Messiah render speed comparison.


#121

nah those are 4 threads. no sense using 1 thread if in real world people would use whatever works the fastest. not messiahs problem that kray cant do multithreading.

anyway i think we can all agree that there is plenty of headroom to extract even more speed out of messiah. but i think this scene has been beaten to death. I am working on something a bit more interesting not only as a good benchmark, but to see what messiah can do. i think we need more user created unique renders here.


#122

I think this scene bores me even more than it does you :wink: Wegg and me had beaten it to death months before you started this thread.
I reread your first post and found some data wrong. I rendered with LW and FPrime and Messiah with 4 threads and Kray used only 1 thread ofcourse (until multithreading is added in Kray1.7)
I used 4 bounces in FPrime and Messiah and Kray used 24 bounces. All this on my dualcore AMD64X2 4800+.
So indeed that it’s not messiahs problem that kray can’t do multithreading, but it doesn’t help it in the end. Kray still renders way faster on one core and way more accurate.
I showed earlier a rendering from labuzz redone in Kray. There you see too can that Messiah lacks accurate radiosity.
I am glad that 2.4 update fixed some of the very bad reflection errors, but in the field of accuracy nothing is solved because it needs a new final gathering method of the photons.
Messiah already has an accurate Monte Carlo if you leave the (experimental according to docs) GI Noise Reduction off. It means long rendertimes and MC noise though.

I noticed that in your renders you took a wrong lighttype and you used very high GI radius value. Basically that means the local GI is blurred so bad that the spheres have no contact shadows anymore.

Here are my (really final!) Messiah results:

notice contact shadows under spheres


this one is best I could reach with all settings.
GI samples 24, GI depth 4
12000000 photons, gather radius 3, gather count 10000, tolerance 1.0

here a lot of different values, it is interesting to see that if you want to rescue the contactshadows, you have to lower GI radius and that gives a lot of photon noise, you can also see in the reflections how bad the rest of the scene gets, like walls and ceiling.


#123

here is the Kray render:
http://www.xs4all.nl/~ezwart3d/LWbenchmark/Radiosity_BOXKrayFinal.png
maybe you can change the url in the first post in the thread to that.

for the people who just say, who cares about those contactshadows?!, think about it. In some scenes like an interior almost the whole picture can be an space that is such a contact shadow. So when this is not accurate, everyhing appears to “float” on the ground and lacks contrast.

here is a sequence that shows the direct light as white and the GI in different stages of the render:

first pass, contact shadows

the GI noise reduction eats them away, only under the left larger sphere remains some faint one


to get them on all spheres, the GI radius gets so low that there are photon splotches anywhere…


#124

i still think there is plenty of potential to use for animation. afterall the fastest render was in 2:34 minutes. just up the quality until you get a time you are comfortable with.

also kray is limited in its shader support and geometry. with messiah you get all the shaders.


#125

I think it’s a great observation with the contact shadows. That shouldn’t be too much of an issue figuring out how to optimize our photon solution to retain them! Great thread! I’m sorry I wasn’t paying attention to this one from the start…but now that we’ve got a milestone done we’re finally open to check deeper into other sections…just like this one! So thanx! :slight_smile:

I’ll keep you updated! :wip:


#126

hey taron, you think you can also look into a more natural refraction solution? something more like a checkbox to eliminate the need to create “AIR” polys?

and the render time is still not calculated with the photon emission pass included.

thank you for all the work btw, really nice to see bugs get squashed :slight_smile:

p.s. lets talk about krays multithreading when its done :slight_smile:


#127

Yeah I´ll kill for that one :bounce:


#128

I am going to this thread once more, because I didn’t read the first pages thoroughly.
The Kray render has 24 bounces, but with a Monte Carlo renderer (first pass) that is insane.
I tried to set it lower to get more speed, but then the arealight shadows under the spheres and the ceiling and upper corners get way too dark. A light in such a box should light up the whole thing and wash out the shadows.


#129

24 bounces…iiik. That’ll be some sort of “trick” in terms of “intermediate baking”, which is something I thought of not too long ago. I do see that it should very much be a better type of photon distribution and reception/evaluation. Sounds like a curious and fun project to research into… coincidently this is along the lines of something we are preparing ourselves for these days. SO: VERY COOL, thanx a lot! :lightbulb

As for the refractions, yeah…hmmm…let’s see how we can implement that the most elegant way. I’ll have to think about that, but it’s something that should benefit greatly from being implemented into the core… …we’ll discuss the possibilities of a direct integration into the shading procedures, or if we simply make that an extra little node thing. But it will come and most likely very shortly, too! I seem to remember that Thomas Helzle was doing something like that already, if I’m not wrong. He also made some beautiful renders with it, I thought…hmm… anyway, it should be a standart part of the rendering. So, again, THANX, great call again! But I really want to point out that Thomas was thinking about it, requesting it and even doing it himself a few years ago, if I’m not wrong… …so don’t think I would have forgotten. There were plenty of little things falling a bit behind and as we build up the major elements we develop the space to fill in all the furniture piece by piece! That’s a pretty one, too! :thumbsup:


#130

Hi Taron,

this sounds awesome! Photon renderers like vray and kray use lightcache to store the first irradiance pass, that works great and fast. I don’t expect Fori to write a new photon renderer overnight, but I feel Messiah:render is almost there for me, if it spreads the photons to places where they are needed! I started testing Messiah again, because it can run on my BNR farm and Wegg told me it has matured now after I complained about not being able to use FPrime and Kray on the farm. The new shaders and all are great, and also the pure Monte Carlo renders are good now.

look at this picture, this is a testrender in Kray showing only the photons (200000 24 bounces):

the next is the first pass in Kray showing the gather count: (13secs)
notice the large “gather radius” that cannot hold the shadow under the smaller spheres

This is how the first pass looks filtered and this pass is stored for the final gathering upcoming second pass:

If you would add the direct Messiah lightshadow you see exactly what Messiah make of this scene, interesting, this is about where Messiah photons ends and where Kray starts to do the actual render.

It spreads a grid of points from where it sends off rays all around to calculate the actual light there and it uses this precached irradiance for fast “lookup” and interpolates the samples.

this render was made with “only” 200000 photons to make the photonmap and from those white dots between 200 and 400 rays are shot to calculate the exact illuminiation. You see that there the contact shadows are back in the picture.
It’s true that every scene “needs” its own parameters but we found a very easy recipe to get there, most scenes all work with same settings and the only thing changing is the number of FG rays and the spread of that, with “small overkill” it works for any.
I hope a similar thing can be added to the Messiah renderer and pictures like in this post should be no problem anymore: http://forums.cgsociety.org/showpost.php?p=3261217&postcount=84
see also my Kray render of that, it shows well what we miss in Messiah now.
http://forums.cgsociety.org/showpost.php?p=3274961&postcount=101


#131

exactly, kray uses the same “trick” as FPrime to interface with LW, so no access to shader tab and such, hopefully LW9 node editor will help there. I expect FPrime and Kray updates to hook into that.
BTW if there is anyone realising kray’s shortcomings, it’s me. You are the one constantly talking about multithreading, not me. I just rendered in Kray like it does at this moment on my dualcore and used more threads in Messiah and FPrime and kray times hold up well. btw for this particular scene I saw in FPrime no visual improvement when using more threads compared to one in the same rendertime:
1 thread:
http://www.xs4all.nl/~ezwart3d/LWbenchmark/LWBenchmarkKrayFPrime5m36qual5.png
4 threads:
http://www.xs4all.nl/~ezwart3d/LWbenchmark/LWBenchmarkKrayFPrime4threadsSAME.png

strange huh?! In Messiah this is really parallel it seems, I didn’t try, but it looks like that.

Also, maybe you should look more critical to the latest renders you produced in this thread, those fast renders look nothing like realistic lighting. Use a Spotlight and only thing you can change then is bounces, GI NR settings and AA. Your result would then be about the same as mine or Wegg’s.
you write: “just up the quality until you get a time you are comfortable with.”
That’s kinda strange after I showed you in a long post there IS no setting that ups the quality. I have to use a compromise now to get the result I showed. The time is less important, but using Monte Carlo without GI Noise Reduction just takes too long on “normal scenes”
I agree we should try prettier scenes now, but I just used this scene to illustrate some problems in any renderer, it already led to the fix of the reflections, so it helped :wink:
I tried a render of a room earlier even before Wegg and me tried to tackle this LWbenchmark, will show later.
I am doing all this to be able to use Messiah for supposedly nobody else wants it for :wink:


#132

I must be crazy but here it comes anyway:

There seems to be more demand that I thought. :shrug:

Questions:

  • How many of you would be interested in buying TLHPro if I would release it next week?

  • What would you be willing to pay for the collection - more, the same or less than for AoN?

  • Would you be willing to accept preliminarily docs for now, with full docs (not as big as AoN though!) later? The tools are very easy to use IMO, so you should have no problems.

  • Two plugins from the originally announced set may be missing for now or reduced in functionality (Wireframe and Ambient Occlusion) but you would get:

[ul]
[li]TLHPro: BiasGain (old, updated, combines Bias and Gain in one node)
[/li][li]TLHPro: Channels (old)
[/li][li]TLHPro: Color (new, simply gives you an animateable color when you don’t need a full gradient)
[/li][li]TLHPro: Distance (measures Distances in several ways)
[/li][li]TLHPro: Doublesider (texture the front of polygons different than the backsides)
[/li][li]TLHPro: EasyGlass (makes the setup of simple glass objects extremely easy - no more doublesided objects needed)
[/li][li]TLHPro: EasyMath (allows you to stack up to 10 mathematical functions in a row so you don’t have to clutter your shaderflow with multiple of my older math nodes)
[/li][li]TLHPro: Merger (28 PS-like layer modes for your combining pleasure :wink: )
[/li]You could call it also an advanced Mixer Node.
[li]TLHPro: HSVtoRGB (old)
[/li][li]TLHPro: RayType (switch between different materials for camera, reflection, refraction, GI and custom rays)
[/li][li]TLHPro: RGBtoHSV (old)
[/li][li]TLHPro: ShowValues (a helper I use for development - shows the min, max and average of values and colors that go through it. It doesn’t do anything to the values)
[/li][li]TLHPro: SplineCurve (use the messiah graph editor as an superfine spline editor for textures)
[/li][li]TLHPro: SuperStep (old updated, Clamp, Step and Smoothstep in one node)
[/li][li]TLHPro: Thickness (measures Thickness in various ways)
[/li][li]TLHPro: Time (allows you to use time as part of the shaderflow - with multiple options for intervals etc.)
[/li][/ul]

The old TLHPro Development Thread is here, browse it for more info on the included nodes:

Information about AoN:Studio, the procedural texture set for messiah:studio can be found here:
http://www.screendream.de/AoN_studio.htm

Maybe let me know on the TLHPro Thread what you think.

Taron: would you guys be able to finally make sub-groups available for the right mouse button shaders list in the 2.4 patch?
With my TLHPro tools added, the list grows longer than the screen is high and this list doesn’t scroll, so the first and last in the list can’t be accessed.
I think the best approach would be to allow for sub-directories under plugins and the directories are reflected in the list as submenus (like the materials in the palette). So a sub-folder under plugins called “TLHPro” would show up as a submenu “TLHPro” in the shader menu. If that is impossible, a specific entry in the plugin itself would be also okay, but not as flexible for the user to organize shaders just as he wants…

Remember: this is no promise, just a question to gather some intelligence…!

Best regards,

Thomas Helzle


#133

I’d be interested, though it would be less attractive if Ambient Occlusion weren’t included…
That’s a damned handy shader…

Price? The same as AoN, I’d say.


#134

I’d be interested. I’d say the same price. Yep.

By the way. It would be incredible if you did release them.


#135

As a beta tester (so to speak) for the TLHpro shader set I can only recommend them to everyone. They make again a big improvement in working with the Messiah shader system. :thumbsup:

Great work Thomas! :bounce:

A price equal to the AoN shaders is totally appropriate.


#136

THLPro shaders !? :bounce: Where do I have to sign !? :bounce:

Sure, Thomas, get them out as soon as possible ! I would pay a similar price for them as for the AoN shaders.


#137

Well at least $300 right there Thomas. I’d say its worth it. At least you’ll get some money. With all the time you spent on them, I’d say its worth releasing them.


#138

Sign me up!!! :bounce:
I would give same or more as the price for AoN.
I´m earmarking cash for it right now.


#139

Count me in as well.


#140

Thomas,

I’ve already stated that I would purchase given the opportunity. But I did notice that the two listed below was not on the list, is this intentional?

TLHPro: Occlusion (brandnew - ambient occlusion and more!)
TLHPro: Wireframe (tons of options for wireframes, vertices and polygons)

R