Mental ray render times making me MENTAL

[THEPRAYINGMANTIS]

Hello all,

I’ve been testing out mental ray in max 2009 design for architectural renders here at work and I’m quite happy with the results. My only real concern is trying to keep render times at a reasonable level comparable to some of its competitors (Vray) and I’m really having a tough time. The attached image @ 3000x2250 took almost 40hrs to render and that’s just too much. I’ve attached my render settings. If anyone can suggest some settings that would give me a good balance between quality and speed that would be great. I’m using MR sun/sky with sky portals at the window openings and AO on all materials. Also, does anyone know how to gamma correct Promaterials? I don’t mean the textures, I mean the reflectance colour for semi-gloss paint let say. There doesn’t seem to be a way to add the gamma/gain modifier to the reflectance colour. Also, I would like to thank Zap, SandmanNinja , Justin Hunt and the others for helping me on my last thead.

http://www.cgarchitect.com/vb/35364-interior-lighting_washout-wall-colours-unexplainable-artifacts-blotches.html

[aaronrumple]

Are the interior lights lights or just a self illum material?

40 seems a "bit" high, but what is the processor spec. Single? Dual Core? Quad Core?

 

3300 px wide here take me on average 15-20 min. But that is using a farm. So that is about the same as 5 1/2 hours based on the number of cores for a single system.

[Macer]

I consider the Pro Mats to be dumbed down A&D mats, so just upgrade to them for more control. You can probably up your spatial contrast to .08ish - that will improve times without loosing too much detail. How about lighting? What are your shadow settings on your portals? Can you lower the samples? Better yet can you turn off shadows for at least some of them? This will have a massive difference on speed. Thats a start anyway.....

[John Dollus]

well, your refraction limit is way too high and is probably the culprit. should be one level for each layer of traced transparency and nothing in the scene has more than 4 that i can see

[THEPRAYINGMANTIS]

Actually both, self-illuminating material on the sides of the light fixtures and 46 photometric lights inside the fixtures to illuminate the ceiling. The photometric lights are only set at 5% intensity (1595.1cd) Photometric data taken from manufacturers website. It seem pretty high for these fixtures but anyway. I don't have a renderfarm but I'm using a Dell Precision PWS490 Intel Xeon 5140@2.33GHZ (2 processors) with 4.0GB of RAM, running on VISTA 32bit.

[THEPRAYINGMANTIS]

All sky portals have 16 shadow samples and the photometric lights are casting mental ray shadow map shadows.

[Macer]

Oh yeah, I missed the refractions! Limit your max trace depth, you can also reduce reflection in the FG rollout to 3 (you could even drop the preset to draft without much loss of detail).

 

And what do the rest of your Photon settings look like? I've found with complicated lighting I often get faster renders using the lowest possible sampling radius (a couple of hundred mm as opposed to a few metres plus)with more photons per light shot. This leads to a longer photon calc, but quicker renders (probably of more use for animations though!).

 

That aside, photometric lights that use web distribution will increase render times hugely too. In a scene like this without much light splash on the walls I'd try setting the lights to not use webs for distribution - spot lights with 'point' shape is the generally very fast and will give you very quick results, but experiment to see what looks acceptable.

 

Try knocking your shadow samples to 8 and see if they aren't too grainy, or if you using internal lights too, I would turn of shadows for the portals all together. You can probabbly see where I'm going with this - reducing the amount of 'area' shaped lights (especially casting shadows), will have a huge affect on render time. You could try ray tracing shadows on the internal lights too. Its just a matter of finding the balance between speed and quality. You just have to mix it up a bit and run loads of tests!

 

There's some really helpful info on Master Zaps site about creating great water in MR too.

[Crazy Homeless Guy]

That aside, photometric lights that use web distribution will increase render times hugely too. In a scene like this without much light splash on the walls I'd try setting the lights to not use webs for distribution - spot lights with 'point' shape is the generally very fast and will give you very quick results, but experiment to see what looks acceptable.

 

If you do use Photometrics, set the shadow distribution to point, otherwise area shadows will be cast, which will slow down the rendering. Often the distribution switches to match the light profile when a light is loaded.

[Spooner04]

The max refraction is what caught my eye too. that seems really really high. I think I remember reading somewhere that it doesn't need to be more than 6. 4 would probably be good. Is there any reason why you went with 17?

[Crazy Homeless Guy]

The max refraction is what caught my eye too. that seems really really high. I think I remember reading somewhere that it doesn't need to be more than 6. 4 would probably be good. Is there any reason why you went with 17?

 

I am not sure, but I think the refractions may become double the problem when you consider this image is rendered with caustics.

[THEPRAYINGMANTIS]

Re: Interior Lighting_Washout wall colours and unexplainable artifacts/blotches

OK a few things to fix

 

Turn Scanline OFF, Saves a bit of memory

Reflection/Refraction Trace depth try Max trace depth 20, max Reflection 3 Max refraction 17

 

Final Gather

When using Photons set the Diffuse bounces to 0, any thing above are disregarded at render time

Leave noise filtering to standers, High takes way too long

 

Photons

 

Turn ON Optimize for FG

Max Trace Depth 20

 

jhv

 

That was the advice given to me from Justin Hunt for my last thread. I must admit I thought the values did seem a bit too excessive but I took his advice anyway since I'm a noob using mental ray. I did not switch on caustics though.

[Macer]

I'm sure he had his reasons, he more masterful than most with MR. However this post is enquiring about how to reduce render times, that one probably wasn't. Here is the link that I mentioned before:

 

http://mentalraytips.blogspot.com/2007_06_01_archive.html

 

Cheers,

 

Mark

[Justin Hunt]

Yea I did have reasons for high refractin numbers, there are alot of levels of transparency. Bare in mind that once it passes through the levels it needs it wont calculate any more, however if it reaches the limit before it passes through all levels of transparency it will return the backrouund colour (or black).

 

What I cant see is how many Photons per light you are shooting, or if you are using Automatic or manual for each lights photon calculations. With that many lights large photon numbers will take ages to calculate and even longer to render.

 

You could also drop the FG to "Draft",

 

A nice trick I have been teasting, when using photometric lights with a Sun and sky system. Crank the Photometric lights' Energy value right up. Keep the MrExposure Physical scale set to "Physical Scale". This way you can keep the lights multipiers at there default values and still see them light the scene and not be overpowered byt the suns intensity.

 

jhv

[thomasj]

i once was a mental ray freak, but now that i got introduced to vray, i think i'll never go back.

[Justin Hunt]

and this helps how?

 

jhv

[THEPRAYINGMANTIS]

Thanks for all your suggestions guy. I re-rendered the image and it took just under 5hrs with no noticeable difference in quality. To Justin, average photons per light is 50,000.

 

For future projects, is there a general formula or ratio that can be applied to determine the Max. Num Photon per Sample and Max. Sampling Radius and its relation to the side of room or scene? How do you go about determining these values? I just want to streamline the time it takes for test renders. When running a business, I have to justify the time it takes for project completion and that requires me to keep test renders at a minimum.

 

Again. thanks guys for all your suggestions.

[Spooner04]

just a side note for the future. The average number of photons isn't actually the number of photons per light. You can open up the mental ray message box and it will tell you how many photons each light is emitting. Alot of the time if you are using a daylight system the sun can hog alot of the photons and your other lights might be emitting 15 or something really low, making them pretty useless. you can adjust the sun multiplier down to get some more photons from those other lights.

 

I'm not sure if that was mentioned in your other thread or possibly you already knew that. Glad you got it figured out.

[mattclinch]

your spatial contrast values seem very low too. try 0.035 or 0.04 - it should still resolve those little struts in the roof but wont oversample unnecessarily.

[Justin Hunt]

just a side note for the future. The average number of photons isn't actually the number of photons per light. You can open up the mental ray message box and it will tell you how many photons each light is emitting. Alot of the time if you are using a daylight system the sun can hog alot of the photons and your other lights might be emitting 15 or something really low, making them pretty useless. you can adjust the sun multiplier down to get some more photons from those other lights.

 

I'm not sure if that was mentioned in your other thread or possibly you already knew that. Glad you got it figured out.

 

Good point, although reducing the multipier breaks the physical correctness. If your not worried about that then go ahead. Rather set the lights photon emmission to manual and specify a photon number and energy value.

 

The message window is an often overlooked source of information about whats going on in your scene.

 

jhv

[Crazy Homeless Guy]

Good point, although reducing the multipier breaks the physical correctness. If your not worried about that then go ahead. Rather set the lights photon emmission to manual and specify a photon number and energy value.

 

Are photons physically correct?

[Justin Hunt]

They should be if you leave mr to automatically calculate and you have set the exposure physical scale correctly. But seeing that you have the ability to change the energy values it is possible to break the laws of physics to get the results you want. Artistry over Accuracy.

 

This is the cast when using a combination of MRSunand Sky with Photometric lights. In order to see the photometric light you either have to crank up the light multipier or crank up the photon energy values to match the sun. The latter is easier to understand as the multipier values are what you'd use in real life. The former the multipier does not relate to what you would expect in real life but 100's to 1000's X bigger. (Hope this makes sence:o)

 

The other thing you can do is decrease the number of photons the sun will emit by setting the GI Photon value to less than 1 and increaseing the Photometric GI Photon value to more than 1. This is using Automatically Calculate Energy and Photons.Thus balancing the overall photon emission to favour the photometric lights over the sun, meaning that there will be more light detail from the photometric lights and less from the sun. Which is good for detailed interior scenes and gives more even spread of light detail through out the whole scene.

 

Bare in mind that if you are not using photons and just Final Gather, changing the lights energy and GI values has no effect, thus you end up cranking the light multipiers instead.

 

jhv

[Crazy Homeless Guy]

I have often poo-poo'd MR's ability to adequately bounce light around a scene. It wasn't until I started manually controlling the individual lights did I start to get results that I considered close to reality. Maybe my mind is just warped from years of using Vray, but I don't really think so. When I work Irradiance Particles, and Importons the light bounce around very gracefully. Much nicer and more natural feeling than simply working with Photons.

 

I also am incredible suspect of saying that the reason you can't see a photometric light with out cranking its value is because this is a normal condition of the real world. Even on a sunny day, I can look out the office windows, and see lights on inside of the building across the street. They may not be the brightest, but they are still easily noticeable.

 

This is not the case when leaving both the IES lights at their defaults while using a MR sun. Or Vray for that matter. It could be the the way our eye compensates is that much more advanced that simply tone mapping, but I doubt it. A photograph could verify this. It could also be that I am just focussing on the actual light source when I am looking at the building on the other side of the street in the real world, and I am not actually seeing the light that is cast.

 

Either way, at the end of the day it is important to remember that the engines we use themselves are just approximations of the real environment, and not actual representations of a real environment. Though some engines are designed to do a better job at this than others, but typically at a cost of speed and efficiency. Meaning Maxwell and Fry.

[Justin Hunt]

well said, which leads us to the need to learn how to get the best out of your renderer of choice.

 

I must admit that alot of it I just dont understand and its only through experimenting that I come to my explinations. SO if anyone has a better explination I am all ears.

 

jhv

[Justin Hunt]

just to illistrate what I was saying.

 

jhv

[Macer]

Thanks for the info Justin.

 

I'm looking forward to trying this on my next indoor/daylight scene.

 

Cheers,

 

mark