Risc Chips: Chip and Change
Risc chips have yet to take the commercial sector by storm and topple the mighty Intel from its desktop throne. But, says Sean Hallahan, the technology is evolving to survive
Hewlett Packard set the Risc ball rolling in the 1980s when it began designing a new chipset and the technology was speedily adopted by other companies. HPs intention was to optimise the number of instructions on a CPU chip.
During the development of the technology, the company ran a series of applications on one of its own HP 3000 machines, a Digital Vax and an Amdahl mainframe. The experiment revealed that all three machines used over 300 instructions, but that less than 100 were regularly used. The logic of this revelation led HP to conclude that it could trim the number of instructions to between 70 and 100.
Although HP pioneered Risc, it did not invent it. Risc (reduced instruction set computer) was actually invented by IBM for its 801 minicomputer, but never used until the launch of the RS/6000 in 1990. The 801 was born out of development work carried out in 1974 by a team headed by senior IBM researcher John Cocke to develop a switching system capable of handling 300 calls per second.
In 1975 the 801 emerged from the project as a standard machine, but IBM failed to capitalise on the development lest it cut across the sales of mainframes based around the Cisc architecture, which were still highly profitable. When asked why IBM did not get into Risc sooner, Cocke made the company?s position clear.
?We had the goose that was laying the golden egg ? it?s very difficult to make any changes when you are doing well. The antitrust group of the US Justice Department [investigating IBM for monopoly practices] wanted to split Ploughkeepsie [an IBM development location] into four groups. I often wonder what would happen if this had occurred.?
It was not until 11 years later that IBM introduced its first commercial Risc-based machine, the RT-PC. The machine was a commercial failure, largely because it was poorly marketed and because IBM did not know where to position the system. The RT-PC, as its name implies, was considered a part of the personal systems division, where it sat uneasily among the Intel-based solutions. The 11-year gap between the 801 and the RT-PC probably cost IBM a small fortune, even if it did protect the more profitable mainframe and PC sales.
But if IBM was slow to see the advantages of Risc, other firms were not. HP, Sun, Digital, Mips, Pyramid and Sequent moved in quickly to exploit the market. Sun in particular was able to capitalise on IBM?s shortsightedness. Its first workstations appeared six months after the RT-PC and were as powerful as IBM?s systems, considerably less expensive and well marketed. Initially Sun recognised what IBM had overlooked ? the demand for Risc-based Unix systems lay in the scientific, engineering and higher education sector rather than in the commercial market.
But by the mid 1980s things had changed. The Risc workstation was no longer seen as being confined to the scientific and engineering ghetto. There were predictions by analysts and confident claims by the Risc vendors, most notably Sun, that Risc-based machines would challenge Intel?s supremacy on the desktop. Intel remained steadfastly wedded to the Cisc (complex instruction set computing) technology which had made it a fortune throughout the 1980s.
In 1993 IBM woke up to the fact that Risc was here to stay and formed an alliance with Apple and Motorola to develop a Risc chip which would be smaller and more powerful than the Intel Pentium. A consortium was formed by the three companies, the Power Open group, which began signing up software developers.
But the Power Open group seems to have pretty much dissolved, with Apple being the only company fully committed to the Power PC chip on the desktop, and IBM the only player committed to the technology on the server side.
In January, IBM, Motorola and Bull announced that they were discontinuing support for Windows NT on the Power PC platform. The following month, in a tit-for-tat move, Microsoft abandoned development work for NT on the Power PC. Any hopes that Power Open had of becoming a standard and a rival to Intel on the desktop were set back, if not destroyed, by these moves.
Martin Burns, Motorola European operations manager for the Power PC, acknowledges that there is little likelihood of the Power PC un- dermining Intel?s dominance on the desktop within the next five years. ?There is not going to be a turnaround in the market in the next five years. We will not see Intel losing 80 per cent of its market share and Motorola gaining it,? he says.
According to Burns, whatever processor a machine is running is irrelevant to the user, providing the user gets the performance he requires at a price he can afford. Intel?s success on the desktop is not based on the power of its processors, but on the fact that the majority of independent software developers have ported their application software to the platform. ?The reason Intel has the market dominance on the desktop is because of the wealth of software available on the 80x86,? he says.
But newer development in the field of computing may yet boost the life of the Risc chip, according to Burns. He believes that Risc chips will play a significant role in the internet, intranet and the terminal replacement markets if and when the network computer (NC) revolution happens.
The concept of thin client computing, espoused by Oracle chairman and chief executive Larry Ellison and bolstered by independent analysts such as Bloor Research, could give Risc a new lease of life. Burns believes that initially thin client machines will replace the large number of dumb terminals that still abound in the corporate environment, but that gradually they will also eat into the Intel desktop market.
One of the issues which NC supporters argue will drive the market forward is cost. The devices connected to the server will cost about $500, according to Ellison and his cohorts. But to produce such a machine means reducing the costs of components, including the microprocessor that powers it. ?One of the good things about Risc technology is that it means we can make a smaller and more powerful chip for well below $50,? says Burns.
Burns? view is supported even by those who might be thought to be hostile to the whole concept of the NC ? the PC manufacturers. Speaking at a con- ference last year, Roger Priebe, communications specialist with Dell?s Advanced Technology Group, outlined the company?s thoughts on the NC. ?To meet the target price of $500, the computer will need a microprocessor on the trailing edge of the price/performance curve, where prices generally fall quickly. Use of a high-performance Intel Pentium or Pentium Pro chips is not currently viable ? an inexpensive computer must use an inexpensive microprocessor.?
He suggested that possible candidates were the NEC VR4300 used in the Nintendo games machines; the R4640 from Integrated Device Technologies comparable to a 100MHz Pentium chip; or the Strongarm processor from Advanced Risc Machines.
Chip architectures tend to leapfrog one another technologically, and Intel believes that it is just as capable of producing powerful solutions as its Risc competitors. ?Risc performance was supposed to be so compelling that it would blow Intel off the desktop, but when we produced the Pentium Pro in 1995 it was the most powerful chip on the market, more powerful even than Alpha,? says Chris Hogg, a market development manager with Intel.
What matters to the user is the base of application software available, says Hogg. ?Apple is still the biggest company shipping Risc on the desktop, but walk into a dealer and see how many application packages are available for the Mac.? He also points out that Apple is adding Pentium processor boards to the Mac to widen the availability of applications.
Firms like Digital and HP have staked their future on selling Risc systems. Digital?s Alpha is generally acknowledged to be the most powerful architecture on the market. According to Martin Finn, commercial director with reseller Hamilton Rentals, it is selling well. While he admits that the sales of Alpha-based systems cannot be compared with the volume sales of Intel-based systems, he thinks Digital is doing well in the server market, particularly in the commercial sector.
The price of Alpha systems has fallen in recent months, and while they are still more expensive than a comparable Intel-based solution, the gap is narrowing. But Finn identifies some drawbacks to the sale of Alpha. ?For some applications we would not recommend Alpha over Intel because it does not perform as well. I don?t want to be drawn on what applications, but if you tried running Microsoft Office on Alpha you would probably be disappointed,? he says.
Finn believes that what is obstructing sales of Alpha is the lack of a 64-bit operating system, Windows NT, to run with the 64-bit Alpha chip. ?There is no point in having a 64-bit chip if you only have a 32-bit operating system to run on it,? he says. Microsoft had originally promised that a 64-bit version of NT would be available during this year, but is now promising that it will be available by the end of the calendar year.
According to Phil Payne, analyst with Sievers Consulting, there are rumours that IBM will incorporate some Risc features in the next generation of its mainframes, but he does not think that ?one technology will wipe out the other?.
There are a number of competing Risc technologies vying to become the industry standard: Digital Alpha; HP Precision Architecture; Sun Sparc; Mips, now a subsidiary of Silicon Graphics and a supplier of chips to a number of Risc vendors; and the Power PC Alliance made up of IBM, Motorola and Apple with support from other companies such as Bull.
According to Finn, the clear market leader is Sun by sheer dint of installed base and aggressive marketing. The Power PC Alliance, which hoped to counter the predominance of Sun, is unravelling. And Apple, which still adheres to the Power PC architecture for its desktop offerings, has far too many financial problems for it to become a serious challenger.
Finn says HP could become a major player once the next iteration of its architecture appears, but that is probably about two years away. ?Who can predict two years ahead in this business?? he asks.
The cost of building a wafer fabrication plant to manufacture a new chip is now put at something between $1 billion and $1.5 billion. To amortise the cost of research and development, a semiconductor company needs to produce chips in massive quantities. The Power PC Alliance attempted to overcome this problem by signing a deal with a major US motor manufacturer, which would then use the chips as embedded systems in its cars. But as Hogg points out, the requirements of embedded chips and desktop chips are totally different.
Risc-based systems were originally designed for the high computational requirements of the scientific and engineering market. Predictions that they would take the commercial market by storm have not come to fruition, although some firms have had a measure of success.
Risc is certainly not dead, nor will it ever be, but it will be some time before it achieves the mass market penetration of Cisc and pushes Intel out of the centre of the picture.