Some Thoughts on INTEL
November 29, 2010
Russell Fish III
Intel is the world's largest semiconductor manufacturer. It employs nearly 100,000. It just reported record revenue and profits. Gross margins for the most recent quarter were 65%. It maintains a near monopoly position in its primary market of microprocessors for personal computers. Life is good, or so it would appear.
Intel's astounding 30 year run of financial success stems mostly from a lucky break combined with one of gutsiest executive decisions ever made by two public company executives.
The break came in 1978 when Motorola's 68000 microprocessor manager rejected IBM's suggested product modifications for what would become the IBM-PC. IBM then visited INTEL and adopted the 8088 for the PC.
The gut check came in 1983 when Andy Grove and Gordon Moore decided to drop out of the then profitable DRAM market and focus entirely on x86 microprocessors.
Intel, IBM, and Microsoft subsequently rode their symbiotic relationship to a quarter century of PC dominance.
However computers did not begin with Intel.
In the Beginning
First there were mainframes. They were large multi-million dollar computers requiring special buildings and used mostly by governments and the military. They were sold by the dozens.
Then there were minicomputers.
Ken Olsen of Digital Equipment recognized that a smaller, simpler, and less expensive computer might sell in higher volume. DEC created the minicomputer business with the $16,000 PDP-8 which eventually sold about 300,000 units. The PDP-8 and its descendants remade the computer industry and created immense wealth for dozens of minicomputer companies.
The Minicomputer Model Was Simple
Combine off-the-shelf integrated circuits in a unique way. Write software to make the circuits perform useful functions. Then turn the production crank and collect $20K or more at each pop. Along the way, pick up extra revenue with leases, upgrades, customization, and service contracts. From time-to-time refresh the products with the latest off-the-shelf integrated circuits and repeat the process.
Using this model DEC grew to be the second largest computer company in the world behind IBM. Employment grew to 113,000.
The Minicomputer Model Broke
The minicomputer model began to fall apart in 1974 when Intel introduced the 8080 and Motorola the 6800 microprocessors. At $300 each these single chips contained most of the logic to build a simple computer. The economics of computing were forever changed. Despite the compelling marketing opportunity for computers owned by individuals, DEC president Ken Olsen opined in 1977, "there was no reason for any individual to have a computer in his home".
In 1981 the largest computer company in the world introduced the IBM-PC home computer. From that moment forward DEC was a dead man walking. It didn't die immediately and even posted 38% year-to-year growth in 1986. They belatedly tried to enter the microprocessor business and even produced the well architected DEC Alpha. Despite all efforts they slowly, painfully, and embarrassingly bled money for a decade before expiring in the arms of a computer company started on a party napkin.
The PC Model
The PC Model was similarly simple as the minicomputer model, with a slight wrinkle:
Combine off the shelf components in a standard way. Offer with standard software to perform useful functions. Then turn the production crank and collect $2K or more at each pop.
The wrinkles were the standardization on the x86 microprocessor, the PC architecture, and the Microsoft operating system. A cascade of riches rained on Intel and Microsoft from time-to-time when they refreshed the PC with their latest improved x86, Pentium, or Xeon and newest Microsoft DOS or Windows.
The PC Model Broke
The continuing PC revenue model depended on "Moore's Law" which projected increasing microprocessor performance each year. INTEL pushed performance primarily by spending prodigiously to develop and manufacture the very best semiconductor processing. Processing improvements eventually began to run into the laws of physics, specifically the laws that define the current required to charge and discharge capacitors at high frequencies.
"Moore's Law" was replaced by "Moore's Suggestion" in 2004 when Intel cancelled the Pentium 4 and then its replacement codenamed Tejas. The Tejas reportedly ran at 150 watts TDP with a pipeline in excess of 40 stages and a package approaching 1,000 pins.
INTEL then announced that its microprocessor focus would be on multi-core CPUs. This decision only aggravated the memory bus bandwidth problem by requiring multiple CPUs to access memory on the highway previously used by one.
Berkeley's David Patterson succinctly summarized INTEL's problem in "Patterson's Three Walls":
Power Wall + Memory Wall + ILP Wall = Brick Wall
The "Power Wall" referred to the heat generated when a microprocessor goes faster and faster. The "Memory Wall" referred to the number of memory bus pins that will fit on a package. The "ILP Wall" referred to instruction level parallelism as indicated by the number of pipeline stages. Optimizing one wall aggravates another in an apparent endless loop.
Desktops to Notebooks to Netbooks to Tablets
The PC business built on Moore's Law was now constrained by Patterson's Walls. INTEL's processing competitive advantage could no longer cover up the x86 architectural dead end. From that point forward, innovation required extending usability, reducing power, and lowering cost. INTEL was at a competitive disadvantage.
In 2006 Nick Negroponte introduced OLPC (One Laptop per Child), a small form factor computing device. OLPC used a low power x86 processor from AMD, but instead of Windows it used a Linux derived operating system. OLPC was always more of an "Electronic Peace Corps" project than a commercial venture, but OLPC cracked the spell that prevented PC makers from making smaller computing devices without Windows.
Based on OLPC's success, INTEL stripped down the x86 architecture to reduce power in their new ATOM. Acer and others quickly sensed a market opportunity and "netbooks" were born. The early netbooks shipped with versions of Linux. Millions of unsophisticated users unsuspectingly browsed the Internet on a computer not running a version of Windows.
Microsoft responded quickly by bombing the OEM price for XP and regained the netbook business just in time to be crushed by the man from Cupertino.
Apple and ARM Nibble INTEL's Low End
Once again Steve Jobs did what only the very best CEOs do, see the future and then create it to meet his vision.
Steve correctly defined the computer future as being small, wireless, and mouse-less. The iPhone was a baby step, but the iPad was a giant leap. In one move he stuck a fork in both INTEL and Microsoft. INTEL was dragging around a legacy of heavy, high wattage, steam powered CPUs. Microsoft had made its living for two decades consuming INTEL's Moore Law CPU cycles in ever more inefficient ways to implement MS Office, the ultimate keyboard and mouse-centric app.
The high growth niche of computers is now tablets, primarily the Apple iPad. The Consumer Desktop business looks like Detroit. Notebooks are growing only single digits. Netbooks are headed for eight-track player land.
Furthermore Steve didn't just invent a new segment. He erected a very high barrier to competition. He bought a low-power fabless CPU company called Palo Alto Semiconductor and directed them to strip out all the power consuming circuits from their ARM implementation. Two years he owned the world's most power efficient CPU by a substantial margin.
Those who choose to compete in the tablet space have two CPU choices, INTEL ATOM or plain vanilla ARM. The red sunsets in January 2011 will be from the post-Christmas bonfires of ATOM equipped tablet inventory such as HP's Slate Windows 7. To take market share your most distinguishing feature must not be possessing half the running time of the market leader.
ARM equipped tablets will not fare much better. As best as can be determined from outside Apple, TI's OMAP, Qualcom's Snapdragon, and all the other ARM clones consume as much as twice the power as Steve's A4 ARM implementation. Furthermore he can tune his CPU decisions to his tablet design. He can tune his OS optimization to his CPU design. The result is a very tall hill for competitors to climb.
INTEL responded to the ARM threat by acquiring Infineon's ARM making division in August 2010. The move appears reminiscent of the wounded DEC's attempt to enter the PC business by reselling Radio Shack computers as DECstations. The acquisition came after INTEL sold its previous ARM venture to Marvell in 2006. Somebody's crystal ball is broken.
Marvell Multi-core ARM Nibbles INTEL's Cloud Computing
Not only did INTEL sell a business for $600M that it had to buy back for $1.4B, but the customer for their previous ARM business has now targeted INTEL's high-end server business. In November Marvell announced a quad-core ARM CPU that consumes one fifth the power consumption of an INTEL x86 solution.
Abu Dhabi financed Calxeda has completed a design for a similar chip. ZT Systems has announced a 1U 16-core ARM based server that consumes less than 80W.
However, INTEL's lead in the high-end floating point scientific market is safely protected by the Itanium.
Does INTEL Become DEC?
In a word, NO.
DEC's business was minicomputers. When minicomputers ceased to exist, so did DEC. DEC was also crippled by a CEO who lost his crystal ball around 1977. (That seeing the future skill is very rare and does not last forever.)
However INTEL is really two businesses, fabulous fabs and legacy microprocessor designs. Even faced with Patterson's Walls, x86 sales will do a nice replacement business for years. INTEL may enter a Xerox or Kodak like business phase producing commodity ARM chips ending in an acquisition by China or Dubai, but probably not.
It is more likely that INTEL's other business of fabulous fabs will continue to prosper as the counterbalance to TSMC and GlobalFoundries. INTEL already does leading edge foundry work for IBM and Achronix.
On the down side, the foundry business is capital intensive, and the primary competitors tap government treasuries for their Capex. In addition foundry profits are nothing like the 65% gross margin of the leading edge CPU business.
INTEL could buy ARM. The price would probably be between $4B and $8B, but an acquisition would probably trigger anti-trust concerns.
INTEL and TOMI(tm)
If INTEL wants to remain in the high gross margin CPU business. They must have a new architecture. Unfortunately INTEL has a bad history of developing new CPUs. The iAPX 432 was supposed to be a 32-bit replacement for the 8086 in 1981. Two decades later the Itanium was supposed to be a 64-bit replacement for the Pentium. Both were embarrassing expensive belly flops.
Computer architecture is pretty tough, even for a behemoth like INTEL. Fortunately it was even tougher for their competitors, but even the x86 cannot escape Patterson.
The TOMI architecture is the only new CPU architecture in a decade. It seems to nicely solve all three of Patterson's Walls, and it does so on inexpensive DRAM processes. It makes ARM look like the 20 year technology it really is.
Abu Dhabi thinks computers are the new oil. If Steve correctly foresees a computer future chock full of lightweight, wireless, power sipping gadgets, then TOMI could be Ghawar.
INTEL could build it through their current joint venture with Micron. TOMI runs anything that can compile with gcc. It scales nicely from tablet size to cloud server size chips. Its acquisition would probably pass SEC scrutiny. It could maintain INTEL prosperity for another two decades.
It will also never happen. Noyce is dead. Grove is an old man. Moore is an old man. So is INTEL.