Disposing of 'Intel Inside'

[tw]

Intel may be in serious trouble. It’s just not obvious on the spread sheets and will not be apparent for four and more years later when work performed this year starts appearing on those spread sheets. Symptoms start when you look at its top manager. Paul S. Otellini is the first non-engineer to run the company. Otellini studied economics in U of San Francisco and then an MBA from University of California at Berkeley. Many new employees include software developers, sociologists, ethnographers, even doctors to help develop products. He lays particular emphasis on marketing expertise because he thinks the only way Intel can succeed in new markets is by communicating more clearly what the technology can do for customers. This is the same mentality that Lucent used to even undermine the Bell Labs. Obvious application only undermined and destroyed the genius in Bell Labs. Otellini is talking the same way having no experience where the work gets done.

Instead of remaining focused on PCs and other processor related functions, Otellini is changing Intel to play a key technological role in a half-dozen fields, including consumer electronics, wireless communications, and health care. And rather than just microprocessors, he wants Intel to create all kinds of chips, as well as software, and then meld them together into what he calls "platforms."

Intel got where it is by doing what it did productively AND nurturing others to do same or compatible functions into a system. For example, computer chip sets, modems, video cards, etc are all contributors to making Intel innovative and productive. Intel defined a strategic objective, then designed and controlled the heart of a system design that all others contribute to. Some of those successes are USB, AGP video, PCI bus, power supply standards, video standards, Ethernet, wireless computing, most of the mobile computer functions, and the so many ways of powering and powering down a system.

So Otellini will take all this work from others to make Intel a better company? Somehow compete against its own partners?

Some of Intel's problems were already created by Barrett - the previous boss who followed a legendary Andy Grove. But symptoms of a company in even deeper trouble are demonstrated by how Otellini will solve these problems. They intend to blow up Intel's branding; a fifth-best-known brand worldwide. Intel will "clear out the cobwebs" and kill off many Grove-era creations. Intel Inside? The Pentium brand? The widely recognized dropped "e" in Intel's corporate logo? All will be eliminated - as if redesigning fenders on a GM car will fix the problems inside. But remember, Otellini is an MBA and does not come from where the work gets done. His solutions will only be what he can see - not where problems really exist. Somehow rebranding Intel will solve everything?

Intel's problems started with Pentium 4 when the architecture got so large that a chip could not be produced reasonably. The solution was to optimize compilers so that Pentium 4 would remain faster. Intel's genius was in processing and manufacturing that others could not do equal or better. Smaller transistors, less heat, faster switching times, introduction of the most advanced new process technologies and materials. This combined with partners who were some of this nation's best corporations - Microsoft, Dell, Compaq, HP, the so many video and BIOS manufacturers, modem designers, chipset designers, memory manufacturers, and networking companies. All partnered with or had their designs defined by Intel.

Well Intel started to hit a brick wall. Its advanced manufacturing abilities were confronting limits of transistors. Its architects had not kept up their microarchitecture. And its management under Craig Barrett apparently were too busy looking at spread sheets to, instead, see a technology barrier approaching.

Suddenly last year, Intel canceled all new chip designs because heat and other problems made the Pentium too difficult. Suddenly the new Intel architecture was created due to an ignored problem - and created in an emergency solution - a dual core chip design. This being ironic because Dr Craig Barrett's background is material science with over 40 papers in the science. He should have seen it coming.

Well Intel recently hyped a whole new microarchitecture to fix the weakest part of a Pentium design. Will it solve Intel's recent loss of one title - 'fastest server processor' - to AMD?

So Intel must address its product line - technically. Instead Otellini wants to diversity the company into doing what its so many product partners accomplish. Somehow he thinks the GM corporate strategy will save Intel. Somehow putting Intel in direct competition with Texas Instruments and other so innovative companies will make Intel better? The guy must be an MBA in the tradition of Carly Fiorina - who received a same early analysis from this author for same reasons. She too hyped the word innovation while doing the classic bean counter shuffle.

This is not good for Intel, Intel consumers, or America. It reminds me of how AT&T decided all other computer manufacturers were incompetent and that AT&T would become the new power in computing. AT&T was also destroyed by MBA management with the same new philosophies. Their soulution - buy and destroy NCR.

AMD - a German processor manufacturer - is quietly eating away at most of Intel's product line. Intel's only promising processor markets are in its Pentium M series - mobile computing. Even its non-volatile memory business may be in trouble without a breakthrough new technology product.

[Torrere]

AMD is based in Sunnydale, California. California is not now, nor has it ever been a part of Germany.

Intel's problems did not start with the Pentium 4. The design of the Pentium 4 was motivated by marketing: because consumers judged CPUs by clock speeds, they sacrificed speed and power for high clock speeds. The early 1.5 GHz Pentiums 4s were, if I recall correctly, slower than the 1.0 GHz Pentium 3s.

[tw]

Originally Posted by Torrere

AMD is based in Sunnydale, California. California is not now, nor has it ever been a part of Germany.

I recall AMD (before making processors) was a TX company - one of the little companies in the shadow of TI. But AMDs processors are now designed and manufactured in Dresden Germany. This is also where AMD is building their newest fab. For obvious reasons, AMD does not much discuss where these processors are designed and manufactured. Yes AMD also has fabs throughout the US. But...

Intel's problems did not start with the Pentium 4. The design of the Pentium 4 was motivated by marketing: because consumers judged CPUs by clock speeds, ...

That is not the story of Pentium 4. Architects designed a chip choke full of ideas including (if I remember) a preprocessor that would rearrange op-codes. As the chip was designed and taped out, it was clear that the IC would be about 3 or 4 times too large. Reasons behind the expression 'too large' are too numerous to list here.

Therefore Pentium 4 architect solution was to remove as many functions as possible. One operation that suffered from a 'diet' were SI/DI operations. To make this 'Jenny Craig' processor work fast enough, compilers had to arrange code in proper order.

Now spin doctors entered. A Pentium 4 executing non-optimized code would run quite slow. For example, SI/DI functions would drag that processor into long wait states if code was not properly ordered by a compiler.

At that time, AMD processors were still significantly slower but were selling at discount prices to attract customers. After the K-5 fiasco, AMD was finally making a marketable processor. But AMD was not making profits since their still inferior processor had to sell at discount prices. K-6 was still a dog. But the first dog that demonstrated AMD could finally design a processor.

To discover reality, always start with the technology - the details. So many instead jump on Rush Limbaugh type hype - never first ask for those technical details. Never challenge those who promote MBA school type reasons. I have been through this same 'myths verses reality' discussion on '8080/8085 verse Z-80', '80x86 verses 68000', and 'Pentium verses PowerPC' nonsense. In each case, the hype about marketing or other myths were proven false. The accurate answer was always found in technical details and who was top management.

For example, so many would praise the 68000 because writing code was easier. IOW they were myopic. All those 'flexible' op codes also meant hardware was more difficult to manufacture and design. As a result, the 680x0 (68050?) self destructed because the 68000 architecture was that poor. But you could not tell that to programmers who typically had no science education AND who did not see the entire picture.

Same reasoning explains why architecture of a Pentium 4 design was its weakness. A weakness that was masked by so many other and superior aspects within Intel Corporation. A weakness all but recently acknowledged by Intel with their recently announced new cores - said to have 'superior' microarchitectures. Without those details, then one really does not know what is and is not better. Those details suggest serious problems for Intel - made worse by a new top executive who talks too much like Carly Fiorina.

[tw]

Originally Posted by jaguar

Demos of the latest chips I saw were spanking AMD. AMD is also now a generation behind in fabs and Intel will have it's next gen out before AMD does.

Those demos were a new Intel design verses an existing AMD product. Then AMD announced their new processors. Whereas Intel is already doing 65 nm chips, AMD is still testing that technology. Again, a crown jewel in the Intel product - superior material processing designs resulting in faster transistors.

AMD is currently winning at the upper end (server) market. Their HyperTransport has been successful and may even eliminate the North bridge. HyperTransport is on the board signalling at gigahertz - which should raise your eyebrow IF you appreciate the details.

Too many recent Intel designs - which once rarely happened - are missing design deadlines. Mistakes have been found too late in the design process. Intel equivalent to HyperTransport, called Coherent Scalable Transport, has been delayed for both Pentium and Itanium architectures. This by itself means little. But lately Intel had been making too many such mistakes - far more than the Intel that was run by Grove, Moore, and (forgot his name).

These details today are what cause spread sheet to reflect those realities four and ten years later.

Cray has recently assessed AMD and Intel products. Their bottom line is that the long term projections from both companies means AMD wins. It suggests what is happening in upper end product markets. But then Cray has never been a successful company once the MBAs decided to use GaAs technology rather than innovate - massive parallel architectures. For example, Cray lost $200million in 2004. They have repeatedly been close to bankruptcy because MBAs took over the company. Cray is playing catchup even to their own employee, Stephen Chen, who proposed a massively parallel architecture (that MBAs rejected) and who is now running a Cray competitor - Galactic Computing. What Cray is calling a major innovation - blade processors - well, if you have a grasp of details, then you appreciate why that may be more hype and less innovation.

IOW when Cray choses AMD, well, one must also look at who might be doing the choosing - another detail.

[tw]

Originally Posted by Torrere

AMD is based in Sunnydale, California. California is not now, nor has it ever been a part of Germany.

From CBSMarketWatch.com on 1 Jun 2006 entitled "AMD raising output in bid to boost share":

Advanced Micro Devices Inc., the world's No. 2 provider of chips that power personal computers and servers, is ramping up factory output in an attempt to grab more market share from larger rival Intel Corp., the company's top executives said Thursday.

Daryl Ostrander, the company's senior vice president of manufacturing, said that AMD's annual wafer starts will increase fourfold over the next three years at its factories in Germany.

"We will be fully positioned to serve one-third of the (chip) market by 2008," he told investors at the company's analyst day in Sunnyvale, Calif.

[dar512]

Originally Posted by tw

AMD's annual wafer starts will increase fourfold over the next three years at its factories in Germany

Coca-Cola has manufacturing plants around the world. That doesn't make CC a Saudi or Russian company.

[tw]

Originally Posted by dar512

Coca-Cola has manufacturing plants around the world. That doesn't make CC a Saudi or Russian company.

Boeing corporate offices are in Chicago. Therefore Boeing aircraft are a product of Illinois? Of course not. AMD’s processors are manufacturered in Germany - a fact that most certainly surprises most every lurker since AMD does not advertise it. I really don't care where the CEO hangs his hat. Only bean counter types feel that is important. A product person instead asks about the important people - the little people. Boeing aircraft are built in Aisa, Washington, et al. Only those who don't know how work gets done (finance types) would call Boeing aircraft a product of Illinois.

Meanwhile, a more interesting point is AMD's strategic objectives. A tripling of production? AMD implies the MBA now running Intel will not solve Intel's management problems.

[dar512]

Originally Posted by tw

AMD - a German processor manufacturer

I still think this is misleading. It implies that AMD is a German company.

I think the competition between AMD and Intel is a good thing. I don't get your hand-wringing over Intel's current issues.

[tw]

From EE Times of 12 Jun 2006:

AMD to invest $2.5 billion in new Dresden fab
Advanced Micro Devices said Monday (May 29) it will invest $2.5 billion in its fabs here [Dresden], adding a new 300-mm fab to replace an existing 200-mm facility that will be closed. AMD said the expansion will quadruple its processor production here within the next 30 months.

Construction of the new Fab 38 will not require an entirely new factory. Instead, AMD plans a thorough transformation of Fab 30. ... At the same time, the company will increase the capacity of existing 300-mm fab 36 by 25 percent to 25,000 monthly wafer starts. Fab 36 was launched in October 2005.

... By the end of next year, the new facility will produce 20,000 wafer starts a month at full capacity.

Combined, AMD's Dresden fabs will produce 45,000 wafers a month. According to plans announced here, the new fabs initially will produce 65-nm chips, but AMD will switch to 45-nm process technology by mid-2008 and eventually move to 32 nm.

[tw]

Intel is addressing previous mistakes directly traceable to top management. For the first time in a while, Intel may actually succeed in some engineering accomplishments - may have products that turn back the AMD challenge. Intel is testing 45 nm IC designs when the competition is still struggling with 65 nm designs. Intel is finally addressing multiprocessing as a solution to what would otherwise require a 10 Ghz processor.

Tomorrow, Intel will announce an accomplishment that other have been struggling with for years. IBM was actually in production when their high-K (Dupont? designed) material pealed off ICs during manufacturing. Intel will be using a high-K material in 6 months in processors. AMD-IBM-Toshiba consortium probably will not have same for at least another year.

Original transistors were germanium. Silicon replaced germanium for one key reason - glass - silicon dioxide. Germanium oxide (as a glass insulator) was not possible.

Push (or turn) on a facuet so that water enters from mains and exits via tap. Transistors work similar. On one side of glass (silicon dioxide) is a channel that permits electrons to enter on one side and leave on other channel end. To switch (on and off) that channel, electrons are removed or piled on the other side of that glass. That other side is called the 'gate'. To make a transistor run faster, that glass was made thinner. Less electrons into the gate could switch that channel faster. Well that glass is as thin as 3 atoms thick. Therefore electrons piled into gate were leaking across glass. You feel that leaking as heat; CPUs created as much heat as a 60 or 100 watt incandescant bulb.

Semicondutor manufactures have done just about everything to solve that smaller and thinner glass problem. But everyone knew what was needed. Glass needed a high-K material. Intel has been experimenting for years with hafnium - having eliminate other possible solutions long ago. In six months, Pentiums using that new material will be sold maybe using 45 nm transistors.

The actual buzz words are high-K and low-K materials. To make that glass requires higher-K materials so that glass can be thicker; gate operates like 3 atom thick glass without electrons leaking through that glass. Low-K materials means signals travel across the IC faster. Low-K materials are not as necessary since 10 Ghz processors are not in the pipeline. But high-K materials have again averted a brick wall. An innovation schedule called Moore's Law is met again.

[jaguar]

Yea the netburst arch sucked balls, everyone knows that but the newest chips are looking might fine, centrino was very nice, the core duo and upcoming quad ones that will be in the new power macs are looking like very nice procs indeed. Demos of the latest chips I saw were spanking AMD. AMD is also now a generation behind in fabs and Intel will have it's next gen out before AMD does.

[jaguar]

Which new procs did AMD announce? I'm not aware of any AMD proc coming out before them. HT is nice but it's not like AMD developed it alone and the CSI (Scalable Coherent Interface) delay was the fault of the ill-fated Intel India operation which has now been closed. CSI also has some notable advantages over HT and latency is way, way lower.

In a nearer timeframe Merom and Conroe are shaping up nicely and as it stands AMD has crap all in the mobile arena which is becoming larger and larger. Intel still has 80% of the market. AMD has done well to get where they are but I expect it's going to be neck and neck on peformance over the next couple of years.

Put it another way, noone can deliver chips like Intel with the exception of maybe IBM, Apple would not have picked Intel if they didn't see a damn strong line of processors coming up.

[tw]

Originally Posted by jaguar

Which new procs did AMD announce?

I only heard a report of the AMD presentation which, for me, would be too soon for reports with details.

... AMD has crap all in the mobile arena which is becoming larger and larger. ... Put it another way, noone can deliver chips like Intel.

Intel's move into the Pentium M line is probably their most productive move. Intel was losing that market to other 'clone' Pentium manufacturers. This is why a major company wants smaller competitors. That success by the competition got Intel's attention in a market Intel had mostly ignored. That revelation created the Intel Pentium M line. Technology from that processor line is being used in other Intel products as Intel 'discovered' they had a new "unexpected?" energy problem.

But what troubles me have been weak architecture designs in their upper end processors, their failure to make Itanium competitive, their complete avoidance for a 64 bit Pentium, that Intel was surprised by a sudden heat increase and an unexpected need for multi-cores, the Rambus memory fiasco, CSI delays, failures to dominate in the wireless markets (Wi-Fi and Wi-Max), failures to break into the mobile phone business which could be the disruptive technology to replace mobile computing, Intel’s failure to remain competitive in the non-volatile memory market (although rumors persist that Intel may reenter that market with a whole new technology), and numerous other misses.

Especially worrying is that Intel appears to not have a long term strategic plan as it did 10+ years ago that resulted in North/South Bridge, AGP, PCI integration, USB, flat memory models, the many powerdown methods, and simpler peripherals by putting more compression/decompression functions within the processor. Where is the new strategic objective for hardware? It appears Intel has none; therefore shotgunning innovation rather than integrating it in a fashion that made Intel so successful.

Intel's success that included going smaller faster - going directly to 65 nm technology and making it work, its ability to finally close a deal with Apple - Intel has been repeatedly talking and building an Apple with Intel chips, and their success with Pentium M all are achievements. And of course, nobody can mass produce so many more advanced chips so quickly like Intel. That manufacturing and material processing ability has always been an Intel crown jewel which is why previous posts about strained silicon, ovonyx unified memory, high K-dielectrics, and spintronics always include references to Intel.

But AMD keeps eating more of Intel's pie - which Zilog Z-80, Motorola 68000, IBM Power PC, and Transmeta, etc all promised and never could even come close to delivering. AMD is the first competitor to remain and even grow competitive to Intel. Not because AMD is doing things better. I fear a severe management weakness starting to take hold in Intel. This because the new top man talks too much like Carly Fiorina and John Young; both caused bad times in HP.

[tw]

Intel to Undergo Broad Restructuring
... Intel Corp. is planning a broad restructuring that will focus on every aspect of the chip maker's operations, Chief Executive Paul Otellini said Thursday. ...

The restructuring comes as Intel has seriously stumbled in recent months. A host of new products from Advanced Micro Devices Inc. has allowed Intel's biggest competitor in PC microprocessors to steal away market share. ...

Building on the success of Intel's Centrino, a bundle of chips and software for mobile computers, Intel earlier this year introduced Viiv, a chip package designed for media PCs. Earlier this week Otellini unveiled a new bundle called vPro, which initially will be aimed at business desktops, but eventually will include mobile machines as well. VPro is slated for release in the third quarter.

VPro sounds like the new dual core processor design when management finally conceded and terminated nearly finished single core processor designs.

[jaguar]

VPro might be the public name for Merom/Crusoe/Woodcrest, the release window fits perfectly, as does the release order (merom last).