Messiah render speed comparison.

#1

So lets kick things off. There is a thread on newtek forums comparing LW to Kray and to many other render engine speeds using radiosity. Since Messiah isnt even mentioned im taking it upon myself to see how it stacks up.

*Lw rendered this benchmark scene in 26min50 with 4 threads on this AMD64 4800+ dualcore.
Using only 1 bounce.

Kray rendered this with 1 thread and 4 bouces (it doesnt support HT yet). 5:36

Fprime rendered this with 1 thread and 4 bounces in the same same time:

Rendering with messiah using one thread and 4 bouces 5:24

Quality is ouch, considering that im using a dual 3.6ghz Xeon (1 thread though) Messiah seems to have alot of problem with blotchy shadows in reflections. also there is a weird white shape reflecting in all the spheres and the LW logo has a few white spikes (they are black in LW but when i open the object in messiah they are white). any ideas on fixes?

btw, im using a spotlight as suggested by wegg. the spotlight angle is 90 degrees to simulate the area light better. im going to try upping the shadow quality.

Here is the project file:

#2


9:52

Using a Planar Light, Shadow Quality 20,

AA level 3, Threshold .015, AA pattern: Enh Hammer, Size 1.5

Light Sources: ALL GI Type: Monte Carlo, GI Samples 7, GI Depth 4, Auto GI reduction off.

GI Photons 250k, GI Gather Radius 100, GI Gather Count 500, GI Intensity 1.2, GI Noise Reduction .01,

I dont like how im losing the bottom edge of the spheres and the hard edge between the ceiling and walls.

#3

Hi,

I would like to participe and make some tests.
thanks for sharing :slight_smile:

cheers!

note: for white area on LW logo (did you check if have more than 4 polygons (ngons)), overlaped polygons ?etc…

#4

Everyone is welcome to contribute. just make sure you copy my settings and fill them in, i want to make it a resource so that people using messiah can refer to these renders to see how the settings look and how they affect render time. I know with enough tweaking we can get close to kray with quality/time.

FYI the project file has been updated with the fixed LWO. no more white logo pieces or weird shapes.

Also, i see that in LW the light object is using glow effect. anything similar in messiah?

#5


10:31

Using a Spot Light, Shadow Quality 5, Falloff Inverse Distance .0001 Cone Angle 90

AA level 4, Threshold .015, AA pattern: Hammersley, Size 1.5

Light Sources: Direct GI Type: Monte Carlo, GI Samples 5, GI Depth 4, Auto GI reduction on.

GI Photons 70k, GI Gather Radius 140, GI Gather Count 500, GI Intensity 1.8, GI Noise Reduction .1,

Again, the bottom edge is too light, also i reduced the weird reflected shape but a piece is still there. Atleast the LW logos are all black and the upper edge isnt as sharp.

#6

Vroom! Messiah rules. :slight_smile:

#7

awwww… dont take the easy way out. try to match the original

although id like to know your settings…

#8

I honestly can’t remember them. This was a test a friend and I did on IRC in December. He was using Kray and F-Prime.

#9

Stooch: What have you done with the wall behind the camera?

/ Svante

#10

it wasnt planar, i just opened it in modeler and then scaled it using local size. just load up the new .mpj file i linked int he first post. it has the latest assets in it.

now there are still artifacts but they seem to come and go depending on how many photons i use… go figure.

I figured it out. If the GI gather radius is too high, it will average highlights from the spheres into the wall making those shapes… or something.

#11


5:22

Using a Spot Light, Shadow Quality 5, Falloff None, Cone Angle 90

AA level 4, Threshold .025, AA pattern: Hammersley, Size 1.5

Light Sources: Direct GI Type: Monte Carlo, GI Samples 5, GI Depth 4, Auto GI reduction on.

GI Photons 75k, GI Gather Radius 12, GI Gather Count 300, GI Intensity 1, GI Noise Reduction .015,

Getting there. I can probably make it a tad faster or smoother. but i think messiah is as fast as kray. Now, who can figure out how to make the underside of those spheres a bit darker?

Note that anomaly dissapeared. If i make gather radius any bigger, it will come back. So i have to increase photons i guess. Wellp, its 9pm and im still at work, just for you guys so ive had enough for tonight. l8r.

#12

I think you could up the quality of the shadows a lot more. Try for 12. Then you could get away with a higher threshold on the Adaptive AA. I also prefer the RG pattern over anything else. It seems to provide cleaner edges. The rule is. . . more rays. . . smaller gather radius gives you more accuracy.

#13

but this last result is darker compared to Kray ( add more depth in GI panel)
or use the filter to balance the pic…

too much brightness on the floor …

you can add a Fresnel effect on reflection … you will get better and realistic render of chrome ball …

use a gradient( on normal )

#14

How do you figure that Opera? Chrome balls are reflective all over.

Here is my attempt. (Damn you Stootch for twisting my arm.)

Not quite as fast but I think It is of a much higher quality. Especially on the ceiling. That time isn’t accurate. I don’t think it is taking into account the first GI pass. Its more like 14 minutes.

#15

very nice. im trying to basically nail around 6 minutes. watch that anomaly wegg! see how the dark grey reflection has a lighter grey in it? not sure what it is, but it should be a solid dark grey per my example. lol damn gather radius.

but that render looks better then kray otherwise.

also im seeing a weird halo around the LW logo on the green…why is that???

also can you please post your settings? i want to show render noobs settings so that people can use these as a reference…

#16

now keep in mind… we are just using 1 thread… because thats all kray supports at the moment. so quite possibly, messiah has the fastest GI on the market. tomorrow i will unleash all 4 threads and see how fast/quality we can get.

#17

I think the problem is. . . the higher the Gather Radius. . . the more it starts trasitioning from good looking GI to an ambience with a color that is retty much an average of all the walls. And thats why your getting the glow under the balls.

Man those grey stripes are funky.

I wonder if the Ray Depth have something to do with it.

#18

Hey Wegg don’t push me !!!
I don’t have my dongle yet !

 I will kick your a$$ , when I will get my version !
 limited size: render only in 320/200 with the demo.
 unable to save too !
 
 1 week to wait ... (will be mailed today or tomorrow from PMG).
 
But yes good result ! ... I had fast result too ( with montecarlo).

advice for everyone: don’t use large sample ( ie: 50 or 100 ) use tiny value (2 - 5 ) it depends of the size of your scene and you resulotion of course …

How to find the good size ? my tip:
put 100 photon
put 10 for sample size
put 10 for gathering…

 now counts the dots, play with the sample size ...
 when it's done add more photon.
This way , you will undertsand how photon have a relation with smaple size ( then Quality and realism !!)

About the Shadows: to have radiosity in the shadow, add depth : 3 or 4 (that’s enough) (depth = bouncing light).

to reduce the noise, add more GI sample ( ie: 2 or 3 ) (higher value= less noise,  more definition)

as Wegg said: use the R antialias method (better for GI) … to find the good tolerence number, render a sharp edge (use value under 0,1 ) 0,1 is too high ! (somehing like 0.08, you can go under for better result)
Antialias level at 2 minimum (keep the filter at 1.5 … don’t use 1.0 or you will get sharp pixels)

 sampling your "area" shadows at 20 is too much ... 8 or 10 must be enough ...
 check your raytracing depth ( 2 bouces are enough to see reflection on the sphere with a value of 0.5)
 
 
 What's next ?
 
 calm down the GI instensity to 0.9
 change the white diffuse material(floor)   1.0 to 0.9 or 0.8....
 change the spread of the light and it's intensity too ...
 
 following thoses advices you will get fast result ...
 I get good result less of 1 mn 30  with v2.2a. (320/200)

but don’t fake too much to speed up your rendering: more you tweak, less it looks like radiosity ( check corner edge, color bleeding in shadows, check large white area too … to make a good balance bewtween speed and quality)

Wish to help more (sharing file),but I need a dongle !:slight_smile:

EDIT:

reformated the text a bit for a better comprehension (read it again in case of)
Use Ram (not file chunk) to speed up you render
put half of the size available in your computer ie: you have 2 giga of ram put 1024.

To stooch: to reduce the green halo, you need to touche your material ( a smaller diffuse value ie: 0.9 or 0.8)

And it’s good to share the setup, but it willnot be the “magic” values for everything. It depends of the size of the scenes and details ( test an other scene with smaller details on objects … the scene may need smaller radius sample, more photon, less GI,etc… )
the setup shared will be the good values for scene like this one only )

#19

hmm.
rendering at half rez isnt a good way to judge in this case.

when you double the resolution, you actually quadruple the number of pixels… so dont jump to conclusions just yet.

also what halo are you talking about? I got rid of all anomalies on the last 5:20 test.

FYI all tests are done with 1 thread, 1 chunk, ram chunks enabled. not sure if these settings transfer with my project file though. got another render coming up shortly.

#20

I wasn’t able to improve on your renderings in any way. Personally I think Kary has one up on us in how it is able to keep those darks under the balls. I’m able to get those darks if I up my Photons and decrease my Gather Radius but of course. . . that ups the render time. There is no use doing GI if it doesn’t keep the little cracks and details dark. Lets hope pmG has a look at our tests and figures out a solution.