was delayed again in a big way yesterday. After consistent schedule slips, Intel mgmt has finally realized Larrabee didn't have the nut-sack to hang with the big boys.
Arguably Larrabee is Intel's most important project.
First, what we've known as PC CPUs has become Atom, lighter, smaller, more power efficient, but the trend is clear. More and more computing is going portable, fewer CPUs are sold as productivity stations like traditional desk tops or desk sides. Intel's revenue for mainstream will drop precipitously over the next year or two because Atom (and follow ons) are lower selling prices.
Second, there is the whole system-on-a-chip segment, think iPhone. Intel thought hand-held would benefit from Atom, but at one watt it's way too power hungry. So there is no x86 answer for this space and Intel really has nothing to offer. The traditional players like QCOM, TXN, TI etc have good solutions for the foreseeable future, or until some disruptive technology enters.
So what's left? The high end, but what's happening there? GPU guys are claiming 10-100x performance improvements computationally intensive tasks like oil and gas exploration, medical image processing, financial modeling. With the addition of Fermi, an EEC compliant part that will ship early next year, the high end cash and margin cow position Intel has enjoyed is being challenged in a big way. Now revenue is not going to go away tomorrow, but with GPU's power and cost efficiencies 1/10th of a high end CPUs, the margin contributions this segment enjoyed in the past will surely decline from the moment Fermi shows up.
Larrabee was supposed to stave off the high end threat from the GPU guys, this is really the only segment which Intel can count on to generate high margins. Additionally, it was to generate revenue in mainstream through gaming. Sounds like this announcement took gaming completely off the table. What will make up for lost revenue in mainstream?
So the BIG RESET button was pushed this week. Larrabee has already cost billions in R&D, and now is guaranteed to cost billions more. It was supposed to plug holes in an aging product line up. Founder Gordon Moore's law is proving now more detrimental than beneficial as Atom (x86's legacy) can't help but go down the cost reduction path.
Larrabee's delay puts the anticipated revenue stream out to some unknown point in the future. It also gives a larger lead to multi-threaded processors on the market, and delays the software development work.
This delay is a big deal, press and analysts will show this in the coming weeks and months. Intel's margins and revenue are going to get squeezed over the next year or two with little prospects of replacement. I'll bet Paul O never thought his reign would be this challenging.
Go to LWLC message board and read about patent applied for yesterday. Could INTC realize "disruptive" is here with photons soon to replace electrons, ergo Larrabee, in effect, dead on arrival?
(polymer replacing silicon)
Nvdia and AMD have a much bigger chance of failure trying to get GPU's to compete against Intel's CPU strong hold or in AMD's case, fusion of the two, than Intel has of making Larabee an innevitable success.
Intel is just delaying because they don't want to release something that doesn't absolutely blow the pants off the competition.
I really appreciate the thoughtful (cough) reply. The fact that you guys are EVEN talking about NVIDIA should scare the crap out of you.
The sun has started setting on x86. Heterogenous computing will replace what we've come to know as x86 server farms at the high end. At the low end, hand-held devices (driven by ARM - what a mistake it was to give that business up, eh?) will be the portable CPU of choice.
What does that leave Intel? The squishy middle. As a tweener.
Since process enhancements have basically tapped-out on frequency and power efficiency, Intel has to look elsewhere for gain in x86 -- the legacy value-add intel has is software compatibility, and that counts for something. But in the meantime, since x86 can't get faster, Moore's Law will shrink x86 to nothingness. That means x86 becomes a commodity, and lowest price wins.
So how does software compatibility play in these spaces? in HPC big, computationally intensive applications will need to be re-written to take advantage of multi-threading anyway, so programming compatibility means little here. And NVIDIA has a HUGE lead and gains market momentum daily.
In portable, the trend is smaller, lighter, more capable. But Atom (and x86 follow on) devices are not getting traction because they are power hungry, so applications are being written for ARM anyway. Intel loses a big advantage if they re-architect x86, so that isn't too likely either. So where does that leave them? They are stuck with a declining mid-market longer term. 300M PCs today, eroding ASPs and users moving with abandon to smart-phone like devices that are way more convenient than a notebook.
All I'm saying is the future isn't as certain as you folks want it to be. Intel isn't going away any time soon, but with the big Larrabee RESET, the world just got a whole lot more complicated for Intel.
I asked a founder at NVIDIA in 96 or 97 what he thought the future of the CPU was. He said, "One day the CPU will occupy a little space in the corner of our graphics processor." More than a dozen years later, seems like the forecast is on it's way to playing out.
"we remain concerned that cannibalization, from notebook to netbook, from desktop to nettop, and from much higher CPU price points, will mostly or fully offset other Atom-driven benefits."
Yeah, INtel's first GPU was very ambitious !! To come with with the all-new state-of-art parallel architecture and having zero experience with these architectures and drivers.
And...they are only falling farther behind now. They're traditional MO isn't gonna work in the GPU space.
Perhaps Intel believes that ARM coming from the low end is actually a bigger threat in the future and needs to concentrate on low power processors rather than the radiator Larrabee would have been.
Thats a good point.
But my point was margin and revenue contribution from a high-hundred dollar product is hard to replace. How many ARM like things do you need to sell to make up for one HPC CPU? 20? 50? 100?
Going down the ARM path also commoditizes Intel's offering. As x86 does not play here, there is nothing special or unique they can offer this segment. So they are going to be competing on price and sharing the pie with a dozen other guys.