Dunno if its just me, but I see a lot of promise in this architecture. If they get the drivers right...............

It's almost comical, really, to see so many people who ought to know better getting revved up about what is basically a marketing exercise for hardware that is at least 12-18 months away from production...;)

Hopefully, it has occurred to at least some people that anything can happen in a period of 12-18 months in this business, and that such a period is a long, long time in terms of the number of goods and services to be made available by lots and lots of companies aside from Intel, and of course 12-18 months provides Intel with a very adequate time frame to cancel "Larrabee" and offer something else instead. Indeed, such a happenstance would not be new or unique for Intel, as well as many other companies we might name.

Larrabee to me, anyway, doesn't so much resemble a gpu as it does a cpu, which ought to surprise no one because Intel has been primarily in the cpu business for a long, long time. My own inkling of the situation is that Larrabee is Intel's response to AMD's Fusion--and both products at the present time are little more than vaporware, and as such it really doesn't matter what either company claims for either vaporware product, does it? What will matter, of course, is what either product is when it ships, if it ships, regardless of what either company may call it.

One really good thing has happened, though, as the result of this Larrabee 12-18-month pre-release "information," and that is that thankfully nobody is referring to Larrabee anymore as a "discrete gpu" product...;) I guess vaporware parties do have some redeeming value after all!

I heard Dr. Larry Seiler's middle initial is B.

Dr. Larry B., they call him.

/badjoke

So would this mean that the card would not be hardware dependent for future DirectX releases?

Tile-based rendering? Deferred rendering?

Does this sound like we won't be able to use FSAA with this platform, just like the UT engine?

I haven't finished reading the paper yet but how does Larrabee attack the problem of writing parallel programs? We have dual and quad cores soon to be in mainstream PCs but programmers are still not making the most use out of them. Same issue with the PS3 / Cell processor... Still, it sounds very interesting and I hope the momentum keeps building to accelerate the development and the eventual adoption of one or common set of parallel programming languages / methodologies. :-)

things are about to get a lot more interesting.
Not so fast, it's going to take a good while yet before this shapes up in any way.

this is exactly what *every* graphics programmer wants

tile-based rendering?! IIRC the last time we saw anything do tile-base rendering it was the Sega Saturn, and we all know how that went! I'm not so much skeptical of frames/per/second as I am of visual quality. This new render API may produce images significantly different then what we're used to (think the difference between a PS1 and an N64).

It sounds like this Larrabee is practically an entire computer within your computer, too.

And Scott, don't be hard on yourself. You should probably wait until you have a Larrabee in your hands before doing a full writeup anyways. I have a feeling there will be some sort-of last minute to-the-wire change in situation anyways.

An original pentium core running tile based processing.

Can't wait to see the Quake II benchmarks.

forget graphics, this will completely revolutionize the HPC accelerator market. I can't wait to see a completely general purpose teraflop accelerator that doesn't force you to deal with graphics-oriented drivers that are out of place in headless cluster nodes.

With Intel's terrific driver support for their current graphics offerings, Larrabee should be an instant success. Give a pig wings and maybe it can fly??? Larrabee will rise or fall as a GPU on the quality of the software and driver support. And Intel has a deserved abysmal reputation in this direction with all their current graphics products.

For one simple example of many, Intel claims that the 965 Express graphics/core-logic chip-set (widely used in Intel lap-top PCs) is fully compatible with World of Warcraft under Vista, but it still has a major driver issue with WoW - black-screen/driver crash --in spite of Intel's claims to have fixed exactly the same symptoms many months ago.
This crash can easily be stimulated within the game.

See:
http://www.intel.com/support/graphics/sb/CS-025607.htm

The claim here that builds after 1187 do not have the problem is bunk.

Build 1437 and 1472 (the very latest) exhibit the exact same symptoms. However, build number 1332 from Nov 2007 does not appear to have the problem. Seems as if Intel has zero clue as to the root cause and apparently does not check that driver updates do not break previous bug-fixes. Are these the same people you want to trust with supporting your brand-new Larrabee GPU card ??

If Intel will not/cannot fix a show-stopping graphics problem after a year for a very popular game with very low graphics demands and with a pretty simple graphics core, what hope have they with Larrabee ?? Where are they going to instantly hire top-rank graphics software engineers ?

Maybe if Intel helped AMD get out of red ink by buying ATi and thus acquiring ATi's GPU-oriented software gurus, they would stand a chance. As it is, I expect that Larrabee will struggle for at least a year or two trying to gain traction in the GPU domain, not only in the development of support software, but also convincing skeptical developers - regardless of the quality and versatility of its hardware architecture.

Remember that for Larrabee to gain any traction at all as a consumer-GPU, it will have to support both current AND LEGACY games just as successfully as either ATi's or nVidia's GPU offerings. For anybody that reads the nVidia or ATi driver release notes it is pretty obvious that a huge amount of best-effort is put into ensuring that new driver releases do not break the backward-compatibility of previous drivers - and even with that best effort, things occasionally do go astray in backward compatibility. Larrabee will have to deal with backward-compatibility starting from ground-zero. Hopefully the Larrabee development team are buying copies of hundreds of different current and legacy PC games right now and a bunch of AMD-based PCs, just to cover all their test/QC bases as a neutral GPU supplier.

Whatever about GPU success, I am sure that Larrabee will have some significant and quick success in the CPU domain, such as being an effective alternate to the Cell processor

aaaaaaaaaaaaaaaaaannndddddd here comes the ACM paper for those who are interested. Grab it while its hot.

http://www.box.net/shared/egn2hnmw4s

while I'm not a compsci person I found a couple interesting things.

I thought that anand had massacred their block diagrams but it turns out they pulled them straight from this paper.

*Rough/ackward performance estimate # of cores needed to obtain 60 fps: 8 for HL2ep2, 15ish for FEAR, 19 GoW (1600*1200)
*"Although P-threads is a powerful thread programming API, its
thread creation and thread switching costs may be too high for
some application threading. To amortize such costs, Larrabee
Native provides a task scheduling API based on a light weight
distributed task stealing scheduler [Blumofe et al. 1996]"
*The paper also pointed out that unlike nvidia's current offering larabee can directly share memory between processors rather than writing out to the main memory. This is accomplished transparently by cache coherency proticals.
*another unique feature "Larrabee can support order independent transparency (OIT) with no additional dedicated logic by storing multiple translucent surfaces in a per-pixel spatial data structure."
*Improved shadow mapping in the form of irregular shadow mapping. gets rid of artifacts in shadows and can also be used to implement hard shadows.
*performance will scale linearily with the number of cores in fear HL2 and GoW.

Who cares how it runs? If it's anything vaguely competitive, and manages to remain relevant for much of any time at all, it's sure to change the game.

I'm sure Intel hopes this catches on, cause it's a very very small step to go from lots of mini-x86 cores on a 'graphics card' doing the rendering to having lots of not-mini x86 cores on the CPU doing the rendering, and having a simple brainless VRAM/framebuffer/DACs/etc on a card just doing the actual outputting.

Gamers are a tough crowd to please. A well architected chip means little, it all comes down to frames per second, eye candy, and drivers that don't crash. I suspect Intels first go aint gonna fair so well against the competition but hell, it'lll be an amusing ride no matter the outcome.

Larabee, if executed correctly, will hurt Nvidia by leaps and bounds.

Larabee is what Intel sees as the vision of computing and they didn't want to get "left behind" when AMD bought ATI. AMD is already "leaking" what their first hybrid chip will be, Intel is already behind on this platform.

What I like about this is how Intel has learned from 3Dfx and Nvidia's failed use of custom api's.

We saw glide shine and burn out....

Nvida tried this once with their old Cg custom api, and again with "tesla" and Cuda...

True, "Failed" may not be the word... just very specific in their niche uses.....

But this at least is promising to me in that rather than bundle the computing power with a very expensive/limited use video card, they can bundle this with a modified cpu. With x86 as the target core, and adding pipes to handle other tasks its like Cell+Cg+Tesla with a much wider market.. and alot easier to code for.

I have hope this launches well.... I'd like to see this succeed long enough to get passed the 1st couple generations to mature.