Is Intel really worried about AMD's moves to set up an Arab-backed foundry company? Is that why Intel has said the foundry won't have a licence to manufacture x86 parts?
Clearly it's a move to try and wreck the project, but why would Intel want to do that? Intel knows it has to have a rival in the x86 business or it become a monopoly which will attract heavy government regulation - or worse - remember AT&T?
Recently Intel impliedly acknowledged that its cost structure was too high for making certain types of products when it out-sourced Atom processors to TSMC.
Asked why he thought Intel had done that, Rich Beyer, CEO of Freescale Semiconductor, told me:
"Because the cost structure of the netbook and MID market is going to be significantly different than the cost structure of the processor market for laptops and desktops", said Beyer, "it's simply a decision based on: What do our factories do? They make desktop and laptop chips. Do we want to sell processors that sell for $20 and less? Do we want to develop fabs to sell that sort of product?"
Putting the same question to
"It's because Intel's cost structure is so outrageous", said Penn, "it's pushing the leading edge, pushing the technology. Its conventional microprocessors can absorb all that cost - Intel still gets an average of about $100 for its processors - but you turn that structure to making jelly bean stuff and it doesn't work."
Clearly Intel must know that there's only one way for the price of x86 processors to go. In a PC market that has been commoditising for a decade, that way is down. That heaviy manufacturing overhead is going to become more and more burdensome as PC commoditisation continues.
So along comes Intel's only real rival, AMD, with a plan to get the Arabs to fund its fab. What kind of cost structure will the Arab-funded fab have for making x86 processors? No one knows. But, the Arabs have more money than even Intel, they can set the wafer cost to AMD at any level they damn well like.
The prospect of the control of x86 pricing moving from Santa Clara to the Arabian Gulf would be more than a trifle vexing for Intel.
concerning Intel, maybe this legal activity is just another moke screen, in order to dissimulate a bigger problem:
- intel lakes a real low power process (because its would be low power processes are inded derivated from high performance processes anyway)
- intel cannot admits such as it as always marketed strongly its process advantage
(expect bad Wall Street reactions of admiting so)
- intel inks this TSMC deal to supposedly open its ATOM core to the market
- but intel said it can refuse to sell the core to customer it doesn't like
- this is a losing proposition Intel don't expect seriously anyone to subscribe
- the only real TSMC customer which will use the ATOM core will be Intel itself.
- but in the meantime nobody has realized that Intel has no process for the new 'hot' stuff (cold inded, the low power gadgets)
Today Intel has many unused capacity in its precious fabs, so to use this over capacity is even cheaper than to outsource to TSMC. To transfer ATOM production to TSMC is not currently justified at all by economics, but by technical consideration.
Smoke and Mirrors ...
Fabrice, I believe you are right that Intel has oversold its process technology advantage to Wall Street analysts. At the Intel-TSMC proess conference a number of Wall St analysts asked if Intel was transferring process technology to TSMC to make Atom, and they seemed surprised when Intel said 'No'. Intel couldn't say otherwise because TSMC CEO Rick Tsai was sitting on the panel. But in other situations, Intel likes to say it is ahead of eveeryone else on process.
Secondly, I believe you are also right when you say Intel has gone to TSMC in order to get a low-power process. I haven't seen any sign of Intel being able to make true low-power processors.
However I don't agree that there is no economic reason for Intel going to TSMC for Atom. Recently I asked two people why Intel went to TSMC for Atom and the replies were as follows:
Rich Beyer, CEO of Freescale Semiconductor replied: "Because the cost structure of the netbook and MID market is going to be significantly different than the cost structure of the processor market for laptops and desktops", said Beyer, "it's simply a decision based on: What do our factories do? They make desktop and laptop chips. Do we want to sell processors that sell for $20 and less? Do we want to develop fabs to sell that sort of product?"
Malcolm Penn, CEO of Europe's leading semiconductor analysts Future Horizons, replied: ,
"It's because Intel's cost structure is so outrageous. It's pushing the leading edge, pushing the technology. Its conventional microprocessors can absorb all that cost - Intel still gets an average of about $100 for its processors - but you turn that structure to making jelly bean stuff and it doesn't work."
Is it just me or the power comparison between ARM and x86 is drifting between now and some times ago.
In the past I was used to read that the main problem of Intel and the pentium was that x86 in order to be MIPS efficient had to have a huge instruction decoding & fetching overhead which is not at all power efficient, whereas RISC CPUs were the opposite. Now it seems that the whole discussion is focused on foudries and processes. But is it even possible for an x86 CPU, at equivalent MIPS, equivalent process technology, to be as power efficient as an ARM? Or is the RISC/CISC battle finally of minor importance?
I don't know, Djonne.
If you ask the ARM people how they get their low-power rating, they say they've developed loads of techniques over the years to keep power down and their low power rating today is just an accumulation of a couple of decades of design experience which was always focussed on achieving low power..
concerning the power consumption of x86:
yes the nightmare decoding of x86 instructions cost significant power, but another fact is that the x86 is so complex to verify (because of all the accumulated obsolete stuff), that this represent an enormous effort (that why there is so few x86 players), thus for a given budget, when you do just a RISC processor, far easier to verify, you can spent a lot of resource doing low power optimizations (instead to verify x86). This is the main hidden cost of x86: its intrinsic complexity consumes most of the efforts, these consumed efforts are not applied to more relevant characteristics (x86 compatibility requirement resulting only of politics consideration, enforcing the Intel/Microsoft de facto monopoly, and do not correspond to any technical arguments). From an entropy point of view it is very sad to think to all these brilliant minds working to solve problems resulting of the ugliness of an, originally 8-bit*, architecture, dating from the 70's.
(* 8086 was assembly source compatible with 8080, an 8-bit processor).
Interestring stuff, Fabrice, thanks. Maybe Intel's on a hiding to nothing in the low power market.
If it was only an architecture problem, Intel would have made it with Xscale. Maybe there´s something fundamentally wrong with the whole Intel... fabs, processes, business philosophy... that prevents them the real success at low power.
Atom did start a new market, but as many analysts already think it wont dominate for very long. The only thing slowing this development down is the wide SW support that Wintel has over other platforms.
Very wise, Veijo, as always.