Larrabee might not be that great, analyst reckons

I am awaiting the Larrabee, it is finally something console suitable that hopefully will cover many processing needs. But is there a much better way.

Why not a ARM based solution.

Parallel processing solutions on something of the level of a ARM can fit more units in, and more processing power. Nvidia could easily gain a license, and build around the ARM to make a one solution does all. But there are two realities. That they could make an even better general processing solution based around their own designs or others. That more processing power for the average task can be gotten in the same space by using a mixture of specialised cores with general purpose ones.

ARM was a natural successor to the Intel PC processor, it started out the best (apart from the Novix Forth based chip) the worlds top performer. But Acorn dropped their head to head ambitions with IBM, Apple gave it a PDA role and went with the Power PC, and Intel was still spoiling back then with low power embedded nearly PC processors (sound familiar) sacrificing superior quality for PC compatibility (and those processors did not go very far). Now Intel is PC processor mad again, when ARM could have delivered more for more battery life. The things not in the ARM's favour are the increased complexity of parallel programming more cores, and the complexity ramp-up of maximising the potential of the ARM and taking advantage of any space freed up. History can repeat itself, or we can go better ways.

Like the Atom CPU, whit Larrabee technology don't look at the first gen but second ore even the fourth.

Larrabee is not only a GPU but more a GP-GPU

There is no guarantee that it will succeed but Intel is a hell of a compiler writer and if a 22nm 200 core Pentium 1 like design is out there maybe GP-GPU will smoke ati en nv design

I'm sure Nvidia know more about the graphics pipeline than Intel.

Likewise I'm sure Intel know more about X86 (which larrabee is) than Nvidia.

But I'm absolutely certain they both know more about computer component design than investment banker Mr Kumar who should stick to counting his (or others) dollars.

that guy Kumar seems to be one of those nvidia fanbois.

"high efficiency of Nvidia's existing designs".

Wot? He must have efficiency confused with heat generation, or die size, or cost. Those are the things Nvidia has "high" on the designs.

We shall see, but I wouldn't expect it to outperform discrete GPU's. Performance is limited to the slowest component. In this case, it probably will be the memory subsystem. It will be 128-bit shared dual channel bus that is shared between the CPU and GPU elements. Simply put, since the new mid-range is a 256-bit bus (9600GT), there is simply no way it is going to outperform a discrete GPU. Unless Intel is using a 4+ channel XDR2, which I highly doubt since my impression once again is that this is a low end part.

Larrabee is aimed at the IGP market where performance isn't really a factor but power consumption and basic SOHO functionality is required. One problem Intel faces is that Aero does not run on the current IGP solutions, and as a result, OEM's will gravitate toward discrete solutions. 

This is simply an attempt to recapture a market Intel periodically dominates. In any case, it would be interesting to stir up the old Rambus issue again and "test" if XDR2 vs. GDDR5 actually does perform. Hey, second time is a charm.

http://www.rambus.com/us/products/xdr2/xdr2_vs_gddr5.html

If Intel's GPU is even remotely successful, it creates competition, spurs on innovation, and creates choice for the consumer, as prices drop and capabilities increase. it's a win-win for all of us. Intel won't be knocking off NVidia or ATI, ever, just like Apple and Linux aren't knocking Microsoft off the perch.

The two big ones still offer up exceptionally poor drivers for their cards. If Intel can provide a decent performer with more flexibility and much more importantly more stable and hasslefree drivers they have a winner. They're currently easily one-uping the other two on Linux drivers so it sure is possible.

/signed frustrated ATI driver and NVIDIA driver user

Intel may not have the experience and optimization needed for graphic processors but it has the best silicon and own state of the art fabs. Nvidia and AMD may have the best graphic design but in the absence of good semiconductor processes, transistor technology and modern fabs, intel will still lead by processing power and sheer brute force methods. 

The defective Nvidia parts being recalled and for others it is being sued is a testimony of its poor processes. Similar case happened with AMD Phenom when it came up with TLB bug and had to be delayed so long to become uncompetitive. Intel processors are efficient and more powerful , and AMD has lost the lead and money they got due to their revolutionary Athlon64 and Opteron design. If Intel continues to execute and listen to customers and engineers as it has done in the recent past, it has nothing to fear from the underdogs.

Seems like a bad idea to me, their expertise is to track trends in the market, new things are unpredictable. Also, whilst the Pentium may be 'antiquated', Larrabee doesn't just stick 32 586's on a die and get on with it:

It has a 1024-bit ring-bus similar to what was seen in R600, 16 ALUs per core (so that gives us 512 ALUs in total (with of course other logic which will make recursion possible and make it more adept at branching and such)) and an estimated clock speed of around 2GHz.

I think people are all too quick to blam Intel just because they are so successfully at selling crappy chips in the low end and that they're trying a 'different' approach. Although it does show how people are scared of change.

Oh and I can't help but connect this to the whole Nvidia vs. Intel thing going on.

what is keeping oh lets say amd, via, ibm and even motorola's old cpu division from doing the same. After all larrabee is a very old cpu core tweaked for multicore use.

ehhh who owns the old alpha core? 

think about that, efficient fast cheap and no x86 problems


Travel back in time to 1998. The prevailing chip from Nvidia was the Riva 128 and ATI had the crappy Rage Pro cards. All came in 4 and 8MB formats. There were some other dead-to-be cards in the market including the Matrox Millenium II, PowerVR PCX2, Rendition, and Voodoo II. At that stage in the game 3D accelerators were so pansy wet toilet paper made them grind their teeth. 24-bit? PowerVR had it, just not at any playable framerates. Single-pass multi-texturing? Voodoo was the only one.

Then Intel came to the fore with their "incredible" I740 discreet chip. It offered 8MB onboard memory and was specifically designed to utilize the AGP bus to store textures locally. They promised superior image quality and faster frame rates. The reality was that the image quality was ok. It couldn't do 24-bit z-buffer. Single pass multitexturing was to come later. Intel also failed to provide a badly needed OpenGL driver until more than a year after it was released. That was horrible as it was plagued with problems in Quake2 engine games. Direct3D wasn't much better. Nothing came out after the I740 for Intel's discreet line. 

It seems to me that this Larrabee will become Intel's 21-century Laughabee. Aroof.

Larrabee- seems it will be continued complexification of Graphics Engine. Or Engines, actually. its step for Intel & like small bitty baby ati like tenicale for Big Intel. Suddenly Budding from some new multi core demands. Larrabee give Intel challenge of making CoProcessor thats significantly more powerful than Mainboard CPU, from AhHem, Intel?. Its Hop & Step. Heck I doubt if Stock will do much more than Carrie On.
drashek

I have my doubts too. Nvidia's stock on the other hand has been falling nicely since July. Intel can just wait for them to take the full impact and acquire.

I used to own one of the early i740's (a kyro too, I used to have a bit of a fetish back then). It wasn't bad, don't get me wrong, but Intel always had ambitions and problems with breaking out of the low-end, even in the "good times".

You are forgetting one big advantage of intel:
They have the better/faster process technology.
Even if they are 30% slower in design, they could and up equally as fast.

All attempts by Intel to break into the gfx-market have been disappointments performance wise. Remember the i740? And their on board GPU-solutions have been abysmally weak as well. The common factor is the initial claims by Intel that they're going to rock the world and then end up with a failure as far as performance is concerned. I have no hopes for Larrabee either. Remember MMX which would make discrete GPU's obsolete, increase games and media by 400% and so on? Lol!

Larrabee seems intersting on the paper but it has to perform.
If it will be a failure well... than Intel is in trouble because GPU is getting more and more important and traditional CPU's are losing importance.

In this aspect Inte, Chipzilla is just like MS. They want to design something, they design what they think is right without really taking into account what the market/customers really would like. Mind me, they will say they do, but no matter, the results talk for themselves.

Intel was really lucky with their Israeli team.

Now they want to derive a graphics chip from Pentium tech? Right...

I just see the same old Intel again, talking and talking.

I will believe when I see benchmarks from 3rd party testers and even then... :)

Aero will run on current GMA's and run games, they simply still cannot perform on par with the lowest end discrete GPU's. OEM's have gravitated toward discrete graphics.

"Larrabee is aimed at the IGP market where performance isn't really a factor but power consumption and basic SOHO functionality is required. One problem Intel faces is that Aero does not run on the current IGP solutions, and as a result, OEM's will gravitate toward discrete solutions. "

So I guess I'd better stop using it on my Vaio TZ then... Aero glass works absolutely fine on current integrated graphics, you obviously have no clue what you're talking about.

the games marketing on PC represent around 8% of the market, and out of those 8%, only 5% of the games will challenge a GT9600. SPORE is a perfect example of this, a G945 does ok on it, it is not awesome but it works ...
now, people who do not play are interested into getting more value out of each transistor. Most of the people not playing but buying discreet graphics cards never get return on investement on the transistors. X86 is the best return on investment you can do, it is flexible, and new application runs immediatly on it. x86 is the legacy that let you run your DivX 3.0, or your indeo 5.0 video of 5 years ago, or even 10 years ago.
What the GPU vendors provide to you are video encoder GPU based is working ok, it does generate a video with art effect, but it works, the only problem is that they do not deal with legacy, they can not encode for DivX 5.0 (if you have a DivX DVD player, u need this!), they have good tech demos, but non of the legacy part is not taking care. Larrabee is the attempt to deal with Legacy on the GPGPU side, to be uniform and logic into the architecture. The same discussion between MMX or non MMX happened, most of the tech press was saying that it is too difficult to program in SIMD, and they were saying that custom MPEG2 chip are better than MMX ... we all know how it ended: Since Win2000, MMX is required by the OS, and no PC is sold with MPEG2 chip to play or encode video. It is all about return on investment of the transistors you put in the platform. 
In the futur, the GPU will become a coprocessor to do more than 3D, but in an homogenious platform, beeing able to run x86 legacy code, because this is good for the final consumers, and it is good for the programmers.
The rest is 3D gaming only with about 100 technology demos ...
Bottom line: You can not recompile the world for each generation of GPU u are doing ... it was proven not to be possible ...
so, x86 in the GPU is the convergence to GP GPU, no other path can deal with the legacy of the PC.

Francois

The power of larrabee is mainly that it is x86 (don't know x64). It seems to me that Intel is planning to mainly focus on this to get any processinghungry application on larrabee because of it's
'easy' (more correct: available development tools for this) instruction set.

Magius, why do you think Intel are not talking to potential customers about this? They are certainly talking to us about it, and have been for a long time. We are definitely a potential customer. And they have taken all the feedback on board, as we can see by updates they make to their plans.

You underestimate Intel at your own peril.

I'm not getting why everyone is like "OMG Larabee is based on the Pentium 1 it's so old LOL!"

Gee, superscalar in-order pipelined core, was about 3.1m transistors back in the day, strip out the FPU (since they'll have whatever wonderFPU they're taking about) and cache and it's a bit smaller. Maybe they picked it for mips/transistor efficiency? Ever consider that one?

Or maybe it's something about having to get game code running on this thing. Oh if only there were similar chips already running games so coders could be used to their limitations. Oh wait, the PS3 and X360 both use (gasp!) "obsolete" in-order design cores similar to the P1.

But this is about graphics so that shouldn't matter, it's not like graphics cards are going to be running the physics or AI of modern games in the future, what a laughable concept!

Obviously only a company with specialized experience with the very specialized needs of the very specialized graphics pipelines of modern GPUs can possibly compete. I mean it's not like graphics pipelines have branching or integer operations or anything silly like that! I mean that would imply they were Turing-complete and no different in capability than a general-purpose processor, just more or less efficient.

And of course there's no way you'd suddenly want lots of general-purpose processors with familiar instruction sets, because it's not like rasterizing polygons with advanced shaders, raytracing, and radiosity are all different viable approaches converging on the same result with the only difference being the efficiency of specific things. I mean if that were the case you'd see different rendering approaches generating nearly identical fidelity, which would imply you could use whatever algorithm suited your hardware best and get the same image...

And of course, Intel obviously has no clue how to make a high bandwidth memory controller! I mean it's not like they've ever designed one for a current or upcoming CPU or anything! Plus all they've really got at the end of the day is the best process technology, and all that would let them do is fit the entire thing in the same package or core so it can have a ridiculously wide memory path that a PCB can't match, and what good would that be? Besides that approach is silly, or some other GPU/CPU company would have mentioned plans to do that sort of thing.

Ok enough sarcasm. Nvidia is more likely to fall over from having shot themselves in both feet, kneecaps, and the head before getting "knocked out" by Intel at the rate they're going. And AMD despite their improved GPUs are falling apart on the CPU front, and their Fusion project looks like it may get beat to the punch by a Nehalem+Larabee, even if it's a hack like how Kentsfield made it to quad core. Nobody will care if its "real", just that it works.

3D game is most important for a traditional GPU, but is not what intel really care about. On the other hand, general purpose computing and HPC seems more important to intel. Larrabee will have big advantage in these area due to it's design (check the DP flops).

How many of these chips will be stacked inside the xbox720?

Couldn't Larabee be just like MMX on steroids? That would fit perfectly with Intel's comments of ATI and Nvidia still being successful and would further go some way to provide barriers for AMD CPUs if Intel is able to gain industry software support. I imagine it would be very difficult for AMD to incorporate support for Larabee instructions as they did for MMX / SSE back in the day. Just my 2 pennys.

Larry is not targeting the current graphics. It is creating new paradigm: realtime ray traycing. This wil bring ultimate photorealistic quality.

Intel has really only released one good graphics card for 3D and that was a very long time ago. Heck they didn't design the i740, they bought it!

Intel's sending out a LOT of smoke right now, but Larrabee will be like most of their "3D" chips. It will suck for gaming.

If they can fit 20 cores on one die why is it taking intel so long to make a four core die?

quote from stupid analyst 
""Nvidia has had over 10 years to optimize the 3D graphics pipeline, the necessary drivers, the platform connections needed to supply the memory bandwidth required, and to work with the software and apps developers". " 

Now this shows that this guy knows absoloutly nothing about nvidia, from a tech POV, he knows it from a cash, mullah, dosh, deniro POV which is very stupid for an analyst.

Nvidia after 10 years still cant produce a 100% driver, and they cant produce a video card as powerfull as the 9800 etc etc, without it being almost 1Km long, weighing almost 100KG, requiring stupid amounts of power, and also noisy. So tell me Mr ANALyst why is it you know so much about Larabee and how much of a flop it will be when you cant even ANALyse a company you have been looking at for yonks namely Nvidia. 

you know nothing Kumar 

*we are relentless when it comes to flaming us INQuirians 
(people who like inquirer) so before you go running your limited mouth, make sure you know what your talking about

It's unfortunate you published the comments of an Investment W***er. 
Sounds like he and his mob haven't got a particularly blue portfolio.

Intel leaks are talking about using Pentium-derived cores for the graphics pipeline, yet I don´t recall any P54C design of the original Pentium scaling up beyond 233 Mhz due to limits on the architecture, pipelines, etc., the P54C core was all-CISC, not RISC cores running decoded CISC programs like today, they are talking 1.0+ Ghz core speeds... yes it´s going to be a hell of a ride for Intel to pull this out.

remember the i740?
Intel has already tried the discreet graphics business - and failed.
In fact, so far Intel's non x86-cpu history is a long list of failures.

I won't speculate on the performance of intel's latest attempt before you can buy the stuff in regular computer hardware shops.

"I certainly expect Nvidia and ATI to carry on being successful," and adding “We're trying a different approach”.

What's different from being successful? Is there any other option besides failing. This will be a ray tracing beast, but ray tracing is so complex that it probably won't be able to compete visually with similarly priced NVIDIA/ATI products.

The top graphics wizards, Pixar, didn't bother with ray tracing until making the movie Cars, and even then it was only for certain effects in certain scenes. It's more of a last choice technology, since most graphical effects are much easier to produce with other methods. Even getting the fancy lighting and shadow effects that Intel tries to pass off as ray tracing only, can easily be done on normal GPU's. (check out real time radiosity(sp?)) Eventually, ray tracing will become common, but it will only be part of the process.

Plus, Intel's video drivers suck royally, at least on Linux, where every movie studio does their graphics work. Intel's drivers cannot run half of the KDE programs and randomly crash the X server monthly on my wife's computer.

@Pantera,
I think Intel owns the old Alpha IP. After buying Compaq, HP sold new Alpha systems until April 27, 2007, and they still sell refurbished Alpha systems.

"It is precisely this “different approach” Kumar seems to be taking to task in his report, with Intel using software which will have to run properly on multiple processing cores and based on what Kumar calls an "antiquated" Pentium design."

And he is an analyst for whom again? What an absolute idiot! So he thinks Larrabee is simply old pentiums put together? LMAO!!! He is so beyond hope that I won't even bother commenting further.