[aitraaz]

Yep thats a lovely pic, but that LeCorbusier chaise is a high yield object, even pov-ray would have a hard time making it look bad...

[mverta]

Oh my GOD is that going a long way just to get a dig in. A "high yield object." That's classic. Brother did a bang-up job on that image, so give him props and don't minimize it like that.

_Mike

[aitraaz]

OMG.

I was praising the design genius of the chap who designed that chaise (that nutty swiss guy who turned architecture & the industrial world on its head ).

So don't be *defensive* mate, no digs whatsoever, great image, all props due...

[mverta]

Just wanted to make sure my man's getting the love. My bad.

_Mike

[Voidmonster]

Great test Voidmaster.
SJ, Voidmaster if they invite either of you to become A-teamers please resist - otherwise you'll turn to the darkside and spend your days telling us about the next new fabulous release and how great V1 really looks!

I'm making an active effort to SHOW you how good V1 looks right now.

I continue to be completely unimpressed by the assertion that the beta looked better. It looked different, and I'm not even sure that it's look can't be easily repeated in V1 with a clear understanding of how to translate materials.

I'm completely willing to be wrong, though. After this one, I have some other ideas. Mostly because I consider comparing the lighting between beta and V1 to be silly. It's not useful. Comparing against photos is far, far more productive. I am doing this round of tests mostly as an experiment in further understanding the material system and hopefully providing some good info for people who want to get the beta look. I'm not doing it because I think the beta looks better.

[superbad]

So help me out here. Based on what Mihai posted somewhere, I've been using 214 x3 to represent "white" surfaces (on backdrops and such). Looking at this, I'm thinking I should go to 200. How about even lower? Unfortunately, I don't have the time or computing resources to do tests like this. I applaud your effort and thoroughness.

[giacob]

V1 looks pretty freakin' good, actually:



From y_okaue's gallery. We're just all trying to make it better.

_Mike

yes great but what resolutions ad renderetime what is ... 3 days?
besides renders made by very talented people cant be the metre for measuring the value of a render engine.. i am sure that if okaue would apply to scanline could make a fantastic image even with that

[Voidmonster]

So help me out here. Based on what Mihai posted somewhere, I've been using 214 x3 to represent "white" surfaces (on backdrops and such). Looking at this, I'm thinking I should go to 200. How about even lower? Unfortunately, I don't have the time or computing resources to do tests like this. I applaud your effort and thoroughness.

This test uses 200x3 for white only because the original scene did. I would think 214 should be perfectly fine, but I'll throw one in at 214 for comparison in my series.

--Update:

The 214x3 render is still cooking, but at SL 8 I'm seeing none of the problems I saw with the 230x3 image. It should be completely fine to render, but I'll let it keep going just to be sure.

[Voidmonster]

My tests are getting slightly chaotic while I nail down a method. The full 1024x768 res up to SL 18 was overkill and was taking a long time, so I've scaled down to 640x480 and I'm starting at the beginning.

Observations so far (everything will be fully backed by closely monitored stats when I'm done):

Lambertian is about 1% slower than 214 x3 white. 214 x3 white seems to be about 1% slower than 200 x3 white.

Hard data is forthcoming, as well as the original propagation test.

[Frances]

I'm completely willing to be wrong, though. After this one, I have some other ideas. Mostly because I consider comparing the lighting between beta and V1 to be silly. It's not useful. Comparing against photos is far, far more productive.

I may be mistaken, and I'm sure I'll be roundly corrected. But in the past, even the harshest critics of Maxwell (not Maxwell users) had to grudgingly acknowledge the superior quality of light in Maxwell renders. This was a very real accomplishment of the Maxwell Beta, and should be considered as an important standard. Even one comment like "There is something odd about the light" is unacceptable. And the unacceptable has happened since the RC - V1 releases. Several times.

So maybe it isn't silly to try to discover what it was about the quality of light in the beta that sets it apart from V1. It took months before anyone would even acknowledge that there was a difference.

Overbright surfaces messing up GI is an issue with all rendering software that uses GI. So that is also a very important point to explore here. I'm a firm believer that realistic renders require realistic input. Maxwell, in particular, has been conceived based on that idea, extending the input to the camera itself. What you are doing is very productive and worthwhile, but I think that the tests that others are performing are productive and worthwhile too.

[Voidmonster]

60% through the second render and it's looking like the difference in speed between Lambertian material and non-lambertian material is pretty significant.

It's looking like the non-lambert material is going to be done in a bit over 4 hours and the lambert material took a bit over 5. That's significant enough that I'm rethinking whether or not to ever use lambert.

[DrMerman]

Nice tests everybody!

@voidmonster : looking forward to those results mate. Are you trying different roughness values, or one roughness value agains lambertian?

Dr M

[Voidmonster]

The current render is with 99 roughness. Next I'll do 90, then 50, then 10, then 0.

The latter three mostly for my own curiosity, but also because I want to see how the light propagates through the scene with much more reflective surfaces.

[Voidmonster]

See this thread: http://www.maxwellrender.com/forum/view ... &start=105

...for my detailed test results.

[Voidmonster]

Also, I have concluded from this test: never ever ever use Lambert. It's slower and it screws up light distribution.