncing around. Such materials don't exist in the real world (except one called Spectralon but..... )

[Kzin]

I hope I have time to do more renders....I have to finish the plugin manual, I'll let you know when it's done.

Kzin, can you give me a bit more details? Or if you can please send me the XSI scene to test? What were the render times and resolution for both? There is one thing you have to be careful in Maxwell with the materials, it's to not set a too white reflectance 0 color since this will introduce more noise than necessary and also a lack of contrast. A limit I recommend is around 220-225. If you set it at, or close to 255 that means this material will reflect back almost all light it receives so light just keeps bouncing around. Such materials don't exist in the real world (except one called Spectralon but..... )

no white mats, thats right, a rule for all renderers. ;)

the lighting is done with phys sun/sky only.
it is rendering at the moment further so i can decide how long it took to get rid of the noise (a noise level that is acceptable). until here it took 9h 40min on my core2quad@2,667ghz
(forget about the light leeking near the windows, its because of bad geometry, not of false rendering)

[mihaiNL]

This is with Maxwell 2.6? Wondering about the aliasing near the windows. One thing, the bump seems much too strong and this can contribute to the noise. Going to try this scene....

Do you have the Mental Ray render and render time?

[Kzin]

This is with Maxwell 2.6? Wondering about the aliasing near the windows. One thing, the bump seems much too strong and this can contribute to the noise. Going to try this scene....

Do you have the Mental Ray render and render time?

yes, its the latest maxwell version.
strong bump is sometimes needed, so i rendered this to show if maxwell can handle this.

i can upload the mr rendering later this day, rendertime was about 6,5 hours with glossi walls, but without glossis on the board.

[Kzin]

this is the mr rendering. phys sun/sky with brute force ibl and brute force ip's. no portal lights or other additionel light sources. 6 bounces with 4 passes for ip's, 100k importons. ibl quality with 0,5 and 50 rays for ip's.




this is the maxwell rendering after 20hours and 21 minutes. as you can see, there is some noise under the tables that should be cleaned up with additionel sampling levels.
compared to the mr render, its a bit sharper because of the aa filter in mr. i will render another one overnight so you can better compare. of course the maxwell render has no indirekt bounce limit like the mr render, so its a bit hard to compare. i will also try to match the glossies and bump values of the maxwell render.

[mihaiNL]

Thanks for posting that, I'm testing this scene too and I think the user normals need to be deleted first and the smoothing angle lowered for some objects or the shading will look pretty bad. Also I'm not getting the aliasing by the window frame as in your render....strange. I'll post a render soon.

The Mental Ray render besides not having enough bounces also is overall noisier than the Maxwell one, no? (the walls) Question is what would it take to make it cleaner and with better AA?

[Kzin]

The Mental Ray render besides not having enough bounces also is overall noisier than the Maxwell one, no? (the walls) Question is what would it take to make it cleaner and with better AA?

i corrected the noise, had to increase the sampling a bit, its because of the glossy character of the wall shader (which the maxwell render does not have). i render this overnight and will see how long it took.
the noise itself in the old render is smoother compared to maxwell which has a "firefly" character, but the actual mr render is way better.

the classroom needs alot of tuning, thats right, took some time to get it to render clean. some geo problems are in, but its good enough for testing.

[Kzin]

this is mr and took 12 hours. its now noisefree, cleaner then you would need it for animation rendering.
importon density of 0,5 and 6 indirect passes. lighting is a bit different as you can see, tryed to match the sun position of the maxwell render. used wide glossis everythere.
i also used fstop of 5,6 instead of 8 in the last render to make the darker parts brighter for better noise judging.
unified sampling with min1/max300 and quality of 12, 1 local samples for all shaders and the sunlight. ip samples 50 ans ibl quality 0.5, like in the first render.


in general, the classroom uses a lot of indirect lighting, this means you have to render more samples as you would need for more direct lighting like outdoors. so i am sure maxwell and mr will render faster with such a lighting setup.

for maxwell, does it use the linear workflow automatic? have i degamma all 8 bit inputs by hand like in mr? i wonder of this because the textures in maxwell renders are alot darker so i ask myself about that point.
would also be interesting to see some more maxwell renders and the rendertime/machine specs to compare quality/speed. ;)

[Maximus]

Kzin, could you please share the scene? cause i'm havin an hard time to use IBL or IP with ctrl ghost because it doesnt update the buffer so its a total mess..
I'm so willing to take a look at your setup if you are so kind.

[Kzin]

Kzin, could you please share the scene? cause i'm havin an hard time to use IBL or IP with ctrl ghost because it doesnt update the buffer so its a total mess..
I'm so willing to take a look at your setup if you are so kind.

saturn explains this some time ago, because its a geo shader it does not update automaticly, so if you change a value in the shader, you have to refresh the region twice.
and also of this, if you render an animation, the first frame is not showing the ibl lighting. i know its not a good way so lets hope ad will implement all this and the new ibl features in si2013.