Intel shows off Larrabee graphics chip for first time

[hakanpersson]

So that's the future of pc game development, attract the users with eye candy.

[mister3d]

I think they couldn't show something worse than that. A mirror-like reflection on the water. :surprised

[Simon]

Intel have no idea.

They are still pushing their crappy 86 technology on everyone and keep telling everyone that raytracing is the future. They have invested too deep to consider a proper SIMD solution.

Raytracing is not the future of games(well not in the next 6 years). Its just not worth the outlay. Gamers don't stop to check that the reflections or shadows are 100% physically accurate. Half the population wouldn't even know how water should really reflect light. You need to remember that all of pixars films up to cars was rasterised, and actually it has a lot more benefits for realtime than raytracing.

[CompanionCube]

better video of it
http://www.youtube.com/watch?v=b5TGA-IE85o

[R10k]

I guess if game devs can easily use it to their advantage, it might become a popular choice. I'm sure this type of tech can be used for a lot more than basic reflections. Time will tell.

[Venkman]

You need to remember that all of pixars films up to cars was rasterised, and actually it has a lot more benefits for realtime than raytracing.

They did say in their presentation that is has support for Direct X, Open GL, and it has "optimized rasterization pipelines" along with ray tracing and something else that I couldn't understand.

[Simon]

They did say in their presentation that is has support for Direct X, Open GL, and it has "optimized rasterization pipelines" along with ray tracing and something else that I couldn't understand.

Indeed but most the rasterisation is done in software. Unlike dedicated hardware such as a GPU that has the functions build in to it in hardware. A middle range gpu has anything up to 300 cores build specifically for purpose and larrabee has probably around 25 general purpose cores. Noone will use it because noone will support it.

Since ati and nvidia have done GPU's so well it is just silly for Intel to try to muscle its way back on to the market.

[Venkman]

Indeed but most the rasterisation is done in software. Unlike dedicated hardware such as a GPU that has the functions build in to it in hardware. A middle range gpu has anything up to 300 cores build specifically for purpose and larrabee has probably around 25 general purpose cores. Noone will use it because noone will support it.

Since ati and nvidia have done GPU's so well it is just silly for Intel to try to muscle its way back on to the market.

Would it be useful for fast rendering engines for animators/movie makers, or are they just trying to get game companies to build tech around their CPUs?

[Per-Anders]

So that's the future of pc game development, attract the users with eye candy.
That's the current situation of PC game development, it's been that way for a very long time, there hasn't been any real gameplay innovation in mainstream PC games since Wolfenstein 3D, no new genres or concepts, just endless FPS and subtle evolution. That doesn't stop modern games from being damn good fun though!

[Simon]

Would it be useful for fast rendering engines for animators/movie makers, or are they just trying to get game companies to build tech around their CPUs?

Theoretically, but if you want something like that then a cell processor would provide what you need. If you want massive parralisation GPGPU is the best bet. The problem is its great providing an awesome new processor but devs have to integrate it into their pipeline. The code bases of the renderers used in vfx are pretty monolithic and ripping them apart to support new unproven tech is a daunting prospect.

Xbox360, Ps3, Pc all use rasterised graphics through opengl or directX. As you can imagine for a cross platform development thats difficult but not too different in terms of architecture and all are used in the same way. I don't think any company will ever code a raytracer in parralel to their traditional rendering pipeline, just for a few pc users who have Larrabees. ...Just like noone ever supported those silly ageia chips.

Edit: That said me and all the researchers I know are itching to get our hands on a few to see if they are any good for anything.

[InTerceptoV]

Intel have no idea.
Raytracing is not the future of games(well not in the next 6 years). Its just not worth the outlay. Gamers don't stop to check that the reflections or shadows are 100% physically accurate. Half the population wouldn't even know how water should really reflect light. You need to remember that all of pixars films up to cars was rasterised, and actually it has a lot more benefits for realtime than raytracing.

Actually, Pixar used raytracing from Bug's Life... they didn't do it in PRMan, but BMRT...

But as you said, it's not that raytracing is gonna be used in games in next 3-5 years, but things are moving forward, and it will hit VFX market first to earn some money for future developement. All those hitech features we have in VFX will be reflected to games sooner or later, it was happening in the past and it will happen in the future...
I just wanted to say go Intel, Caustics Graphics, Nvidia... someone will get it right!

[DrBalthar]

If you want massive parralisation GPGPU is the best bet. The problem is its great providing an awesome new processor but devs have to integrate it into their pipeline. The code bases of the renderers used in vfx are pretty monolithic and ripping them apart to support new unproven tech is a daunting prospect.

Sorry but you talk like someone who obviously never has coded a single line of production code ever in your entire life. And especially not a single line of GPGPU code.


Xbox360, Ps3, Pc all use rasterised graphics through opengl or directX. As you can imagine for a cross platform development thats difficult but not too different in terms of

Dude there is no such thing as real cross platform even if you stick to DirectX or OpenGL. Xbox360 is DX9, PC is DX9 to DX11 (even the lowest common DX9 has spefic different vendor capabilities and OpenGL is mostly completely vendor specific. And PS3 is actually OpenGL|ES


architecture and all are used in the same way. I don't think any company will ever code a raytracer in parralel to their traditional rendering pipeline, just for a few pc users who have Larrabees. ...Just like noone ever supported those silly ageia chips.

I bet with you Larabee will be in PS4 or Xbox720.


Edit: That said me and all the researchers I know are itching to get our hands on a few to see if they are any good for anything.

Obviously you are not important enough otherwise you would have access!

[hakanpersson]

That's the current situation of PC game development, it's been that way for a very long time, there hasn't been any real gameplay innovation in mainstream PC games since Wolfenstein 3D, no new genres or concepts, just endless FPS and subtle evolution. That doesn't stop modern games from being damn good fun though!

You are right, though I also meant how hardware developers are laying out the future for pc gaming.
PhysX releases for cooler physic effects. And now larrabee with what appears to be, also only visual effects that has no impact on the game itself. The idea of providing additional eye candy where the pc is able to, is of course nice. But imo not something that will affect my choices of hardware.

But after my like 8th play-through of baldurs gate 2, Im not a very good spokesman in this matter anyway right?:)

[rebb]

I think they couldn't show something worse than that. A mirror-like reflection on the water. :surprised

It's odd, especially since they had a much better looking water-material in place a year ago.
http://www.tgdaily.com/images/slideshows/200806121/etqw_02_flyby.jpg

Even tho back then it also had this strange look because it's almost a flat surface with really really flat and big "waves". It didn't look convincing back then either, but at least it wasn't mercury.

*nitpick*

[Simon]

Sorry but you talk like someone who obviously never has coded a single line of production code ever in your entire life. And especially not a single line of GPGPU code.


Ok whats with the personal attacks in your post? Seriously not cool.

Dude there is no such thing as real cross platform even if you stick to DirectX or OpenGL. Xbox360 is DX9, PC is DX9 to DX11 (even the lowest common DX9 has spefic different vendor capabilities and OpenGL is mostly completely vendor specific. And PS3 is actually OpenGL|ES


I never said it was cross platform, I said it makes cross platform more difficult. I said their architectures are similar. I need to get vertexes, indices, tangents, normals and attributes and send them to the renderer. Not mess about with raytracing code or program my own shader systems etc... and telling me the ps3 is Opengl|es when I stated that it was Opengl, congratulations its a subset of Opengl stop trolling.

Obviously you are not important enough otherwise you would have access!

Nobody has access you dolt it isn't released to anyone who wont pleasure intels marketing dept.

[Simon]

It's odd, especially since they had a much better looking water-material in place a year ago.
http://www.tgdaily.com/images/slideshows/200806121/etqw_02_flyby.jpg

Even tho back then it also had this strange look because it's almost a flat surface with really really flat and big "waves". It didn't look convincing back then either, but at least it wasn't mercury.

*nitpick*

Their older demos were created with banks of multicore overclocked processors rather than the larrabee. Perhaps they had to reduce the quality to get it realtime on the larrabee... Or more likely the code was botched together in time for the deadline build.

[DrBalthar]

Ok whats with the personal attacks in your post? Seriously not cool.

Simply because I am sick of people obviously writting bullshit! Sorry it makes me a bit aggressive.


and send them to the renderer. Not mess about with raytracing code or program my own shader systems etc... and telling me the ps3 is Opengl|es when I stated that it was Opengl,

Psst a litte of information for you a lot of people in the game industry write their own shader system. Especially if it is multi-platform. Besides a raytracing system works the same way you need vertices, triangles, same geometry information, etc. There is no huge difference at all! Besides building up the acceleration structure which is one additional thing if you not already do something similar for your standard GPU rasterization engine anyway (since you probably need it for collision or physics calculation as well). So only the ray kernel is different and that's a tiny amount of code, compared to the rest.


congratulations its a subset of Opengl stop trolling.

Yes a very limited subset that in reality you end up writting dedicated code paths.



Nobody has access you dolt it isn't released to anyone who wont pleasure intels marketing dept.
As I said if you had important enough projects that would benefit from Larabee and would be visible enough you would!

[Simon]

Simply because I am sick of people obviously writting bullshit! Sorry it makes me a bit aggressive.

I'm not bullshitting. I have written no falsehoods in this thread and have presented my opinions.

As for the shader system yes I've written several. I also happen to be researching the next generation of shader systems with the best games dev in the Uk starting in a few days, so don't try to call my bluff. You would need an entirely new system to handle raytracing. Go pitch that to your manager on an 18 month production schedule.

As I said if you had important enough projects that would benefit from Larabee and would be visible enough you would!

See heres you bullshitting. Noone has larabee. Noone

[Apoclypse]

Locked in 4...3...2...1...

[ngrava]

Hey why don't you guys take this more personal conversation to PM. No need for everyone else to have to read this.

I agree that the demo was pretty silly. All that time building that chip and all they could come up with was a low frame rate locked camera shot? Silly!

Anyway, I thought the point of Larrabee was to make it more like a CPU/GPU combo so you could do more complex programing? Pardon me if I'm wrong about this but I thought GPUs where actually pretty limited in there abilities and that Larrabee was stepping it up so people could more easily do things like complex skinning, volumetrics, scientific simulation stuff, beyond shading and simple vertex programs? Also, as far as cross platform goes, I think this is going to be more about OpenCL drivers along side their DX and OGL ones. I remember reading somewhere about Larrabee and OpenCL. I doubt they would support Cuda. ;)

[705]

but it's just weird though,

not to speak too deep in technical things, but raytracing in gaming is a lot of wasted energy just for something that people wont notice that much (imho). you can use that much horsepower to do something else like physics or ai or else but to calculate "too precise" graphics.

[Kirby158]

Larrabee will be able to render ~120 million voxels per frame. That's what people mean with the word "holodeck".

[ngrava]

Yeah, that does seem pretty silly to me too. However, I have feeling that that's not really what is going to be great about Larrabee. I don't think people will use it for ray tracing as much as other more complex things. By the way, I think Physics and AI is exactly the kind of thing that could be done on larrabee more easily then on a regular GPU.

I think Intel is just a bunch of engineers and marketing people. So to them, merrored reflections, computationally, are pretty impressive. So, I don't think they really understand what is cool about ray tracing on the GPU so it's difficult for them to come up with something that might impress a bunch of CG industry people like us. ;) I don't think we should judge Larrabee on what they showed us. I do think that in a few years we will see things like better ways to simulate GI, real time Area shadows, better looking motion blur and more effects possibilities that we are just seeing the limits of with current GPUs.

[^Lele^]

Larrabee will be able to render ~120 million voxels per frame. That's what people mean with the word "holodeck".
LMAO, sentence of the day for me XD

I heard, badly, about some spanky new acceleration structure, akin to a bsp or similar space partitioning, mentioned in regard to the moving vehicles, but didn't understand much aside from the fact that it's good for RT framerates.
Considering causticRT claims to have invented new structures and methods to go about it, and i believe, unsustained, that most of the new stuff deals with this part rather than the actual tracing, i was wondering if anyone has more infos on this.

[Kirby158]

i was wondering if anyone has more infos on this.

I estimated that because I wrote a voxel renderer in SSE machine language. It renders 6.4 million per frame on Core i7-920 and is extremely optimized. And Larrabee will be ~20 times faster than a Core i7-920.

[Hauzer]

What an awe inspiring demonstration video.

Seriously, couldn't they have at least moved the camera around?

[richcz3]

Intel is in a very big pinch right now.
Desktop and even Laptop sales have been in decline - surging trend is toward Netbook PCs.
The PC Gaming Alliance and Games for Windows Initiative seem to both be sucking wind.
I know that Intel was really pushing Overclocked Quads when they demoed Alan Wake - now it looks like Microsoft is scrapping the PC release of the game.

For Intel to be angling Larrabee at least in part as a PC gaming solution - Yipes :argh:
There is the PC enthusiasts market and high end game PC rig makers, but with so many developers (including MS) opting for console development first or only. This should get interesting.

[CHRiTTeR]

So that's the future of pc game development, attract the users with eye candy.

correction: present, and for quite some time already

[playmesumch00ns]

LMAO, sentence of the day for me XD

I heard, badly, about some spanky new acceleration structure, akin to a bsp or similar space partitioning, mentioned in regard to the moving vehicles, but didn't understand much aside from the fact that it's good for RT framerates.
Considering causticRT claims to have invented new structures and methods to go about it, and i believe, unsustained, that most of the new stuff deals with this part rather than the actual tracing, i was wondering if anyone has more infos on this.

Well at least this demo looks a lot better than the crap Caustic have been putting out :) I would imagine the acceleration structure in the intel demo is something along the lines of a BVH/BIH with some clever rebuild doohickey going on. Caustic haven't really done anything terribly novel for their stuff (at least not from what the patents tell you).

To those of you questioning why raytracing in games is even worth bothering about when most gamers won't even notice the difference the answer is simple - code complexity. Effects that are complex to produce using a rasterization pipeline only are much easier to generate with a raytracing engine. This means the software is quicker to write and easier to debug, hence production costs are reduced. That's something that's pretty attractive for a developer.

Whether it will catch on or not, who knows? I really hope it does. The thought of having a massively multicore, truly general-purpose coprocessor is very exciting.

[wurp]

Indeed, raytracing is also faster to work with for the artists, doing my work I'd have to spend at least twice as long working on a shot if I had to fake everything in scanline...
Even movie production is going more and more towards tracing and I think it's silly to think that the same won't happen to games, it's just a matter of time before the tech is fast enough, I think games will take basically the same path as VFX has taken, just a gradual move to raytrace everything, given how long it has taken for this to happen in VFX (10+ years) I think it's not unreasonable to think it will take almost as long for it to happen with games,


To those of you questioning why raytracing in games is even worth bothering about when most gamers won't even notice the difference the answer is simple - code complexity. Effects that are complex to produce using a rasterization pipeline only are much easier to generate with a raytracing engine. This means the software is quicker to write and easier to debug, hence production costs are reduced. That's something that's pretty attractive for a developer.

Whether it will catch on or not, who knows? I really hope it does. The thought of having a massively multicore, truly general-purpose coprocessor is very exciting.

[CHRiTTeR]

You cant expect realtime raytracing to be instantly super awesome.

If you are going to compare realtime raytracing vs realtime rasterizing then pls keep in mind how long it took for rasterizing to be where they is today.
Look at how quake1 looked back in the day ;)

Give raytracing some time and it will be awesome!!! ;)

Keeping that in mind i think it looks pretty good and verry promessing.
And like intel already said many times; larrabee isnt 'directly' meant for raytracing, its just cool that it is a possibility now!

[Gunnah]

i dunno, this looks pretty nice:

http://www.pcper.com/comments.php?nid=7801


cheers

g

Well at least this demo looks a lot better than the crap Caustic have been putting out :) I would imagine the acceleration structure in the intel demo is something along the lines of a BVH/BIH with some clever rebuild doohickey going on. Caustic haven't really done anything terribly novel for their stuff (at least not from what the patents tell you).

To those of you questioning why raytracing in games is even worth bothering about when most gamers won't even notice the difference the answer is simple - code complexity. Effects that are complex to produce using a rasterization pipeline only are much easier to generate with a raytracing engine. This means the software is quicker to write and easier to debug, hence production costs are reduced. That's something that's pretty attractive for a developer.

Whether it will catch on or not, who knows? I really hope it does. The thought of having a massively multicore, truly general-purpose coprocessor is very exciting.

[shuggie]

Is it possible this will have any uses for offline rendering?

[playmesumch00ns]

i dunno, this looks pretty nice:

http://www.pcper.com/comments.php?nid=7801


cheers

g

I stand corrected!

[CHRiTTeR]

i dunno, this looks pretty nice:

http://www.pcper.com/comments.php?nid=7801


cheers

g

Really impressive and then he added that depth of field at the end! :eek:

[mister3d]

i dunno, this looks pretty nice:
http://www.pcper.com/comments.php?nid=7801

Frankly speaking I would love to have such a card for around 2000$ and I think it would satisfy my needs for something like 5 years. Provided it can do all the effects.

[ngrava]

Wow, that was the best CausticRT demo that I've ever seen. I have to say that the previous ones where not at all interesting to me before. I need to do some further research because I don't understand how Caustic accelerates Brazil. Did Splutter Fish rewrite Brazil to run on the card?

[^Lele^]

Lol, I stood corrected by CP a few times over, and miss it bigtime... :buttrock:
As i was saying, the -seemingly- main issue with production rendering (ie. huge amount and size of textures, loads of individual triangles, huge topologic complexity, and so on) done in Rt is bandwidth, not ray tracing in itself.
In other words, it's moving the data that alters (due to user interaction, or time changes) to and from the (co)processor in an efficient manner.
Which is where the caustic API seems to be extremely efficient at.
I'd love to know what approach the Intel researchers took to the problem, just for kicks.
Oh, and CP, nice link you provided, a true geek article to sink my teeth in, and a video demo to boot! Two thumbs up!

L.

[Gunnah]

heh :) np :) I was kinda excited to see it myself.

cheers dude

cp

Lol, I stood corrected by CP a few times over, and miss it bigtime... :buttrock:
As i was saying, the -seemingly- main issue with production rendering (ie. huge amount and size of textures, loads of individual triangles, huge topologic complexity, and so on) done in Rt is bandwidth, not ray tracing in itself.
In other words, it's moving the data that alters (due to user interaction, or time changes) to and from the (co)processor in an efficient manner.
Which is where the caustic API seems to be extremely efficient at.
I'd love to know what approach the Intel researchers took to the problem, just for kicks.
Oh, and CP, nice link you provided, a true geek article to sink my teeth in, and a video demo to boot! Two thumbs up!

L.

[aglick]

Splutterfish/Brazil was purchased by Caustic.

[PhilipeaNguyen]

...but raytracing in gaming is a lot of wasted energy just for something that people wont notice that much (imho). you can use that much horsepower to do something else like physics or ai or else but to calculate "too precise" graphics.

That's the smart part about RayTracing in Larrabee. Intel is actually using it for Physics collision, surface contact and a couple of other things that you typically wouldn't use RayTracing for, but because they can accelerate the stuff to faster than realtime in hardware they're just trying to get they're full bang for the GGPU cycle buck. I'm pretty sure that's why they bought Project Offset...so they could have a single development platform that was optimized for their hardware to set the pace for development. It'll only work if the first Project Offset title does extremely well though.

[Spacelord]

Wow that Caustics Video is awesome ! I was wondering when they were going to render a decent project.

[DrBalthar]

Larrabee will be able to render ~120 million voxels per frame. That's what people mean with the word "holodeck".

Hmm that's not too much since today GPU solution can already render > 1 Billion voxels per frame. Well also depending at what resolution you're talking of.

Btw that Caustic Video rocked now I am convinced that they actually have something worth considering.

[CHRiTTeR]

Hmm that's not too much since today GPU solution can already render > 1 Billion voxels per frame. Well also depending at what resolution you're talking of.

Btw that Caustic Video rocked now I am convinced that they actually have something worth considering.

yeah, but i hope prices will go down eventually

[gruhn]

The cube root of 120,000,000 isn't even 500.

[CHRiTTeR]

I wonder how much this will accelerate an off line renderer, since it uses x86 there shouldn't be much trouble in supporting it, right?

[TwinSnakes]

The Larrabee video is so lame. The CausticRT video is really nice showing Brazil on it. So was ChaosGroup and their VrayRT GPU video. Right now, I give the edge to Larrabee.

Once Larrabee's API is published, and we get a better idea of how to take advantage of all those cores in parallel, then we'll get a better "picture" of what this card can do.

I'd love to see NextLimit's Maxwell or RealFlow running on a couple of these Larrabee cards. Even better would be an open source unbiased engine that runs on Larrabee.

Fun times ahead indeed.

[Teyon]

Gamers don't stop to check that the reflections or shadows are 100% physically accurate. Half the population wouldn't even know how water should really reflect light.

Actually, a few pages through GameTrailers.com and you'd think different.

[sekelela]

This is the first time my laptop runs even faster than my desktop computer thanks to the latest intel multicore processors, this larrabe presentation is so clumpsy, what about shader layers? and hardware accelerated physics?

Intel crew, sure you can do better

[CHRiTTeR]

*edit: oops sry it had nothing to do with larrabee

but i see no reason larrabee couldnt be used for physics :)

[DrBalthar]

*edit: oops sry it had nothing to do with larrabee

but i see no reason larrabee couldnt be used for physics :)
Well anyone who had read the Intel Whitepapers on Larrabee would know that, because there is even one specific on that topic

[darkZ]

I wonder how a system would perform with an intel card and one from say nvidia. We could use the intel chip for control based instructions and ones which aren't inherently SIMD whereas the nvidia card for all those instructions which need to be run on lot of processors for different but similar data. Using the 24 (?) gp processors on the intel card we can perform all the insructions not suited for SIMD if they are parallelizable.

[InfernalDarkness]

Whether it will catch on or not, who knows? I really hope it does. The thought of having a massively multicore, truly general-purpose coprocessor is very exciting.

Indeed, and AMD's "Fusion" is just that. Seems like AMD and Intel are at it again, only Intel's on its own, and trying to compete with the two Big Dogs in the market (Nvidia and ATI) seems a bit silly.

Also, Nvidia's OptiX is almost in place already, using existing hardware. Good luck trying to slay that one, Intel...

But I'll admit my negative bias towards Intel openly; for all the other reasons combined, on TOP of those my own father recently quit Microsoft (working on Windows 7) to go work on Larrabee. So the things I see standing in the way, in order of personal importance:

1. My dad's a traitor and deserves to be harshly punished for leaving his family to go work on some preposterous piece of peripheral tech. He literally divorced his wife to take that job. I hope the worst for him; god himself has forsaken you, Pops.

2. Larrabee is the third-stupidest Intel nomenclature, preceeded by "i7" and the "Pentium 2-4" series' name-ology. Pentium 2? You can't name it "Hexium" and follow your own number structure? Good work, you unimaginative, uninspired cash chuckers.

3. OptiX and/or Fusion. Normally, this would be the most important factor, but I simply can't afford to miss an opportunity to take blind stabs at my dad, even though he'll never read them.

4. Quake Wars? Seriously? It's like Intel never even heard of Unreal. If Unreal isn't on board for the so-called "Larrabee", then Intel's in for a big shock when Unreal 4 arrives.

5. This list is just a pile of spew; feel free to ignore it, mock me, or otherwise disregard my pointless opinion, my friends.

[CHRiTTeR]

i really dont care for who wins, all i know is this war is good for us and it was high time something like this happened again.

[Eru]

Larraby laughaby...
That demo was a laugphing stock...
Nvidia's G300 is gonna be a MIMD supercumputer, able to run C++ code natively right out of the box... (don't belive me, check for yourselfs)

And as for ray-tracing? 2 things:
1. Current GPUs are running ray-racing algorithms in real-time, for some time now... Anyonw who ever turned on AO in 3dsmax 2010 has been running rt on his/her gpu (ambient occlusion in a spherical harmonics algorithm that has inherent ray-tracing in it)
2. There wont be any real-time advanced ray-tracing in the near future, as the technological barier preventing it from happenning has nothing to do with computational hardware: As I've recently read someone stating in another forum here about mental-ray, It's the memory latency that's the issue, not even it's bandwidth... And that one is not only not improving in the last decade, it's actually getting worse... (ddr3's latency is worse than ddr2, which is worse thatn ddr1, and its the same for ddr4 & ddr5)

The problem is not in raytracing per-say, but in building the entier rendering engine around raytracing, and using advanced methods of it - i.e. higher-order rays (secondary rays, etc.)
It's true that raytracing can lead to much more elegant and simple code, but wheen comparing it with rasterizing, the bottom line is this:
In raytracing, it is simple to do complex things but complex to make them run fast.
In rasterizing, it is simple to run things fast, but complex to make it do complex things.

The problem of rendering the outcome of secondary rays is that it requiers random access to everything in a rather unpredictable manner, which is a memory access problem, not a computational one, which makes the whole debate on GPU vs CPU irrelevant.
In a month or so, G300 is going to catch up (or so Nvidia would have us believe) with x86 architecture. Larraby is not that different, only it's going to have 32 processors, whereas G300 would have 512 of them.

They both would still not be able to do complex raytracing in real-time, as again, the problem is not in the processor, it's elswhere. Realtime AO, first order reflections (of a fairly static scene and camera, and a fairly flat water mesh), or even "interactive" low-res GI, would not convince me otherwise...

[CHRiTTeR]

Its larrabee not larraby. Someone who doesnt spell the name correctly doesnt sound verry convincing.

The memory problem is a well know problem, if it is really impossible i doubt they would've taken this risk and throw so many money at it.

If RT raytracing is really useless and a waste of resources then why is nvidia also going for it?

Its not about having instant full blown raytracing like we know from off line rendering with all its bells and whistles at realtime framerates.
Its about starting to make 'simple' raytracing available at usable/realtime speeds and build the technologie up from there. You got to start somewhere.

I dont know much about nvidia's upcomming g300 but it having more cores doesnt mean anything. Nvidia doesnt use the same cores and design as intel and maybe 32 intel cores are more powerfull and/or efficient then 512 nvidia cores. Both will probably have their pro's and cons.
Dont forget that more cores also means its harder to divide the tasks up efficiently.

Larrabee isnt meant as a raytracing accelerator. It can also do rasterisation and many other things. Its just nice it is possible to do 'some' raytracing in a way that is starting to become usable.
Im sure we'll see a lot of hybrid solutions before we get fullblown realtime raytracing

[dmaher]

I’d suggest we have to give the Larrabee demo a bit more slack.
Clearly it uses crap art.
And you know engineers are notorious for bad art in their demos ;)
The message here is that they’re using realtime raytracing…a good artist could do a lot with that.

BTW, any time someone mentions “eye candy” I roll my eyes.
“Half the population wouldn't even know how water should really reflect light.”
It’s simply not true.

True the average person can’t describe it or recreate it…but they can certainly pick it out of a lineup.
“That’s real”,” That’s CG”; the average person is very good at that game.
But, with every passing year we fool them more often…and people like that.

And that doesn’t diminish the importance of game play design…why would it?
Don’t buy into the idea Art and Design are at odds.

[InfernalDarkness]

Don’t buy into the idea Art and Design are at odds.

Well-said, my friend. Doesn't change the fact that my dad's a backstabbing traitor, and Intel's hired him to work on this steaming pile.

That said, at least they got someone intelligent. The guy did spawn my own sorry self! (grins) And I suppose "Larrabee" is a better name for a GPU than "Divorcer-Betrayer" would be, after all. Which is what I call it in my head.

To Chris:

Its larrabee not larraby. Someone who doesnt spell the name correctly doesnt sound verry convincing.

That would be "Larrabee." Names start with capital letters, my friend! That way it will sound more "verry convincing". (nods)

[DanielWray]

:shrug:

Anyway, It doesn't matter which way you look at it, close to real-time ray tracing is coming and we can all benefit from it, I'd love to have a machine with either larrabee or Nvidia's GPU 3x series in it, just as long as the software makes fairly good use of it.

[Eru]

Its larrabee not larraby. Someone who doesnt spell the name correctly doesnt sound verry convincing.

You got me there, my bad. But I'm actually disappointed by this personal attack. I thoght people here are intelegent, and would know better then to attack an argument by discrediting the argumentor through a puny spelling error detection... Immature...
(I'm also sure you could easially find much worse spelling in the forum then mine, especially from israelians...:D)

Personlly, I couldn't care less what it's called, it just served the pun there...

I agree that it's only a start, but the attempt to market it as anything more so (as intel is clearly trying to do), is loughable... That was my point.

As for Nvidia, they are clearly pushing any GPGPU advancement, and have better experience and personell for that, especially for rt-rt (real-time ray-tracing, has a nice ring to it, dosn't it? ;) If you consider their past course of actions: Buying Ageia, then Mental-Images, after toying with gelato, and now are developing the integration of iRay (accelerated interactive ray-tracing of mental ray) to mr 3.8 win an attempt to run it on current Tesla, AND g300 series. As for g300 vs Laughabee, I know enough on both architectures to know intel doesn't stand a chance in that one... (and yes, 512 doesn't necessarily means faster then 32, but in this case it does, their clock is going to be in the gigahertz scale, and Nvidia has much more experience in parallel computing efficiencies...)

Either way you toss it around, it feals as if Nvidia is much better suited to tackle rt-rt then intel... Though both of them are currently 'falsly' proclaiming contemporary impossibilities as near-future facts... I'm just calling their bluff.

But Intel's apparent unbased cockyness in this one is bordering on the embarrasing side...

[CHRiTTeR]

You got me there, my bad. But I'm actually disappointed by this personal attack. I thoght people here are intelegent, and would know better then to attack an argument by discrediting the argumentor through a puny spelling error detection... Immature...
(I'm also sure you could easially find much worse spelling in the forum then mine, especially from israelians...:D)

Personlly, I couldn't care less what it's called, it just served the pun there...

I agree that it's only a start, but the attempt to market it as anything more so (as intel is clearly trying to do), is loughable... That was my point.

As for Nvidia, they are clearly pushing any GPGPU advancement, and have better experience and personell for that, especially for rt-rt (real-time ray-tracing, has a nice ring to it, dosn't it? ;) If you consider their past course of actions: Buying Ageia, then Mental-Images, after toying with gelato, and now are developing the integration of iRay (accelerated interactive ray-tracing of mental ray) to mr 3.8 win an attempt to run it on current Tesla, AND g300 series. As for g300 vs Laughabee, I know enough on both architectures to know intel doesn't stand a chance in that one... (and yes, 512 doesn't necessarily means faster then 32, but in this case it does, their clock is going to be in the gigahertz scale, and Nvidia has much more experience in parallel computing efficiencies...)

Either way you toss it around, it feals as if Nvidia is much better suited to tackle rt-rt then intel... Though both of them are currently 'falsly' proclaiming contemporary impossibilities as near-future facts... I'm just calling their bluff.

But Intel's apparent unbased cockyness in this one is bordering on the embarrasing side...

So you think your post was mature?
Changing larrabee into laughabee... ok...


Anyway...
That those 512 cores are running in the GHz scale is just a rumour.
And the core speed is rumoured to be 700/1600/1100 MHz (Chip/Shader/Memory).

But those are no official numbers.

Lets presume these numbers are correct, then you still cant know its better or worse than larrabee.

I always thought 'intelligent' ppl used facts to base their opinions on, not gossip.

Other important things beside performance of a single unit are powerconsumption and price off course.
If one performs 2x better than the other but it consumes 1.5x the power and costs 3x as much than, in my opinion it isnt really better.

But we'll see how it goes... intresting times!

*edit: ok, the 512 cores seem to be official now.

[Eru]

So you think your post was mature?
I AM trying, where exactly was it not?
The G300 having 512 cores is only a rumour
That those presumably 512 core are running in the GHz scale is also just a rumour.
I think you should revisit your assumptions:

http://www.guru3d.com/news/nvidia-gt300s-fermi-architecture-unveiled/ (http://forums.cgsociety.org/%20nVidia%20GT300%27s%20Fermi%20architecture%20unveiled)
http://www.guru3d.com/news/nvidia-fermi-architecturemore-details/ (http://forums.cgsociety.org/%20NVIDIA%20Fermi%20architecture%20more%20details)
http://www.theinquirer.net/inquirer/news/1556946/nvidia-reveals-fermi-gpu-architecture (http://forums.cgsociety.org/Nvidia%20reveals%20Fermi%20GPU%20architecture%20)
http://www.theinquirer.net/inquirer/news/1557197/nvidia-shakes-supercomputer-economics (http://forums.cgsociety.org/Nvidia%20shakes%20up%20supercomputer%20economics%20)
http://www.tomshardware.com/news/nvidia-cuda-gpu-fermi-geforce,8766.html (http://forums.cgsociety.org/Nvidia%20Betting%20its%20CUDA%20GPU%20Future%20With%20%27Fermi%27)

Here are some highlights:









3.0 billion transistors
40nm TSMC
384-bit memory interface
512 shader cores [renamed into CUDA Cores]
32 CUDA cores per Shader Cluster
1MB L1 cache memory [divided into 16KB Cache - Shared Memory]
768KB L2 unified cache memory
Up to 6GB GDDR5 memory
Half Speed IEEE 754 Double Precision

GeForce, Quadro and Tesla chipsets will jump from 240 cores to a much larger 512 and should be much faster in each core courtesy of some industry-first techniques.

I don't thnk these are rumours anymore...

But you're right, I don't know for a fact that g300 would kick-ass larabee, but "intelegent people" do allow themselfs to make educated guesses. And I actually was talking about emotional-intelligence when refering to your personal attacks which you continue to do.
You may attack an argument, but it is immature to personally attack somone you know nothing about. I'm not saying you are imature. I don't know you. I'm saying that the approach you have chosen to attack my argument (implying I shouldn't be taken seriousely as I misspelled Larrabee...) IS immature.

Also, to support my claim that we're no where near to full RTRT, here's a quote from one of the "Caustic" articles:

Caustic isn’t afraid to acknowledge though that we are still many years from seeing a fully ray traced game engine – they estimate a solid 4-5 years will pass before their hardware iterations will be fast enough to handle 60 FPS at resolutions like 1920x1200.


Anoher interesting tidbit in the article, is that the hardware architecture of the CousticOne that was design to accelerate ray-generation and tracing by solving the memory access issue, included a DDR2 memory (which today would be considered out-dated) with only a single precision (64bit), which supports the claim of prefering better latency to bandwidth. This means it would be hard to attempt to solve the poblem with a high-end DDR5 hardware that the g300 (and probably larrabee as well) would have, or even sharing the DDR3 system memory of the core-i7 architecture (should Intel choose to integrate Larrabee in the motherboard)

[Eru]

Here is another testament that intelligent men are basing their claims on predictions based on assumptions and/or roumors - John Carmack on ray-tracing and Larrabee, practically saying similar things that I am - and that's from May 12 2008 mind you:

I’ve been pitching this idea to both NVIDIA and Intel and just everybody about directions as we look toward next generation technologies. But this is one of those aspects where changing the paradigm of rendering from rasterization based approach to a ray casting approach or any other approach is not out of the question but I do think that the direction that Intel is going about it as a conventional ray tracer is unlikely to win out. While you could start doing some real time things that look interesting its always going to be a matter of a quarter the efficiency or a 10th of the efficiency or something like that. Intel of course hopes that they can win by having 4x the raw processing power on their Larrabee versus a conventional GPU, and as we look towards future generations that’s one aspect of how the battle may shape up. Intel has always had process advantage over the GPU vendors and if they are able to have an architecture that has 3-4x the clock rate of the traditional GPU architectures they may be able to soak the significant software architecture deficit by clubbing it with processing power.

John obviously feels that Intel's current stance on using the traditional ray tracing algorithms for gaming isn't likely to win out for the next-generation rendering technology. This obviously isn't great news for Intel since even though John isn't the only game programming genius around, his opinions seem to have more weight than just about anyone else in the industry. John more or less backs Dr. Kirk's view that rasterization has too many innate performance benefits to be taken over by ray tracing UNLESS Intel is successful in stretching the Larrabee architecture to incredible performance levels. John mentions Intel would need something to the effect of a 3-4x clock rate advantage on their design versus conventional GPU clock rates to achieve that goal - and while we can generalize and see that he means Intel would need 3-4x the general purpose computational power over competing products, that means a GPU speed of 3.0 GHz or higher when we use today's AMD and NVIDIA GPUs as a reference.

You can catch the whole article here:
http://www.pcper.com/article.php?aid=532

[DanielWray]

I guess it isn't as clear cut, What if the Nvidia processing (Units? cores?) are a very simple and only allow the processing of certain types of data and/ or in a specific way. What if Intel's 32 chips are very complicated and allow the traditional method of code execution, but have certain bottle knocks in the design, I think there are so many things that could affect the performance, going off specifications isn't enough to say "Well this is the clear winner".

The only way to find out which is better, is to wait until some one/ company produces the software which takes advantage of all it's features. Comparing the chips however is pretty pointless as i think it would be down to the software and algorthms in most cases, as to which were the faster design (I'm not expert, so this is pure speculation).

Anyway, I wasn't impressed with the Intel demo, but i haven't seen an Nvidia demo yet, so i can't really compare what is on offer.

[Eru]

I agree with everything you said, but not in the degree of certainty underlying you claim.
Sure, untill we see optimized working software, running on final optimized hardware of both sides, we can not know for sure. However:
Nothing in life is clear-cut (that's my philosophy anyways) and speculation in itself is no different. Every speculation has a degree of cerainty, measurable verifiability etc.
Ask any manager, dealing with risk assesment calculations, it's an un-precise sciece, but a science non the less, economics would not have been able to exist had it been otherwise.
You gather information, analyze it, and make a prediction. The more info you have, the more precise it is and the better you analyze it, the better and more precise your predictions would be, and the better the chance that you would be right. So, the question is not "can we know for sure who would win?", but rather "do we have enough info, and is it precise enough, and is our analysis intelligent enough to make a solid prediction?"

If I can make a solid prediction, I allow myselft to state it as a verifiable claim. It's not completely certain, but has a high degree of certainty.

I think my prediction in this issue has a solid base, if there are variables of unpredictable traits (first generation MIMD architecture on Nvidia's part, and first x86 GPU on intel's part), but they are only a subset of larger variables (like total calculable GFLOPS, based on processor-count times their clock, even if you only know the "range" of the value for each processor it could be enough to eliminate options) you can come to a conclusion that the impact of the larger more-verifiable variable, would certainly overcome whatever advantage an unknown smaller variable could possibly have, if you can asses with a high degree of certainty that it's range of impact would most probably be negligible in comparison.

Tech Report seems to think that the GPU clock speed will be about 1500MHz, a reasonable frequency target for Fermi's stream processing core and about the same as that of the GeForce GTX 285. If we assume Fermi reaches that speed, its peak throughput for single-precision math would be 1536 GFLOPS

As previously reported, Larrabee x86 cores (each Larrabee core is actually a full x86 core) are based on a modernized dual-issue Pentium design with a short execution pipeline. The chip design was enhanced with a vector processing unit (VPU; 16 32-bit ops per clock)

John claimed his prediction a year and a half ago against contemporary hardware (aka. g200 series, which has 240 processors running ~1.3 ghz), based on the stated 225w tdp and all other g300 specs (40nm and all), you could calculate that the clock would not be lower then g200 but rather slightly higher (you even have a direct statement claiming each g300 processor would be faster then a g200 one, even though there would be more then twice of them, and another specific prediction of gflops). Taking it all in, the clock Larrabee would have to have in order to compete with g300 in pure throughput, is no less then 3ghz. (512*1.5ghz*2[singe-presision-operations per-clock]=1536GFLOPS / 32*16[singe-presision-operations per-clock] = 3ghz). Good luck running 32 cores in a single die of 45nm at 3ghz... Thiey just started doing it with 4 without overclocking, jumping away to 8 times that within a rational thermal envelope is not really feasible, and even if so, it would be pushing it way too far in terms of heat. Also, ATI's existing flagship can push 2700GFLOPS, but is not a complete x86 architecture, so it's not directly comparable, but interesting to see how that would go...

(Yo, CHRiTTeR, how's my spelling? Is it to your satisfaction?? ;))

[InfernalDarkness]

Anyway, I wasn't impressed with the Intel demo, but i haven't seen an Nvidia demo yet, so i can't really compare what is on offer.


Nvidia's OptiX videos, demos, and info:

Information (http://developer.nvidia.com/object/optix-home.html)

Review with Screenshots (http://www.geeks3d.com/20090827/nvidia-optix-demos-available-for-windows/)

Pool video (http://www.youtube.com/watch?v=pt1jKQpcAXM) on YouTube

Fractal realtime (http://www.youtube.com/watch?v=_jD0qSAsvr4) manipulation on YouTube

After the last one, it really seems like Intel just doesn't have any idea what else is going on in the real world.

[DuttyFoot]

those videos were pretty cool. why is it in the pool video you see the balls vibrate when the camera moves in the scene.

[Baothebuff]

those videos were pretty cool. why is it in the pool video you see the balls vibrate when the camera moves in the scene.

It's to get the DOF effect.

[DuttyFoot]

It's to get the DOF effect.

ok, i see what you are saying

[CHRiTTeR]

we've seen realtime raytraced reflective balls for years already now, thats nothing new or impressive.

(Yo, CHRiTTeR, how's my spelling? Is it to your satisfaction?? ;))

Im just going to ingnore you from now on because you just dont want to understand anything of what i or others are trying to say to you anyway.

Trying to provoke me wont help.

You clearly already have it all figured out already that nvidia is going to kick everyone's ass and cant be possibly be beaten. Good for you, go invest all your savings in nvidia stock, if you havent already. I wish you all the luck you can get.

btw, i know that JC interview verry wel. I think you need to re read it and really try to understand what hes saying and when he was saying it.

[Eru]

hhhhhhhhhhhmmmmmm, nVIDIA my love..... :love::love::love:
How I wish the whole wide world would adore thau, as i'th love thie.... May you be blessed, for all eternity...:love:


:p

Actually, according to the predictions, the gt300 is gona suck big-time when compared to ATI's 5870... Especially as far as tesselation is concerned... I just think it would suck much less than Larraee, that's all...
Cool demos though...
Cheers!

[InfernalDarkness]

we've seen realtime raytraced reflective balls for years already now, thats nothing new or impressive.

And an even simpler realtime reflective water surface is impressive? But if you watch that video, it's all the other features they're showing off, not just the reflections.

Not to naysay you out of hand, but if you were going to ignore someone, why would you write a lengthy post to them anyway? Just ignore him if you feel annoyed or threatened, don't lecture him as well. He wasn't trying to provoke you, but it's evident that you took provocation anyway, even while saying you weren't going to do so. Relax. Laugh a bit. Make fun of me, if you need to. Key word is "fun".

And it appears you missed the other info and the other video I posted; not to worry, that's not my work, so I don't mind if you don't want to absorb the extra information. But please don't just insult others, because without hard evidence that Larrabee is going anywhere, all this is just idle speculation anyway. We already know what Nvidia can do; it's yet to be seen if Intel can keep up.

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.