SILICON HEART

If you look at the core of a microprocessor, it might seem like an aerial view of a patchwork of fields, a clutter of jagged ends vying for space with angular lines. It all comes down to this: an area, a few square millimeters wide that houses the engine, nay, the powerhouse of the digital revolution. Ubiquitous yet invisible, 30 processors tick away in any average household be it in the remote, refrigerator or your fuzzy logic-driven washing machine. We've come a long way from the vacuum-tube machines that would fit snugly into an oversized building, to the Compaq on your desk, which incidentally packs more computing power than the rocket that landed Neil Armstrong on the moon. Thank the Integrated Circuit (IC), a truly disruptive technology, in every sense of the word, for the processor, as we know it today.

With the IC the processor

just kept getting smaller

and its use expanded dramatically. Today there's one in anything electronic. The business as a result is populated with niche players. But who are the leading players who shape the exponential growth in the processor market? How long can current technology keep pace with Moore's famous observation? And where is the processor headed? These are some of the questions critical to the digital economy, indeed the economy as a whole, and the future.

RISCy Business

In the world of processors, there is one major distinction resting on the design philosophy: RISC (Reduced Instruction Set Computer) and (Complex Instruction Set Computer). John Cocke of IBM Research in Yorktown, New York, originated the RISC concept challenging the existing CISC architecture. CISC used complex and detailed instructions to compute and consumed less time to accomplish a task.

The business advantage of RISC at that time was that with a set of simple instructions one could build, relatively easily, a small processor that consumed little power. SGI, Sun Microsystems, Hewlett Packard, Digital and IBM jumped into the fray. In fact, in the eighties, when most of these companies entered the RISC market, one could thrash out the beginnings of a viable RISC design just as you did a course project in a university computer science class. The major CISC manufacturer rose to be and remains Intel, along with the many clone companies it inspired like Advanced Micro Devices (AMD), Integrated Device Technology (IDT), Cyrix(erstwhile National Semiconductor chip division now owned by Via Technologies) and Rise Technologies. Though, to be fair, AMD is trying to outgrow the incubus. Other important examples of the CISC design include the Motorola 68000 (68k) family used initially in Macintosh systems and Digital Equipment Corporation's (now owned by Compaq) VAX (virtual architecture).

Over the years, however, performance demands have taken away the simplicity and consequently the cost advantage. Both schools have borrowed heavily from each other. The difference now, believe experts, is really definitional rather than technological. Dr Pradeep K Dubey, Research Staff Member, IBM Solutions Research Center, New Delhi, says, "It's not as fundamental a difference as it is made out to be. People were claiming all kinds of things in the RISC category, which were orthogonal to being RISC. Essentially speaking, (RISC is) a single load store type of architecture with a simple reduced instruction set. You could do all those things with CISC as well."

This is not to say that the RISC design did not serve it's purpose. The first 64-bit processor, the fastest processor available today, and virtually all processors comprising the enterprise class systems and high-end graphics workstations owe their allegiance to RISC. But hard economic realities are compelling the players to review their business strategies. Further, to add to their worries, chip behemoth Intel is eyeing their sheltered and captive market. Till now, Intel has been cranking out performance from it's IA-32 line pretty much in tune with Moore's law, snapping at the heels of companies like Sun and MIPS that supply processors for mid-range servers and workstations.

By mid-2000, Intel is expected to release a 64 bit processor (Intel 64 bit architecture, IA 64) code named Merced, aimed at the very high-end.

"Merced has been taped out and we will supply engineering samples to our key customers and partners in about three months or so and our target continues to be production launch by the second half of the next year", says Atul Vijaykar, Director, South Asia, Intel Corporation.

"Tape-out" is industry jargon for a chip design that has been completed and sent to the factory. The next development stage is called 'first Silicon' or when the first prototypes are sent out to partners and customers. There is typically a one to two month gap between tape-out and first silicon. In May 1998, Intel had put back the launch by six months from end of 1999 to the middle of 2000.The Merced tape-out is significant to quell reports that it may be delayed further.

Even then analysts believe that for a completely new architecture, one year from first silicon to first systems, is rather optimistic. A more real scenario would see systems based on IA 64 appearing in the fourth quarter of 2000. A more real scenario would see systems based on IA 64 appearing in the fourth quarter of 2000.

The RISC League

While outlining the RISC sce-nario, Digital Equipment Corporation's (now acquired by Compaq) Alpha processor is a good place to start, for it is credited with having started the exodus to 64 bit computing. The Alpha was meant to be a 25 year architecture that would scale a thousand times in performance during it's lifetime. After it's announcement in February 1991, every major RISC player moved towards 64 bit computing. The only company conspicuous by it's absence in this space was Intel.

The recent acquisition of Digital by Compaq has given a fillip to Alpha with Compaq assuring customers that it will back Alpha and use it in place of MIPS chips for new servers from it's Tandem division. That is good news for under Digital, Alpha's stellar performance record was in stark contrast to it's marketing. "Thanks to the merger, we are ahead of our roadmap," says Suresh Babu, Business Manager, High Performance Servers, Compaq Computer (India) Pvt Ltd, "the current Alpha architecture called EV6 has been announced ahead of schedule and with it we have doubled the performance of the previous architecture, the EV56.

"

The Alpha along with HP's PA RISC have been the performance stars of the RISC galaxy. Iluminata Inc, a US based IT consulting firm carried out an interesting study which tracked the performance of RISC processors vis a vis IA 32. It concluded that over the years very few RISC players have lived up to their performance claims. Alpha with PA-RISC very close at it's heels are best suited for these integer intensive computing defined as commercial, server and computing tasks. IA 32 is the value and volume winner and beats MIPS, POWER and SPARC in absolute terms. In the floating point (read calculations required for graphics intensive applications) arena it's Alpha, PA_RISC, POWER, MIPS and SPARC, in that order. IA 32 comes in last but still makes a strong value play in the technical workstation space.

The other performance leader,

HP, has, since the inception of PA-RISC in 1986, been a pioneer of sorts in RISC technology for commercial applications. The company's current high-end and enterprise class systems are based on the PA-8500, the third major generation of the PA-8000 family. However, in spite of the performance it has wrung out of the PA-RISC family, HP maintains that it costs $3 billion in R&D, design and fabrication to produce one iteration of an advanced RISC processor. It is this cost that drove the company to tie up with Intel, in 1994, to develop a 64 bit Instruction Set Architecture which serves as the technology foundation for IA-64. Interestingly, two years ago, HP had felt the same way about operating systems development costs and had tied up with Santa Cruz Operations (SCO) to develop a 64 bit Unix called 3DA. The project was later shelved. The IA 64 will initially coexist and eventually replace PA-RISC in HP systems.

Like HP, SGI is looking towards IA-64 for it's future processor needs. Indicative of it's future plans and in a major departure from it's MIPS and Unix heritage, the beleaguered visual computing major has announced workstations and servers running Windows NT on Intel's Xeon chips. Like HP, SGI is also extending the life of it's MIPS processors because of delays in Merced. SGI is currently building servers around the R12000 with new MIPS chips expected every six to nine months.

IBM's POWER and PowerPC de-

signs have sadly not lived up to the name. It's all the more surprising for a company that has been a design pioneer. Most recently, they have been the first to announce, implement and bring to market copper processors. As far as the POWER line is concerned, IBM has planned out multiple iterations of the POWER3 and plans to add a high-speed interface that will make it easier to run multiple chips together. Further, the company announced it's "GIGA Processor" last year, which should come to market as POWER4.With the GIGA Processor, IBM desperately needs to gain a performance edge to regain some of it's lost luster.

The PowerPC architecture is interesting for just what it seeks to achieve, to meet the very different needs of the members of the PowerPC alliance-IBM, Motorola and Apple servers for IBM, multimedia applications for Motorola and desktop computing for Apple. Another client in this list is Nintendo. In a billion dollar deal, IBM and Nintendo have come together to put the PowerPC into the latter's next generation gaming console code named Dolphin. IBM has said that the processor, code named Gekko, will be heavily customized for Nintendo. It will run at 400 MHz and will be built using 0.18 micron copper process. When the new platform hits the market it will compete with the Sega Dreamcast which uses a Hitachi processor and Sony's Playstation 2000 which runs on a custom designed chip by Sony and Toshiba called the Emotion Engine. The Playstation was earlier based on a MIPS processor.

The real hope for the PowerPC in the home computing space, comes in the form of AltiVec technology-a multimedia instruction set akin to Intel's MMX, of which Dubey is a key architect. AltiVec expands the current PowerPC architecture through the addition of a 128-bit vector execution unit, which operates concurrently with existing integer and floating point units. The new engine allows for simultaneous execution of up to 16 operations in a single clock cycle. As with all such instruction sets, success will depend on support from the developer community that must write programs that utilize the instructions.

Sun's game has been consis-

tency. Although it's not a performance leader, it has listened to it's customers very closely and optimized it's systems to meet their needs. It is also the only processor company that has utilized the same basic core in three different processor lines.

Between Intel and a hard place

As far as Intel is concerned their roadmap for IA-64 kicks off with Merced in 2000 followed by McKinley in late 2001 or early 2002. In 2002 Intel splits the IA-64 family into two lines based on pure and price performance. When it comes to bringing a new architecture to market, no one is better placed in terms of economics and industry support than Intel. "I believe from a 64 bit raw performance standpoint we will be absolutely competitive if not leaders in a couple of years. Certainly by McKinley we'll be right there," says Atul Vijaykar, Director, South Asia, Intel Corporation.

For the RISC Players, it really boils down to joining them since it one cannot beat them. Compaq has two business divisions, the Industry Standards Organization, which markets Intel based servers and the High Performance Division dealing in Alpha based systems. The former will ship IA-64 based systems as when the processor is available. Babu, however, feels that they are not antagonistic since each caters to a distinct market. "There existed a performance gap between Intel and Alpha and that will continue. There exists a market for both these architectures," he says. Industry experts believe that for the near future, Alpha will continue to hold on to it's performance edge.

As far as HP is concerned it doesn't even publish a separate roadmap any longer. PA-RISC is intertwined with the corresponding IA-64 products. The company already has systems like the N-class servers, which can be upgraded to Merced as and when it ships. Fellow first adopter, SGI's chief scientist, John Mashey has gone on record as saying that MIPS based machines will continue well into the middle of the next decade to meet customer demand, in tandem with the IA-64 based machines.

The only company with no hardware plans for IA-64 is Sun. The company has committed, however, to port Solaris (Sun's proprietary Unix version) to IA-64, which is seen by many as a way hedging it's bets. The sad fact, for Sun and some of the other players is that they don't have the market share to lose. "The economics are very significant because the costs of maintaining a proprietary architecture are so prohibitive. By 2002 some of the competition would have a tough case justifying the economics of keeping a proprietary line," says Vijaykar. To cut down costs both Sun and Compaq have out sourced manufacturing. Intel and Samsung make Alpha, while the UltraSPARC line is manufactured by Texas Instruments.

However, Merced could still trip over itself. "IA-64 may fail because of the complexity required at the compiler level. They will need to show significant performance gain for people to shift architectures'" says an industry analyst based outside Silicon Valley.

The nine month slip in schedule has given the RISC players a small window of opportunity. Merced will now hit the market amidst much stronger competition-awkwardly enough, from it's own brethren as well. By the time Merced arrives at it's 800 MHz frequency, Intel would have 32-bit processors available at that speed. It is in many ways a test bed, which is why analysts expect McKinley to be the real breakthrough product for Intel. Buyers of this class of processors are conservative and would wait till Intel irons out issues.

The X86 Business

The 32-bit desktop and server market is not multifaceted like the high-end 64-bit space, since there are very few players other than Intel. It has gone from strength to strength ever since IBM selected the 8086 and 8088 for it's PCs. So synonymous is Intel with this market that even the term x86 is derived from it's naming convention where the last two digits are eight and six(80286 and 80288). In this intensely price sensitive and competitive market Intel's economic advantages far outweigh any other competitor.

Despite it's hold over the market, Intel did have it's dominance threatened last year. The company was caught napping when the sub-$1000 PC market began to boom. Competitors like erstwhile National Semiconductor division Cyrix (now acquired by Taiwanese chipset manufacturer Via Technologies Inc), IDT and AMD were better placed with their cheaper chips. Intel reacted quickly by splitting it's product line into the server class Xeon', performance grade Pentium II (now Pentium III) and entry-level Celeron processors. Yet they did not get the low-end strategy right the first time. Intel committed a blunder by bringing to market a cache-less Celeron processor which was a sluggish performer.

It was not until the introduction of the Celeron, a processor with a 128KB on-die cache that things began to look up. With it's broad product portfolio Intel unleashed an entry-level price war. Throughout Intel's average selling price (ASP) has hovered around the $250 mark. It's closest competitor AMD earns an ASP $78, Cyrix $49 and IDT$39. Since the launch of the Pentium III processor the frequency of Intel's price cuts have gone from quarterly to monthly. "The incidence of price drops is a way of offering better value to the user, you are getting better technology at a lower price. Now, because of various factors the frequency of these price drops has been much more aggressive. It's a question of competitive forces but ultimately, it benefits the end user," opines Kanwaljit Singh, General Manager, Marketing, Intel (India). Linley Gwennap, Vice President, Publications at MicroDesign Resources (publishers of Microprocessor Report) called Intel's strategy as it's "scorched earth" policy in a May editorial.

Consequently, National Semi-

conductor has opted out of the chip business, and so has IDT, which is on the block. Rise is stuck with chips whose manufacturing costs exceed their market value.

Ever since National put Cyrix up for sale, rumors were rife that IBM would step in and fulfill the need for an x86 microprocessor core in it's product portfolio. But amazingly, a company of the caliber and size of IBM opted to stay out with it's Microelectronics Group concentrating on embedded applications. In July, IDT announced that it plans to license or sell it's WinChip technology and assets of it's Centaur Design chip subsidiary by the end of the current fiscal quarter.

The scenario leaves only AMD as a competitor of any reckoning. It now has a performance grade product, in a market where the margins are higher, beyond the entry-level market. AMD's Athlon or K7 as it was code named, by most accounts, outperforms Intel's PIII processor. The real question is whether AMD can avoid the manufacturing problems that plagued the K-6 line and drive home the advantage.

Intel's response to the Athlon code named Coppermine will not appear till November because of delays.

Cyrix under Via could be significant though it still remains a bit of an unknown quantity.

According to the business plans announced by the company, National will continue to produce the Cyrix MII chip for Via at speeds of 400 and 433 MHz. Perhaps even the next generation Cyrix processor code named Gobi , that is expected to hit the market by December and will fit into the same motherboards as Celeron processors. Via has the credentials to present competition for Intel. It is part of the Taiwan-based conglomerate,the Formosa Plastics Group and also owns the motherboard maker, First International Computers (FIC) and DRAM manufacturer Nanya Technology. Those connections could be very helpful when Via's David takes on Intel's Goliath.

"We can't rule AMD and Via out. The market is so large and lucrative that we will continue to see competition," says Vijaykar. His statement is echoed in Gwennap's editorial, when he says,

"Using the low end as a base to attack the performance segment is no longer a viable plan. This change has led to the common perception that competing with Intel is a money losing proposition...a vendor with the technology to compete in multiple segments at once can easily and profitably supply 10 to 20 percent of the PC processor market." The lure of this $20 billion market will continue attract companies like moths to a flame.

End game

Whether in the high-end or low-end it would be hard for other players to prosper in Intel's shadow. In the near future there seems little that can change the status quo. Both Babu and Vijaykar agree that in the next few years ecommerce will determine end-user and corporate computing needs. The PC as we know it will continue to play a part though perhaps not as central as today. "The desktop will aid other ways of information access and flow. Devices will extend beyond the desktop into things like the cellphone," says Babu. At the back-end the demand for power and scalability coupled with reliability will continue to grow as businesses transform into non-stop

ebusinesses.

The world will also get a lot jazzier. Information will be synonymous with multimedia. Requiring smaller processors to do more. Whether the PC goes the way of Larry Ellison's network computer or follows Michael Dell's "cheap PC with a fat pipe" vision, the future will belong to the company that makes the chip for a Palm-like device that understands your voice.

ABHAY SINGH,



Cyber News Service.