Intel's Larrabee Graphics Chip Delayed Indefinitely

[MacRumors]

http://www.macrumors.com/images/macrumorsthreadlogo.gif (http://www.macrumors.com/2009/12/08/intels-larrabee-graphics-chip-delayed-indefinitely/)

ArsTechnica reports (http://arstechnica.com/hardware/news/2009/12/intels-larrabee-gpu-put-on-ice-more-news-to-come-in-2010.ars) that Intel's Larrabee graphics chip technology has been delayed indefinitely: Specifically, Larrabee v1 is so delayed that, at the time it eventually launches, it just won't be competitive as a discrete graphics part, so Intel plans to wring some value out of it by putting it out as a test-bed for doing multicore graphics and supercomputing development. Intel will eventually put out a GPU, but may not be the one we've been calling "Larrabee" for the past few years.Larrabee had been (http://www.macrumors.com/2008/08/04/intel-details-upcoming-gpu-project-larrabee-due-in-2009/) the codename for a new graphics card technology that would compete head to head with NVidia and ATI. Larrabee had a unique hybrid design that was said to scale incredibly well with multiple cores. Apple's Snow Leopard was well poised to take advantage of this multi-core design and Apple had been rumored to be planning on adopting the chip upon its release:
And I've heard from a source that I trust that Apple will use Larrabee; this makes sense, because Larrabee, as a many-core x86 multiprocessor, can be exploited directly by GrandCentral's cooperative multitasking capabilities.Obviously, with this development, we're not going to be seeing this technology in Macs anytime soon. In fact, Intel is not even planning on announcing details about their followup graphics product until 2010.


Article Link: Intel's Larrabee Graphics Chip Delayed Indefinitely (http://www.macrumors.com/2009/12/08/intels-larrabee-graphics-chip-delayed-indefinitely/)

[xlii]

Not good news. Just goes to show that it isn't easy to design a top quality graphics chip. There is a learning curve here. Even for experienced CPU designers.

[PeckhamBog]

Shame, the more competitors, the better?

[dernhelm]

One word: Wow.

I knew they were struggling a bit getting traction under Larrabee, but I didn't quite understand the situation as well as I thought, I guess.

More competition is always better, so its a loss from that perspective.

It's also odd with Intel's strained relationship with NVidia, I figured part of the reason for that was Larrabee. Apparently not.

I can see it being released in future on an OpenCL compatible add-on card. That could be killer stuck in an Xserve or MacPro, doing high-end video work. It would also obviate the need for a "mature software stack" in order for the technology to be useful.

[Data]

So what does this mean for macs and graphics for 2010, with the whole Nvidia not getting licenses and such ?

[iMacmatician]

Waiting for info on the "new Larrabee."

I can see it being released in future on an OpenCL compatible add-on card. That could be killer stuck in an Xserve or MacPro, doing high-end video work. It would also obviate the need for a "mature software stack" in order for the technology to be useful.I can also see that happening.

[arn]

So what does this mean for macs and graphics for 2010, with the whole Nvidia not getting licenses and such ?

The Nvidia thing shouldn't affect much with respect to graphics as that's an issue with the chipset. Apple is still free to use their graphics cards in their computers.

arn

[borcanm]

Wow. This means were not going to see Larrabee in macs for a very long time. If Intel is delaying it so long, its obvious when it comes out it will be sluggish behind other GPUs.

[andiwm2003]

yep, you can't just walk in and be better than ATI or NVIDEA. They have years of GPU experience.

Well, I hope that Intel still makes progress in providing eventually a great integrated solution that can make up for the 9400. The future a light notebooks or netbooks and there integrated GPU's is the future.

[eastcoastsurfer]

Not good news. Just goes to show that it isn't easy to design a top quality graphics chip. There is a learning curve here. Even for experienced CPU designers.

We don't know the constraints under which the engineers had to work. Intel is a CPU seller first and foremost. They don't want anything they sell to allow you to use a cheaper CPU (many systems nowadays only need graphics card updates to stay competitive). Anything coming out of Intel will be processor dependent. See USB vs. Firewire.

[Speedy2]

This could mean that Intel is buying Nvidia now. Strategically, it would be a wise move for both companies.

[iPhysicist]

This could mean that Intel is buying Nvidia now. Strategically, it would be a wise move for both companies.

...and a slap for AMD/ATI. I would like to see that happen. If Intel can manage to take advise of nVidia's brains it could work out so we have a well performing integrated graphic chip in 2011 (and a decent chip for you PRO consumer with the real macbook pros :rolleyes: )

[J the Ninja]

I'm glad they canceled it, cool as it would've been. It was very quickly turning into another Itanium. I heard tell it ran as hot as the GTX 295 or the 5970, yet could barely keep up with the single GPU versions of those cards. Looking forward to what else the project puts out.

[Riemann Zeta]

This certainly isn't surprising. Given that this Larrabee GPU would have been consisted of a bunch of simplified (in-order) x86 cores, it would have been utterly incompatible with any existing GPU architectures. Hence, the drivers must have been a bitch to write--and it is unlikely that the first few iterations would have yielded performance anywhere near modern ATi and NVIDIA hardware.

[HONDAxACURA]

If Intel can truly make a decent graphics card, that would be great news, but I still will buy a product with nVidia graphics card inside.

An Intel GPU will be a turn-off to the computer people. Stick with nVidia Apple.

[MorphingDragon]

This could mean that Intel is buying Nvidia now. Strategically, it would be a wise move for both companies.

No it wouldn't be, nVidia has pushed themselves into irrelevance. The promised Fermi seems like vapourware at this point. They would probably be pouring money into something that wouldn't work out.

---

http://images.nvidia.com/blogs/ntersect/feature_image03.jpg

[Eidorian]

Was anyone expecting a model outside of GPGPU usage?

There's no relation to Intel GMA. I wonder what the IGP core on Sandy Bridge will be like since it is on die now.

[Len Da Hand]

With a name like larrabee what was Intel thinking!!!

in get Smart he was even more incompetent than Max - i wonder what larrabee's agent number was!!!

Sounds like someone at Intel has lost CONTROL and KAOS is taking over. Under a cone of silence perhaps?

[iMacmatician]

I'm glad they canceled it, cool as it would've been. It was very quickly turning into another Itanium. I heard tell it ran as hot as the GTX 295 or the 5970, yet could barely keep up with the single GPU versions of those cards. Looking forward to what else the project puts out.Fudzilla said 300 W some time ago.

[MM123]

This could mean that Intel is buying Nvidia now. Strategically, it would be a wise move for both companies.



http://www.cringely.com/2009/12/intel-will-buy-nvidia/

[Speedy2]

No it wouldn't be, nVidia has pushed themselves into irrelevance. The promised Fermi seems like vapourware at this point. They would probably be pouring money into something that wouldn't work out.



What the hell are you talking about? Nvidia might be a few months behind but that doesn't mean they are out of the game. Did you say the same thing about ATI when their HD 2000 and 3000 series couldn't even nearly beat Nvidia's competing cards? Look where they are now.

I'm just saying that Intel needs a new strategy if they want to compete in the graphics / GPU computing market. Buying Nvidia would give them a very good CPU platform AND a good GPU. I'm pretty sure Nvidia got a Nehalem chipset running in their labs, but they can't release it without a license. AMD has had the better CPU platform for ages now, which partly made up for the weaker CPU. Time for Intel to do something about it. Intel's chipsets + IGP suck big time compared to ATI + Nvidia. (That's why Apple ditched them) And don't forget about Tegra, that might a worthy acquisition for Intel, too.

[Data]

The Nvidia thing shouldn't affect much with respect to graphics as that's an issue with the chipset. Apple is still free to use their graphics cards in their computers.

arn


Ty for clearing that up for me ;-).

[MorphingDragon]

What the hell are you talking about? Nvidia might be a few months behind but that doesn't mean they are out of the game. Did you say the same thing about ATI when their HD 2000 and 3000 series couldn't even nearly beat Nvidia's competing cards? Look where they are now.

Nope... See there WERE actual ATi chips by now and they were very good value for money. If Fermi EVER comes out and doesnt turn out to be another 2900XTX, itll be over priced and probably one huge let down. See all we have so far of Fermi is an engineering sample, and that was at nVision. nVidias chips usually go to the A3 revision until they're released.


I'm just saying that Intel needs a new strategy if they want to compete in the graphics / GPU computing market. Buying Nvidia would give them a very good CPU platform AND a good GPU. I'm pretty sure Nvidia got a Nehalem chipset running in their labs, but they can't release it without a license. AMD has had the better CPU platform for ages now, which partly made up for the weaker CPU. Time for Intel to do something about it. Intel's chipsets + IGP suck big time compared to ATI + Nvidia. (That's why Apple ditched them) And don't forget about Tegra, that might a worthy acquisition for Intel, too.

Intel admit defeat, funny man. Funny.

You do realise that Tegra is just a re-branded ARM chip?

[macerroneous]

Maybe Intel sued NVidia to try to get license to use some NVidia IP to make Larrabee work. NVidia didn't bite, evidently.

[holmesf]

What the hell are you talking about? Nvidia might be a few months behind but that doesn't mean they are out of the game. Did you say the same thing about ATI when their HD 2000 and 3000 series couldn't even nearly beat Nvidia's competing cards? Look where they are now.

Seriously, and outside of gaming it's ATI that is becoming less relevant. Nvidia offers some damn nice high performance computing solutions (eg Tesla (http://www.nvidia.com/object/product_tesla_s1070_us.html)), and Fermi (when it materializes) will cement this advantage over ATI. OpenCL was also based on Nvidia's CUDA, as opposed to ATI's stream, and in general CUDA provides a great cross-platform programming environment. For these reasons Nvidia is gaining rapid adoption in the scientific computing community and academia -- perhaps not the most important markets, but it's certainly winning the mindshare of some super elite nerds.

[Speedy2]

Nope... See there WERE actual ATi chips by now and they were very good value for money.
(lots of blah blah deleted)

Yeah right, if I could actually BUY an 5800 card anywhere that would be true. Is this just another blind fanboy speaking here who cannot accept criticism or facts about the "other" side? Jeez.
Fermi WILL come out, it will be like 5 months behind ATI and it can be expected that the fastest cards will be quite a bit faster than ATI's current offerings and also more expensive. Just like it was before 5xxx came out.



Intel admit defeat, funny man. Funny.

You do realise that Tegra is just a re-branded ARM chip?

Another one of these empty statements. Intel already accepted defeat big time when they killed the Pentium 4 and continued developing the Pentium-M (later to be known as the Core) instead. They accepted defeat when they ditched Rambus and are now ditching FB-DIMMs in favour of DDR-RAM. You are just talking BS there.
The CPU part of Tegra is not what Intel is after but the GPU part. It can play HD flawlessly and consumes far less power than Intel's forays into the low power sector. And it's already THERE, I can't buy any Intel systems-on-a-chip for mobile devices.

[Eidorian]

It looks like Clarkdale and Arrandale will at least have limited GPGPU functions (http://www.xbitlabs.com/news/video/display/20091204232215_Intel_s_Next_Generation_Platform_to_Support_GPGPU.html).

[flopticalcube]

It looks like Clarkdale and Arrandale will have at least limited GPGPU functions (http://www.xbitlabs.com/news/video/display/20091204232215_Intel_s_Next_Generation_Platform_to_Support_GPGPU.html).
But the benefit will certainly be limited since they share the same module which has a finite power capability.

[TMar]

I have said for years that Intel needed to try to move Nvidia under their thumb, the same way AMD did with ATI. Intel's more to not license Nvidia is just a tactic to weaken their market value.

As far as "nVidia has pushed themselves into irrelevance", maybe you haven't been around long enough but Nvidia and ATI have always had this seesaw game of going back and forth with each other. And every time one of theses flip-flops happen someone like you always points to this nonexistent sinking ship. Welcome to 15 years ago.

[Eidorian]

But the benefit will certainly be limited since they share the same module which has a finite power capability.TDP should be already calculated for full CPU and IGP load. You won't get Turbo Boost from either component though.

[MorphingDragon]

OpenCL was also based on Nvidia's CUDA, as opposed to ATI's stream

Proof?

BTW, the so called high performance CUDA cards are just different drivers and more ram. Same goes with the Quadro cards. Proven by the fact you can get a Quadro card by soft modding its GeForce equivalence.

[holmesf]

Proof?

BTW, the so called high performance CUDA cards are just different drivers and more ram. Same goes with the Quadro cards. Proven by the fact you can get a Quadro card by soft modding its GeForce equivalence.

Of course the official position is that OpenCL was developed in collaboration with both AMD and Nvidia. But having used both, I can tell you that OpenCL and CUDA are practically identical, the main difference being that OpenCL is compiled at runtime.

Even AMD admits the two are practically the same (look at the tables comparing concepts and interfaces between C for CUDA and OpenCL):
http://developer.amd.com/documentation/articles/pages/OpenCL-and-the-ATI-Stream-v2.0-Beta.aspx

[MorphingDragon]

(lots of blah blah deleted)

Yeah right, if I could actually BUY an 5800 card anywhere that would be true.

It all depends on wherre you live.

Is this just another blind fanboy speaking here who cannot accept criticism or facts about the "other" side? Jeez.
Fermi WILL come out, it will be like 5 months behind ATI and it can be expected that the fastest cards will be quite a bit faster than ATI's current offerings and also more expensive. Just like it was before 5xxx came out.



Theres a difference between Critique and noise.

[MorphingDragon]

Of course the official position is that OpenCL was developed in collaboration with both AMD and Nvidia. But having used both, I can tell you that OpenCL and CUDA are practically identical, the main difference being that OpenCL is compiled at runtime.

Even AMD admits the two are practically the same (look at the tables comparing concepts and interfaces between C for CUDA and OpenCL):
http://developer.amd.com/documentation/articles/pages/OpenCL-and-the-ATI-Stream-v2.0-Beta.aspx

Similar and same are different words, look them up.

If you're calling the programming Syntax similar then I agree with you. But if you're saying the the APIs are EXACTLY the same then I have to digress.

[holmesf]

If you're calling the programming Syntax similar then I agree with you. But if you're saying the the APIs are EXACTLY the same then I have to digress.

The syntax is similar, but more than that the most important abstractions (thread blocks = work groups, local memory = shared memory, etc) have a one to one mapping. I wouldn't say the APIs are exactly the same, because they aren't. But the APIs don't differ in fundamental ways either (other than the fact that CUDA is a C extension and OpenCL is compiled at runtime).

[twoodcc]

i guess they hit a road block with this. i'm sure apple is pushing intel to make this, and make it right. we'll see next year hopefully