Michael Slater, industry guru and editor and publisher of the US journal Microprocessor Report, in London last week to give his views on the RISC marketplace, had some disparaging comments to make about Intel Corp’s and Digital Equipment Corp’s RISC processor offerings. The Intel 80860 he described as a good implementation of a poorly thought-out […]
Michael Slater, industry guru and editor and publisher of the US journal Microprocessor Report, in London last week to give his views on the RISC marketplace, had some disparaging comments to make about Intel Corp’s and Digital Equipment Corp’s RISC processor offerings. The Intel 80860 he described as a good implementation of a poorly thought-out microprocessor architecture. As a supposedly general-purpose chip, he said, the 80860 falls down on programmability, no thought seemingly having been given to the compiler and operating environment. Slater echoed industry doubts about Intel’s commitment to the unsuccessful device, which Olivetti was going to use to build a workstation until the last minute when it scrapped the whole project in favour of DEC’s Alpha (CI No 1,945); the 80860 also formed the basis of Oki Electric’s server line until it folded last month.
Apart from use in supercomputers and embedded applications, the 80860 is struggling to garner support. Intel was the last to join the RISC bandwaggon, Slater notes – repeating a comment made some time ago by Intel’s Dave House, that RISC was the last hope of the have-nots – now it seems Intel is the have-not of the RISC marketplace. Indeed, as Slater was only too quick to point out, Intel is shifting people from the 80860 development project onto the iAPX-86 line, on which Intel is concentrating most of its resources now. DEC’s Alpha, meanwhile, was given a hard time for being late into the market. Three years ago, Slater said, I predicted that there would be no more major RISC announcements, and then came Alpha – I could still be proved right. He is cautious about all the hype that has surrounded Alpha from the time of announcement and warns that too much is being promised for a chip that was designed in the first instance to provide a Unix migration path for VAX users. He said DEC has pulled out all the stops to achieve the fastest possible clock speed – 200MHz in a CMOS silicon implementation – and a pure 64-bit architecture. In striving for this goal, however, DEC has sacrificed cache memory capacity – a third of the chip is actually taken up by clock buffers which dissipate half of Alpha’s notably high level – 30 Watts – of heat emission. Although Alpha is packaged to the same die as the Viking SuperSparc, Alpha manages only 16Kb cache while Viking has 36Kb. Nothing, according to Slater, will make up for the fact that Alpha is three years late to market.
By Sue Norris
He pointed to a couple of other technical inferiorities, and went on to note that DEC still hasn’t managed to convince any of the semiconductor manufacturers to second-source fabrication of Alpha – it looks as if DEC will have to make it itself. According to Slater, the semiconductor suppliers just aren’t confident that a market exists for Alpha. Whether or not Slater was aware of Wednesday’s announcement of Olivetti’s endorsement of the Alpha chip, he commented that, since no major workstation vendor had latched onto Alpha, other than DEC itself, there is no-one to really drive the market. Slater pointed to the Motorola 88110 – not yet formally announced due to some lingering problems – as perhaps the best of the current crop of microprocessors, adding that it is a shame that it won’t make an impact because it doesn’t have the customers. The 88000 Slater described as the most highly integrated of early RISC implementations with a very clean architecture and no major warts, except perhaps the combined integer-floating point register file. Its failure to take the market, in Slater’s view, was due to strategic and not technical reasons. Adding that no-one seems to know what really happened, Slater went on to say that it was an apparent lack of early commitment by Motorola’s management that led to the loss of Sun and others as takers for the 88000. The processor also suffered slow product development and, as a result, potential customers lost patience. This, coupled with internal conflicts at Motorola over the 68000 programme, led to a weak positioning of both product line
s. With something of a domino effect, this lack of success in the Unix market, according to Slater, caused Apple to switch to the PowerPC. As strengths, Slater pointed to the 88000’s high level of integration, good single-precision performance, multi-processor support and Motorola’s advantage of single-vendor control, while as weaknesses, he listed the expense of building a large cache on the 88100/200 implementations and the slow speed of the floating point in double precision mode. Slater held up Hewlett-Packard’s Precision Architecture RISC as being the basis of the fastest workstations shipping today, as having a rich architecture and a large primary cache in current implementations; on the downside, despite the Japanese and Korean licensees, there are no useful chips on the open market and no other workstation vendors support the part.
No commitment for MIPS
Also, current implementations are costly, the system design for high clock rate primary cache is weak, and there is a lack of strong US partners. For Sparc, the weaknesses listed were not significant, except to say that Sparc is unlikely to support Windows NT. Its strengths meanwhile include its application software base, Sun’s substantial investment, and the availability of second source suppliers. MIPS Computer Systems, on the other hand, has the first RISC that will support Windows NT. The R-series is also supported by a wide range of applications, and the MIPS market is a level playing field, not dominated by a Sun Microsystems. Sun’s very presence in the workstation market, however, means that MIPS will unlikely ever be number one. The disbanding of ACE, of course too, has done little to boost MIPS’ credibility, and there exists no major vendor that’s committed to MIPS as a desktop processor. The RS/6000 and PowerPC lines are credited with the commitment of three large industry players, and a high-performance floating point. On the downside, high performance can today be only achieved with a proprietary multi-chip implementation, early PowerPC implementations will not be performance leaders and IBM is caught up in internal conflicts over the AS/400 and its mainframes. Finally, at the low end, Acorn spin-off, Advanced RISC Machines Ltd, seems set to battle it out with AT&T’s Hobbit RISC in the handheld computer market – ARM has been taken up for Apple’s Newton range, Hobbit is expected to be the RISC target for Go Corp’s PenPoint operating system.