Intel Kills Off Larrabee Discrete Graphics, For Real

[martel80]

They could have used the Itanium (IA64) architecture for Larrabee since they already have a compiler for that and the instruction set is not so bloated.

 

[vic20]

Unfortunate news.

Just the ridiculous notion of paying a few hundred bucks for a supercomputer-on-a-chip that could possibly give a 15 year old 10 million dollar Cray system a run for its money, at least for a few functions, sitting in my PC running games makes me smile. Could have been a fun toy for folding and other stuff too if a compiler was released for it.

But for a business perspective it makes no sense at all to go after gamers when you could sell the exact same chip to cloud companies like Google and Facebook, data centers, Universities or the military for $4000 per chip for HPCs that may incorporate anywhere from dozens to even thousands of CPUs.

I'm speculating, but if Larabee could allow Intel to build something that can keep up with Cray's Jaguar or IBM's Roadrunner at a much smaller size they would be swimming in new market share and revenue.

It has been said this was the original plan with Fermi before it became a gaming GPU, but I'm sure Larabee's networked x86 design would be more compatible with existing software and super computing infrastructure.

Good business decision, but I'm bummed none the less.

 

[slabbo]

Title is misleading. It should be "Larrabee Kills off Intel Discrete Graphics" cause you know...it sucked.

 

[neonvii]

Here is what i think REALLY happened, Larrabee is probably everything Nvidia and ATI feared it would be and more, because you dont pour that much money into R&D and publicize it before you have something tangible...but as usual, the three made a deal and colluded where the two GPU makers would give large sums of money to intel every year to keep a lid on the project or to "shelf" it for a given amount of time.
i know this sounds like a conspiracy to many of you out there, but the truth is that Nvidia and ATI have been price-fixing for a long time now and all of silicon valley knows it.

 

[V8VENOM]

They were unable to solve the power/heat issues - it's really that simple. They had specifications for 5 versions of Larrabee going out to 2020. I know one of the many engineers working on Larrabee 2. They had FIVE floors of staff working on the various versions -- that's A LOT of R&D talent working on what really is not a very large nor very profitable market place.

Apple's success in the Mobile hardware market is a HUGE wakeup call for Intel, that's another reason why Larrabee is now dead. The real question -- is it too late for Intel to enter this market?

Using Microsoft to help with the software side of developing Tablet OS and/or other mobile devices isn't going to work as Microsoft simply can't make an small footprint (memory) and fast performing OS ... just too many layers and layers of compatibility in the code. The penalty of trying to retain ongoing compatibility for over 25+ years.

Apple's advantage is that they can control the hardware and software, this is a HUGE advantage. The very same advantage that many of you seem to hate for no real valid reason?? Apple's success comes from understanding that the end user really doesn't care what goes into a device so long as it works and delivers what they want ... and Apple have succeeded in doing exactly that ... and they've done it with Samsung CPU and PowerVR SGX licensed by Imagination Technologies and parts that have nothing to do with Intel's own designs.

Intel are now probably looking the market numbers and going ... "OH, this MobileDevice market is apparently bigger than we thought" they really don't have a strong foothold in that market.

And it really doesn't matter if you like Apple or not as this isn't about you, it's about the reality of supplying and demand.

 

[stasdm]

Really Larrabee was a kind of bluff from the very beginning - universal Pentium architecture is not quite suitable for gaming graphics (by definition it will be much hotter than any specialized solution).

By they had to try - at least for the sake of learning to harness the multy-multy-core chips (current technologies do not allow to break the heat barries, so multi-core is the mainstream and multy-multy-core is a near future (at least for a time)).

So, take a closer look at Intel's 48-core (and more cores in near future) processors - here the ideas behind Larrabee may result in very effective ray-tracing software for architecture and cinema production applications.

 

[aaron686]

[citation][nom]martel80[/nom]They could have used the Itanium (IA64) architecture for Larrabee since they already have a compiler for that and the instruction set is not so bloated.[/citation]

What I meant was that with the extra competition from Intel it might drive ATI and Nvidia prices down a little and yes I'm aware of Intel's prices, Im shopping based on the x58 platform ATM>

 

[aaron686]

Sorry @ martel80 that was meant for rambo117.