Intel is entering GPGPU market - how it will effect consoles

It clearly says "GPGPU", not GPU. So I don't see how you two are confused.

rexoverbey

No, I didn't edit the title... Maybe you misread it. While you may call it a VPU, Intel has marked it as a GPGPU on it's powerpoint slides.

You can't say one is faster than the other eithout giving some sort of basis. How many objects, and are they dynamic? What kind of acceleration structures are you using? (I assume your "exponentially faster" algorithm is using some sort of acceleration structure, because without them ray-tracing is basically rasterization but performed with every single pixel. And I'd love for you to explain how that is somehow faster). And I hope you're not going to argue that ray-tracing is somehow faster in practical scenarios, given the huge gap between traditional HW-assisted rasterization and any real-time ray-tracing implementation in existance.

Please don't take offense from this, but most computer science departments generally don't exactly have their finger on the pulse of the professional industry. Teufelhuhn

When graphical complexity of 3D scenes intersects with both the efficiency of ray tracing and rasterization, ray tracing from there on out will see an exponential leap in performance in comparision to rasterization. This is what I mean by jumping over the "ray tracing constant". As the ray tracing algorithm itself, which is a linear formula - O(n) - has a very large constant attached to it. Rasterization algoithms however, are of complexity O(n^2). Of course, it's a little more complex then that, but that is generally the case. This does not require any kind of 'acceleration structure' of any kind, seeing as it's theoretically what will happen, which can be seen when looking at the two sets of algorithms side by side.

As for asking me if I would argue if it's faster in pratical scenarios, I would require some clarification as to what you mean? Do I think it is practicle in the future of games. Yes, of course. Today, no. However, as cpu parrellelization continues to become the future movement for computation, and three dimensional scenes are becoming increasingly complex -- rasterization will make less sense moving forward, and a scalable approach for rendering will have to replace it. I'm not saying how soon it will be, however I think the transition will happen sooner then people realize.

I would argue this is precisely why nVidia is pushing SLI as the next forward movement. It's their way to survive in the future, by having each gpu render a section of the scene. However, going forward... this is only going to work so well, until it becomes impractical. Rasterization techniques do not scale well with parrellelization.

I'm not going to take offense as to your comment regarding "most computer science" departments, considering UMass Amherst 's graduate program is ranked as being one of the top in the world. Just about every professor there has multi-million research contracts with the professional industry. Heck, most of the rasterization algorithms are formulated through research groups at universities. Theories / Algorithms starts at the universities, and then somewhere along the road to implementation they move over to corporations.

With that said, the concepts revolving rasterization versus ray tracing doesn't require you to be working for an architect for Intel or Nvidia. The concepts themselves are really not all that difficult -- it's theory, the thing that is difficult is getting things like this implimented.

When graphical complexity of 3D scenes intersects with both the efficiency of ray tracing and rasterization, ray tracing from there on out will see an exponential leap in performance in comparision to rasterization. This is what I mean by jumping over the "ray tracing constant". As the ray tracing algorithm itself, which is a linear formula - O(n) - has a very large constant attached to it. Rasterization algoithms however, are of complexity O(n^2). Of course, it's a little more complex then that, but that is generally the case. This does not require any kind of 'acceleration structure' of any kind, seeing as it's theoretically what will happen, which can be seen when looking at the two sets of algorithms side by side.

rimnet00

As for asking me if I would argue if it's faster in pratical scenarios, I would require some clarification as to what you mean? Do I think it is practicle in the future of games. Yes, of course. Today, no. However, as cpu parrellelization continues to become the future movement for computation, and three dimensional scenes are becoming increasingly complex -- rasterization will make less sense moving forward, and a scalable approach for rendering will have to replace it. I'm not saying how soon it will be, however I think the transition will happen sooner then people realize.

rimnet00

I would argue this is precisely why nVidia is pushing SLI as the next forward movement. It's their way to survive in the future, by having each gpu render a section of the scene. However, going forward... this is only going to work so well, until it becomes impractical. Rasterization techniques do not scale well with parrellelization.

rimnet00

All I see is Nvidia trying to push further into the GPGPU space. The obviously bought Ageia as way to deliver physics through CUDA, and in the future I wouldn't be surpised if they make some acquisitions that allow them to integrate x86 capabilites into their hardware in order to combat Fusion and Larrabee.

the thing that is difficult is getting things like this implimented.

rimnet00

The engineering / CS answer: a long explanation of technologies with logarithmic notation of computational complexity and discussion of efficiency techniques in real-world versus theoretical scenarios.

The business answer: Intel is "throwing down" a term that basically means "we use graphics cards with some general computing" in order to charge a higher-consulting fee. It's the same reason information systems doesn't called rows / colums in databases "rows and columns" or why it's called a "capitol gain" and not "you made some money from stocks" :P.

As cynical as it is to say - a really detailed discussion on techniques, hardware, and the companies involved is essentially meaningless. You want to know whether or not Intel will play a part in the hardware of next-gen consoles? Look at their partnering, and look at their bottom-line costs - can they deliver the hardware-spec in 2011 for less than AMD, IBM, or nVidia?

Further - look at their licensing - will they be willing to turn this thing over to MS / Sony / Nintendo - and is it even going to be the product they want - or are they going to pull an nVidia?

-

Common sense says maybe, if Intel sees money to be made in the console market, and are moving in the direction of higher-performance GPUs, then it's common sense for them to gun for it. If they're just jumping in in say, late 2010 though, I don't know that Intel will play a part in all of the consoles GPU next time around or not - at this point it's pure speculation.

-

Business Answer - Intel will get into consoles if there's money to made, and they can negotiate a deal - GPGPU is just a technical buzzword, it has nothing to do with answering the question of "will Intel be involved heavily in the next-gen console hardware?".

This is bad news for Nvidia and AMD/ATI.

Greedy Intel is gonna monopolize computer hardware.

This has gone way over my head even when it was explained directly to me ....

Make it simple, how does this benefit me?:twisted:

Interesting to see what the future holds.

[QUOTE="mjarantilla"]

[QUOTE="rimnet00"][QUOTE="MrGrimFandango"]Man thats sweet, Ray-tracing, thats the new gameplay of the 21st century right?Senor_Kami

To most PS3/360 gamers, graphics = gameplay. :)

Whoa, take the 360 out of that and add PC. Go in a thread about PC games and all you'll see is people mentioning framerate and polygons per second.

and yet who argues that cod4 on xbox360 is the best console version and fable 2 is better looking than crysis

[QUOTE="rimnet00"]Lastly, once we step over the 'ray tracing constant' problem with fast enough CPUs, GPUs as we think of them today will likely not exist. Teufelhuhn

Intel has been making onboard graphics for years and they have always sucked. I think this is just marketing hype much like the 128-teracore processor.

rexoverbey

No this is not hype, esp. if you knew what their upcoming larabee archecture is capiable of, at least on paper. I'm sure some of that technology is being intergrated.

[QUOTE="rimnet00"]I never realized how vaguely defined GPGPUs really were. More precisely, I was refering to GPGPUs which have dedicated components for doing both linear and vector calculations, as opposed to mapping to across paradigms which is very ineffecient -- ie CPUs doing vector calculations, and GPUs doing hardcore linear computation. Teufelhuhn

Yeah but it's never been very efficent, or have had much use for...this is the first time I think we'll really see it blossem.

People seem to be missing the point here:

Discussing the technical details of GPGPU, rendering techniques, and computational effeciency to "determine" if Intel will be a major graphics hardware player in the next-gen console race is like having a debate on eschatology to find out if your friend is coming with you to church on Sunday.

who cares, any decent gamer plays on a console for the GAMES.razu_gamer2

correction gamers like YOU play on consoles

Intels...got some serious $#!+ going on at their labs...insane processors....and IBM as well (IBM said they have devices that are able to transfer a whole blueray disc of data a second..like you can download full length movies in seconds)anasbouzid

[QUOTE="Teufelhuhn"][QUOTE="rimnet00"]I never realized how vaguely defined GPGPUs really were. More precisely, I was refering to GPGPUs which have dedicated components for doing both linear and vector calculations, as opposed to mapping to across paradigms which is very ineffecient -- ie CPUs doing vector calculations, and GPUs doing hardcore linear computation. Innovazero2000

Yeah but it's never been very efficent, or have had much use for...this is the first time I think we'll really see it blossem.

What people are trying to do now is try to figure out just which kinds of programs would benefit from a GPGPU approach. Currently, the most favored routines involve anything that involves a lot of simple floating-point calculation that benefits from being done in parallel. But like we both said, they're still probing the limits.

1. I'm glad that companies like intel and sony come up with new technology that tries to change/push the industry forward. Not in an unessesary dvd-blu ray way, but in a snes-ps1 way.

2. Alot of people say graphics cant get much better than they are today. But with brand new technology surely someone is going to make a breakthrough with all the r+d that goes into these projects. It will be interesting to see how this goes.