Any details on AMD Leo demo?

[jlippo]

(back to topic.. )
Just a thought, but wouldn't this technique be nice for a transparent pass when using deferred rendering?
First get max Z in tile to cull lights, then store min-max depth of each light in tile to use during forward rendering pass?

Method might also be nice for Vita as you render in tiles rendering already, to know which lights are visible in each tile might be more worthwhile than going for a fully deferred renderer.

 

[Rodéric]

MMhhh talking about MSAA, did anyone render at a quarter resolution with MSAA (rotated grid) then use each sample as a pixel ?
Just wondering what it would look like in motion.

 

[Davros]

How would someone attempt that
i did the 640x480 4x ssaa scaled up to 1680x1050 and it just looked ok-ish

 

[Andrew Lauritzen]

I think Roderic is suggesting not SSAA -> resolve -> upsample, but rather a direct upsample from the SSAA subsample data.

I'm pretty sure this would look bad though, even if you could figure out a way to map the subsamples to pixels (they are not a grid after all). You really need some sort of sample-position-aware resampling filter in there, but definitely you could do a better job with a filter designed for that task rather than the naive resolve -> upsample scheme (like what Davros did).

 

[Albuquerque]

You really need some sort of sample-position-aware resampling filter in there, but definitely you could do a better job with a filter designed for that task rather than the naive resolve -> upsample scheme (like what Davros did).

While I do not have the technical prowess or even necessary background to describe it like you, that's what I was simplistically trying to convey here:

I guess it could be more interesting if you somehow muxed the upscale and the MSAA resolve into the same pass somehow, or at least allowed them to be closely intertwined.

Having the MSAA filter being an integral component to the upscale could provide interesting opportunities in both performance and quality of the result. My later ramblings about non-linear space was a separate thought-thread on perhaps doing selective scaling (both up AND down) for low detail / high detail inter-scene data.

The mixed MSAA + upscale piece seems far more intuitive and/or simple to a layman like me versus the computation that would surround selective bilinear upscale / downscale to manage quality of various pieces of a given frame.

 

[Andrew Lauritzen]

The mixed MSAA + upscale piece seems far more intuitive and/or simple to a layman like me versus the computation that would surround selective bilinear upscale / downscale to manage quality of various pieces of a given frame.

Well they are orthogonal concepts really. Separating out terms and evaluating them at different frequencies you have to do anyways - in fact games do this all the time today, especially for particle effects and lighting. Typically these are upsampled using some kind of *bilateral* filter that is aware of geometric discontinuities.

But yeah, if you have a a non-uniform sampling of the scene (MSAA/SSAA) and you need to upsampling that, it definitely makes more sense to do something clever and take into account where those samples come from in the pixels rather than average them all first, then upsample the result.

 

[Rodéric]

Oh
Even with a nice filter ?

And yes, games have been using quarter resolution particle buffers and bilateral filtering since 2008 at least. (Maybe a year or two earlier.)
We did it in Age of Conan and it was released in 2008.

 

[Andrew Lauritzen]

Oh
Even with a nice filter ?

Not sure what you're referring to, but I agree with a good filter that takes into account the sample locations you could do something much better than just resolving. I was just saying if you did something naive like map sample 0 to pixel x,y, sample 1 to pixel x+1,y, etc. it would look bad in motion

 

[Rodéric]

Not sure what you're referring to, but I agree with a good filter that takes into account the sample locations you could do something much better than just resolving. I was just saying if you did something naive like map sample 0 to pixel x,y, sample 1 to pixel x+1,y, etc. it would look bad in motion

It had the benefit of being dead simple and cheap ^^
Try simple, iterate...

 

[steampoweredgod]

In this video we see that pixelation can be done away via a fractal algorithmic analysis of the picture

INFINITE resolution within finite definition.

 

[rpg.314]

Not practical for real time.

 

[3dcgi]

I didn't watch the entire video so I don't know why this can't be realtime, but what I did notice is this is technology from the 1990s. The computers in this video are ancient.

 

[rpg.314]

I guess if this tech was any good, it would in use by now.

 

[steampoweredgod]

Not practical for real time.

Ponder

The brain compresses information from 250M optic sensors into 3M sensory fibers with binary output, this in turn fills a memory array of about a billion cells.

At the same time the information can only move in a finite chain of about 50 transitions from cell to cell in the fraction of a second it takes for a human to react. A human can make finite discriminations in fractions of a second, yet the component neurons operate at 10s of hzs, such that this is the limit of information exchange and propagation, and there is noticeable delay even between the eye and brain, and even between different parts of the brain.

It is said that the basis of intelligence is prediction capability, yet you can close your eyes jump back spin and look at an arbitrary object and you will perceive it insanely fast(and note that only a dime size at arms length area enters at high resolution the rest is low resolution input, and yet the image is quickly reconstructed).

In fact humans seem to react in realtime to very complex input with few cycles between such computations. If we assume classical computation, the embodiment of a fractal simulation of the sensory surface such that it's evolution merely has to be corrected via the external input from time to time, would explains its uncanny abilities.

It would also explain why the brain is extremely active at all times even while not receiving any sensory input, and why it can make a CONSCIOUSLY distinct state in response to any arbitrary stimulation of the sensory surface in a fraction of a second(any frame from any movie, etc).

Also connects with the Integrated information theory of consciousness based on discriminatory capability.

What would be more integrated then a fractal system? It embodies infinite information.

I guess if this tech was any good, it would in use by now.

In this world sometimes results are lost or become obscure or used in a limited way only to be rediscovered later, and new applications applied.

The guy got a 2.5+M grant and it appears to being used in some industries.

Also from another nice documentary fractal processes are also used as an extremely efficient receiver and transmitter, not only can it compress information efficiently it can transmit and receive it very efficiently, theoretically optimal. They're also used for effects in cg industry, not sure if they've been tried for Antialiasing applications.

 

[rpg.314]

Ponder

The brain compresses information from 250M optic sensors into 3M sensory fibers with binary output, this in turn fills a memory array of about a billion cells.

At the same time the information can only move in a finite chain of about 50 transitions from cell to cell in the fraction of a second it takes for a human to react. A human can make finite discriminations in fractions of a second, yet the component neurons operate at 10s of hzs, such that this is the limit of information exchange and propagation, and there is noticeable delay even between the eye and brain, and even between different parts of the brain.

It is said that the basis of intelligence is prediction capability, yet you can close your eyes jump back spin and look at an arbitrary object and you will perceive it insanely fast(and note that only a dime size at arms length area enters at high resolution the rest is low resolution input, and yet the image is quickly reconstructed).

In fact humans seem to react in realtime to very complex input with few cycles between such computations. If we assume classical computation, the embodiment of a fractal simulation of the sensory surface such that it's evolution merely has to be corrected via the external input from time to time, would explains its uncanny abilities.

It would also explain why the brain is extremely active at all times even while not receiving any sensory input, and why it can make a CONSCIOUSLY distinct state in response to any arbitrary stimulation of the sensory surface in a fraction of a second(any frame from any movie, etc).

Also connects with the Integrated information theory of consciousness based on discriminatory capability.

What would be more integrated then a fractal system? It embodies infinite information.

I wasn't dissing fractal comression as such. Just for real time purposes.

In this world sometimes results are lost or become obscure or used in a limited way only to be rediscovered later, and new applications applied.

Hardly.

The guy got a 2.5+M grant and it appears to being used in industry.

Where?

 

[steampoweredgod]

I wasn't dissing fractal comression as such. Just for real time purposes.

Realtime purposes depends on computational capacity and memory(also on the state of algorithms used, for example if something similar to hash-life-like exists for arbitrary fractals it may enable extremely rapid transitions with minimal computational requirements, but the complexity of finding and implementing such may take time for its discovery... keep in mind that the game of life is turing complete and as such can implement any computation and be ridiculously accelerated, though it would likely take quite some memory.). The fact that these make for extreme efficient compression rules, emitters and receivers implies it may have efficiency in data representation and processing.

The brain along with the body is defined in less than 700MB of digital information in the DNA, it appears to be generated in a fractal way and to process information in such a way.

The fact that its components appear to transmit information so slow and yet have no problem functioning in realtime, suggest that a hypothetical theoretically optimal solution may exist for optimal data representation and processing that might be extremely responsive to events.

The 6,366,548,773,467,669,985,195,496,000 (6 octillion) generation of a very complicated Game of Life pattern computed in less than 30 seconds on an Intel Core Duo 2GHz CPU using hashlife in Golly. Computed by detecting a repeating cycle in the pattern, and skipping ahead to any requested generation.-wiki

The connection between fractals and cellular automata is also very interesting, it may in theory be possible to define arbitrary mathematical objects with arbitrary precision via the combination of such in a finite space(that can be evolved at an accelerated rate).

All things can be defined in mathematical terms, such as a face, a smile, even the motions at arbitrary resolution and with arbitrarily many intervening frames. A characters movements, the movements of a fluid or cloth. A fractal description may allow for infinite spatiotemporal precision while keeping the representation size finite.

Hardly.

Where?

They did mention several companies involved and the goverment in the video, though I've not looked into it.

 

[steampoweredgod]

More intriguing are the implications if the hypothetical time crystal can be built

He proposes that matter could form a "time crystal," whose structure would repeat periodically, as with an ordinary crystal, but in time rather than in space. Such a crystal would represent a previously unknown state of matter and might have arisen as the very early universe cooled, losing its primordial symmetries.-sciam

"The papers themselves are perfectly respectable, undoubtedly correct, and interesting," says cosmologist Sean Carroll of the California Institute of Technology.sciam

But at very low or minimum energies, most materials can't retain that symmetry, and they crystallize. The regular geometric pattern of a crystal lacks complete spatial symmetry; the structure does not look the same everyplace. Because crystals have less symmetry than before, physicists say they exhibit spontaneous symmetry breaking. Equivalent processes occur in many domains of physics. -sciam

Wilczek says he started wondering whether the concept of an ordinary three-dimensional crystal could be extended to four dimensions, with the extra dimension that of time. A time crystal would spontaneously break what Wilczek calls "the mother of all symmetries"—the symmetry of time translation, which holds physical laws remains the same regardless of what time it is. A time crystal would change with time but keep coming back to the same form it began with, like a clock whose moving hands periodically return to their original positions.-sciam

He reckons a hypothetical device he calls a time crystal could power a computer that would keep on running long after everything else has succumbed to the pull of entropy.

"It's not the most immediate problem in the world, but the point is, we don't have to take the heat death of the universe lying down," Wilczek says. "We can put up a pretty good fight for a very long time." Such a device might even be able to simulate someone's brain - giving a form of life extension.

The most complex still life of all, even spontaneous change in the pattern reconstitutes the pattern after an indefinite arbitrary set number of states(random pertubations ensure restoration from any sequence towards original state)

In theory if the universe resembles a cellular automata, change from the outside would not make it impossible for the pattern to self restore.

Crucially, such a computer should be perfectly placed to survive the heat death of the universe. For one, it would already be in its lowest energy state, so it wouldn't need extra energy input to keep on running. "It doesn't need a power source," Wilczek says. To boot, superconductors tend to work only at temperatures close to absolute zero. So as the universe gets colder and darker, and all other processes cease, this computer would simply get more comfortable. "It could dodge the heat death of the universe for a very long time," he says.

Some doubt there would be any point to such a device. "It's probably not revolutionary," says Craig Hogan, a physicist at Fermilab in Batavia, Illinois.

It's not hard to dream up some possibilities, though. Like an ultimate time capsule, Wilczek's computer could store information about the universe's former life, the only legacy of what the universe once was. More ambitiously, if the right science and technology ever becomes available, humans might upload their brains to the computer, in a sense living on while the universe outside dissolves.

It wouldn't be much of a life. As the computer wouldn't be able to accept new inputs, it would run the same brain-simulating program on a loop. "My best idea of what we'd want to do is to find some kind of experience that we really like, and program it in, so that experience would happen over and over again," says Wilczek.

Milan Cirkovic, a physicist at the Institute for Ethics and Emerging Technologies, doesn't relish that repetition. Luckily for him, others have come up with some equally wacky alternatives.-time crystal computer

A fixed program with no additional input would be no problem if it is a fractal pattern that navigates all possible sequences in a meaningful and ordered manner, the evolution would be infinitely branched..

 

[steampoweredgod]

I would also say that potentially greater possibilities also emerge:

The fractal-cellular paradigm appears to potentially have the ability to do away with the combinatorial explosion in any domain.

And it seems extremely suggestive that such paradigm could be used to instantiate an event responsive structure of infinitely recursive logical order in symbolic combinations. Hypersymbolics or Richsymbolics emerges, rather than a mere letter or word a fractal description of spatial and temporal properties is used and a logic to combine such into a larger fractal representations. In other words a language of thought.

One has to remember that a word is essentially meaningless except in the way it relates to other words and patterns. In the right context it could essentially mean anything.

 

[Davros]

Going from improving the quality of the amd leo demo to building a computer that can withstand the death of the universe.
Are we getting a little carried away

 

[humbertklyka]

To continue the discussion regarding sampling patterns and reconstructions

But yeah, if you have a a non-uniform sampling of the scene (MSAA/SSAA) and you need to upsampling that, it definitely makes more sense to do something clever and take into account where those samples come from in the pixels rather than average them all first, then upsample the result.

I would think that expecially non photo-realistic rendering could benefit from very non- uniform sampling, e.g from say 4 - 1/9 samples per pixel if you have say detailed edges but fairly flat shading on surfaces.

The idea of the fractal guesswork on the other hand seems like quite bad, since that would introduce artifacts that would be hard to predict, even if it could work out to a better match for specific test cases.