[tom]

I should not have expected you to understand the reference.

Probably, I'm not as clever as you Frances. Could you help me to understand it? May I beg for assistance. Thanks!

[choo-chee]

well I hope the speed ppl post replies here will be noticed by NL... I have a new slogan for you NL guys (and please remember I need speeeeeed in whatever way you program it):
Maxwell 3: the light(speed) simulator.

[Frances]

I should not have expected you to understand the reference.

Probably, I'm not as clever as you Frances. Could you help me to understand it? May I beg for assistance. Thanks!

Well, you'd have to be a girl who reads Jane Austin, not necessarily clever. Lady Catherine was contented with just the idea that had she attempted to learn the piano, she would have been superior, which is something that cannot really be challenged - only wondered at.

That's all I can say without crossing a line of diplomacy.

[tom]

So, why do you think I've not understood the quote at all? Besides, me reading Jane Austin or not was not the case. I thought you are shooting nice quotes and wanted to contribute with another quote. If there is something I don't understand, it's the piano fallen into this thread from above.

well I hope the speed ppl post replies here will be noticed by NL...

For sure! On the other hand, we can't count the number of emails we're receiving only for the quality. There are many big brands which simply cannot do without Maxwell on their pipeline. It's because they're unable to simulate lots of special cases/materials they need to proof on other engines, no matter how fast or cheap they are.

[Frances]

So, why do you think I've not understood the quote at all?

Because the reference was not about learning the piano or music at all and you chose to contrast it with a solo artist vs orchestra reference. Substitute "learning the piano" for ...... something else.

[tom]

I can't subsitute learning the piano, it's just priceless!

[tom]

Anyway, back to the topic...

We underline something clearly that we're not going to implement X technology (and this is not limited to today's GPUs) if such a technology is not mature enough to port Maxwell into. Today's GPU's and softwares supporting them may look exciting to you but it's nothing new to us and we have plenty of information/experience on this area in NL labs. It does not mean we are against GPU support or speed. We won't make a version of Maxwell just for rendering with half of its features but, you should always expect a fully armed leap from Maxwell using all features instead. This could be GPU or not, we're not excluding any technology at all.

Looking through the pages of this thread, and this was what I wanted to hear!

Thanks Lars! I think we're very clear in what we're saying.

[dmeyer]

For us, quality > speed.

Speed is great, don't get me wrong, but unless a GPU based renderer is robust enough to make my life easier, to REMOVE steps from the pipeline rather than ADD them, I'm not interested. I can't imagine it being time efficient to use another renderer to get a feel for a scene, only to completely recreate it in Maxwell.

To me the ideal would be a real-time, in-app and in-viewport preview type of engine, similar to IPR. Take Mental Ray for instance. You can use IPR to tune shaders without launch a whole new render each time, but then kick off a proper render when you want a full on preview or render. Maybe it uses GPU, maybe not, but it does not ADD steps or apps to the pipeline.

[brodie_geers]

Can I ask a stupid question in regards to the response that Tom mentioned regarding never going biased and never faking anything? Isn't the line between 'faking it' and not, a bit blurry? I mean, wouldn't a purist say that bump and normal maps are just ways of faking displacement? Or thinSSS is just a way of faking standard SSS applied to a very thin (but not single planer) enclosed geometry? Etc.

-Brodie

[tom]

Yes, the line is quite blurry. This is why we have displacement and not only bump/nm and
this is why we have SSS and not only Single-Sided SSS. But you can't fake frosted glass easily...
So actually, that border is as blurry as in this image:



What is "fake" and what is not? It's very easy to explain and easy to keep in mind.
What we do is simulation. So, the whole thing is an approach for simulating the reality.
There are features we use as tools and there are different ways of implementing them
depending on your core structure. Some features cost more rendertime than others.
It's in our hands to make one or all of them more efficient but if we pass certain
borders, it becomes unrealistic so, the approach starts to fail. This is what I wanted
to say by using the word "fake". Otherwise a whole simulation itself is a fake.
It doesn't matter how fast you simulate when there are lots of limitations.
Today nothing is perfect but, Maxwell has a set of mature unbiased approaches.

[brodie_geers]

Thanks that clears things up.

(pun intended)

-Brodie

HAHAHAHAHAAH

[tom]

You're welcome.

[JDHill]

Can I ask a stupid question in regards to the response that Tom mentioned regarding never going biased and never faking anything? Isn't the line between 'faking it' and not, a bit blurry? I mean, wouldn't a purist say that bump and normal maps are just ways of faking displacement? Or thinSSS is just a way of faking standard SSS applied to a very thin (but not single planer) enclosed geometry? Etc.

-Brodie

The line is clear for me, and it is always only about the light calculation. That is what you are simulating -- what light does in a scene, and how the camera sensor relates to it. If an engine makes guesses in the interest of reducing the amount of hard calculation -- that is a problem. Only by calculating everything in a natural and random way will you consistently end up with a photographic result, because that's how nature works in the real world.

If you give the user the ability to put ridiculous objects or materials in the scene, that does not mean that you have to compromise the light calculation. So, I don't consider those things you mention as 'faking' -- it is also the case that the world is not made of triangle meshes.

[m-Que]

Every time there's a topic about GPU-based renders, I have this feeling of déjà vu. And here's why:

Once upon a time there was a company called Elemental Technologies that came up with a revolutionary video encoding program - Badaboom.
The point - it used the power of GPU for video encoding and promised SPEED: it was promised to be 18x faster than CPU-based implementations.
Two version were announced:
Standard - just simple settings, ease of use;
Pro - lots of settings, flexible, professional encoding tool.
But as the release was delayed over and over again, the company announced it would be releasing only a standard version, potentially adding professional features to it later.
So, time passed by, but basically no new features where added to it. Users begged for simple, yet vital options like subtitles, 6ch audio output, multiple audio tracks etc. All where promised in future updates, and said to be in development, but eventually never been added....

Badaboom was released almost 2 years ago, but it's development is stuck back in 2008. The only thing it's capable of is simple 1-pass video encoding with 2ch audio output. That's it. No subtitles, no color correction, no filters, no multiple pass encoding, no advanced options, etc....
Being tied up to GPU (nvidia CUDA), it simply turned out to be impossible to add all those promised features at this point. (And it's not clear, when it will be).

As for the speed and quality:
It does exactly what promised - encoding is really fast and is done in no time. BUT the quality is useless, since it uses simple 1-pass encoding. Just an example: I took a HD 720p video (300Mb) and encoded it into PSP resolution - that's 480x272. The output video that had the same video quality as the original went up to enormous 934Mb (although the frame size is waaay smaller). When I set the output file size to 300Mb, video became unwatchable.

So, in the end it IS FAST, but what's the point, when result is pretty much useless.
Conclusion: Faster is not always better

[m-Que]

BTW, talking about new technologies, I really like the idea of ArchiCAD's Virtual Building Explorer -
you can render the whole model/environment. Of course, ArchiCAD's render engine is nothing spectacular, but the greatest thing is, that after rendering is done, you can pack it into single .exe file, and then explore it (just like in PC games). Another cool thing is - it's just an .exe file - you don't need ArchiCAD, players, or libraries etc. to run it. It makes communication with a client VERY comfortable.
You can find some samples here:
http://www.graphisoft.com/products/virt ... -explorer/