3DWorld review: VRay 1.5 vs Brazil V2 -> VRay SLOW!

[rivoli]

ok, here's the scene kindly provided by Lele. all props to him, I'm just hosting the file.

you can download it Here

Here you can find the original brazil and vray ones as provided by 3DW.

[m_hinks]

im rendering it as i type!

Interesting method of IR and LC your using there. I might have some questions about it soon!

[^Lele^]

Err, yes, there are a couple of things to note:
It's the LC calculating the glossies, which speeds it up by a huge margin, while not losing anything in quality.
Notice, though, that for some reason i preferred to script something that changed a parameter ("treat glossy rays as GI rays" set to "always") in the materials, rather than using the one under the LC rollout.
Also, the LC values for samples and sizes make sense only when checked through my LC/PPT calculator (25Px in size, 1 of shading rate).
The rest is fairly normal.
Oh, for the daring ones, you may try and add in a few light portals at the windows, as it will better the light distribution even more, at the expense of a slower IRmap (if set to be stored with the IRmap, of course) or a slower direct computation (the normal RGB pass).

[m_hinks]

Cheers Lele. On the Ir, you are only using 50 for the HSph. and 20 Interp. Is that not a fairly blurred solution, or are you using the DE to tighten the details?

[^Lele^]

It's on purpose, yes.
With DE, there's no need for the IRMap to be overly sensitive.

[RErender]

Lele, I saw your latest render on CGtalk, the table looks great

Now I fell bad I have twisted your arm, hehe, great job on going for speed though. Just to completely put it to rest, have you rendered the original 3Dworld file on your computer, and if so how long did that take?

[^Lele^]

Ahah Eric, thus far your "arm twisting" only brought me good things, so why not? :P

I did.
It rendered in 3:24 on 32cores, versus 3:03 for the optimised, first attempt, on the same farm.

http://forums.cgsociety.org/showpost...7&postcount=52

this is the original post on cgtalk.

[RErender]

I was curious how much faster this one

http://forums.cgsociety.org/showpost...&postcount=104

was than the original 3Dworld scene on your X2, I think their scene was about 20 min on my dual core so i guess your scene is about 5-6 times faster?

[^Lele^]

Roughly 5 times faster.
The original one I stopped at 43 minutes, and it wasn't finished yet.

[RErender]

k, so about 1.2 min in Blogotime, conclusion: Vray is 2x faster than B2


Heh, I had to do it...

[^Lele^]

LOL
well, I sneaked it in as a post, and avoided typing "do the math yourselves".
Now you spoiled it!

[RErender]

Heh, well I wanted to post it on CGTalk in bold letters but i was able to control my inner fanboy atleast that much.

[paulison]

lol a rendering in 1.2 mins that really does prove vray is better ...halarious!

what id like to see is a really nice high resolution rendering that takes at LEAST 1 hour to render and the quality of the image is a huge improvement over both the B2 and vray attempts done to date. Maybe not possible with this scene?

People have covoured speed (which really doesnt matter to me anyway) and showcasing features (which doesnt matter either beyond a certain point) but i think quality has been lost a bit in the process. I think the problem is that many people (from the brazil side of the fence) still assume that vray wont be producing images of as good quality so it would be great to see something of an excellent production standard rather than an insanely optimised preview render. wish i had the time and skills to do this myself but i dont

If you can show people an almost perfect image then the people who count aren't going to worry about the fact that it took more than 5 minutes to render. everything i do takes 1-24 hours to render (even prelim and preview renders) and i think that is a realistic production timeframe for an image that can be printed. Only hobbyists will be seriously going "OMG I have a bigger E-peen b'cos my rendering took 2.5 minutes making it 2x as fast!!!" + crazy fanboy emotes.

would be great if someone could do this just to put the quality side of the debate to rest as well

[RErender]

any of the modern renderers can produce a great image with good mesh and materials and given enough time, it's really the speed that sets them apart. By your test you'd best just go unbiased and be done with it.

[paulison]

i understand that but speed is really pointless when it comes down to comparing 2.5 mins vs 5 mins because no matter what you do (as is so obvious to most people), I really don't think that the quality will be acceptable. My point is that there was such a heavy focus on speed within timeframes that produce such poor results that arnt even worth comparing.

In terms of features, dont you think that many recent rendering engines are actually capable of doing the same things. Only the users workflow changes

The problem is that even though we know vray can produce great results (and it is about the user more than the tool), there are still people out there who think vray cant produce images of a good quality and these kinds of debates don't help when the focus isn't on quality

I just think that any comparisons/results would be more useful to to be done in a way that the results can be compared to simulate an actual project, rather than a low res speed /feature contest. For me its kind of like the debate over rendering times with people overclocking their home computer to be 10% faster and then bragging about it - pointless and irrelevant to professionals IMHO.

I know i'm stating the obvious and everyone probably agrees anyway. just adding my little rant in while everyone else does the same :P not intended to be an attack on anyone at all, but just a constructive suggestion and request for more focus on quality instead of speed.

i think its obvious that speed is going to be resonably similar in current rendering engines while methods of calculations are similar between products which is why I've said speed doesnt matter to me much beyond a certain point (anymore).