10th > March > 1999 Archive

Intel roadmap reveals plans up to Y2K

A year agoFrom The Register No. 70, March 1998 More details have emerged about Intel’s future roadmap, taking the plans up to the last quarter of 1999. According to sources very close to Intel’s plans, Intel will introduce its 450NX chipset for servers, using Slot 2 architecture in June. Details include support for four processors, 8GB of memory but without AGP (advanced graphics processor) support. It will support four way interleaved memory, HotPlug PCI and up to 2MB of level two cache. In the same time period, Intel will release prices for its Slot 2 architecture. This will be $4,500 for a processor with 2MB of Level 2 cache, and $2,850 for 1MB cache, with a low end, 512K version selling for around the $1,400 mark. Again in June, Intel will announce its so-called Nightshade project. This is a reference design for a Slot 2 server motherboard with twin Deschutes support, using the 440BX chip set. August this year will be noticeable for the departure of the ceramic Pentium MMX processor, according to the source, barely a year and a half after it was first launched. Things then go quiet until Q3 this year, when Intel will reveal its Marlinspike project. This is an AT based Slot 2 reference design for motherboards, using the 440GX chipset which will support up to 2GB of memory in four DIMM configurations. Intel will release a PII/450MHz processor which supports the 100MHz bus for both Slot 1 and Slot 2 configurations. That is expected to be priced at around the $750/1000 mark. In Q4, as previously reported here, Intel will introduce the second of its Celeron processors, codenamed Mendocino, which will run at 300MHz and have 128K of Level 2 cache integrated on the unit. Q4 will be a busy period for Intel. It will introduce the second rev of its i740 "Auburn" chip, codenamed Portola. Any PC98 compliant systems will lose ISA slots. Again, in the same quarter, Intel is expected to announce an 8-way motherboard using the 450NX chipset and codenamed Saber. This uses Corollary’s Profusion chip. Intel took over Corollary. At the back end of this year, it will introduce a 66MHz 64-bit PCI bus, using an upgraded 450NX chipset. Intel is also expected to announce a portable Pentium MMX processor with a revised form factor and running at less than four watts. This mobile innovation will be the precursor to a project called Geyserville, slated for the first half of 1999 and intended to make even thinner notebooks possible. The first half of 1999 will also make Device Bay a reality. It is expected to come in three configurations aimed at different markets like desktops and notebooks and will eventually supersede the ISA and IDE standards. It will also support FireWire. Katmai, the processor with additional MMX instructions, is now projected for the second half of 1999. It will start at clock speeds of 450MHz, rising to 533MHz, initially appearing in a .25 micron flavour but later moving to .18 microns. Initial pricing suggests a $700/1000 level. Mintel hopes to speed this processor to around 750MHz by the millenium. Intel’s Camino project will be based on the 440BX chipset, and adds a south bridge PIIX6 which will also support 66MHz Ultra DMAS IDE. It will accept dual Katmai chip cartridges, come with AGP 4x support and use Direct RamBUS memory running at around 800MHz. That, again, is slated for the second half of next year. The chip giant will then announce its Colfax project, a 440BX notebook motherboard with AGP, FireWire and RDRAM. The last stage in Intel’s ambitious PII plans, the source said, is the Williamette project for the end of 1999. This will use a .18 micron process which will then be shrunk to .13 microns and anticipated speed advantage of up to 50 per cent. Clock speeds will go from 800MHz to 1.2GHz, and it is expected to include new bus designs. The source said this could be the last of the IA-32 architectures as Intel phases in IA-64 support. It will bridge the last years of this century and the first few years of the next decade. Intel always refuses to comment on unannounced products. ®
Mike Magee, 10 Mar 1999

HP takes risc on PA plans

Hewlett-Packard is attempting to respond to threats from the Alpha microprocessor by bringing forward the release of its PA-8600 chip. The company issued a statement yesterday intended to tantalise and tease its end-user base. It hopes the processor will speed up data intensive apps like data warehousing, CAD and e-commerce. Curiously, Compaq's Wildfire, slated for the end of this year, also aims for these markets. The company said it will include high availability features, and come in 500MHz and 550MHz by 2000 Q1. It will have faster cache algorithms and cache prefetch technology. It will also include error checking and correction (ECC) on its 1.5MB of cache on die, and include lockstep capability. While HP's new guard hopes that the PA Risc technology will pass into history, its old guard thinks differently, we are informed. ®
Mike Magee, 10 Mar 1999

Compaq workstation prog in chaos

Reports reaching The Register from insiders at Compaq are suggesting that its workstation strategy is in chaos. According to the insiders, the Compaq workstation (x86 and Alpha) is "officially toast" today. This is nothing to do with the Alpha versus x86 argument currently raging. Compaq will make an official and very subdued announcement towards the end of this month. The insider tells us that there will be commercial desktops (read professional) and servers in a Brave New Windows World. It's too early in the morning here for us to contact Compaq for the official no-comment we expect. But we'll try later in the day. ® RegisTroid 911 'Brown Bread' is Cockney rhyming slang for dead, sort of a bit like toast.
Mike Magee, 10 Mar 1999

The issues Intel sidesteps with its FTC deal

The legal background to the anticipated Intel-FTC Consent Agreement gives some insight into Intel's strategy, while the minutiae provide just a few glimpses of what we all missed as a result of Intel's capitulation. Intel was formally told that it was being investigated by the FTC in September 1997. This was not the first time that the FTC had turned its attention to the chip giant: in a 1993 investigation, the FTC concluded after two years that no action was warranted at that time. The FTC issued its Complaint against Intel on 8 June 1998, less than a month after the Department of Justice began its present case against Microsoft. There really is no connection, and it would be wrong to construe the actions as an anti-Wintel move. Unlike the first Microsoft case, where the FTC commissioners could not agree about how to proceed (resulting in the DoJ taking over the case), this time the commissioners under Chairman Robert Pitofsky voted 3-1 in favour of legal action, with Commissioner Orson Swindle dissenting because he thought more information should have been gathered before the FTC made a "reason-to-believe" determination. The Complaint states that in the FTC's view, Intel has monopoly power, and "wilfully maintained its monopoly power in the general-purpose microprocessor market" through exclusionary conduct. The legal basis of the Complaint was that Intel had violated Section 5 of the Federal Trade Commission Act by "unlawful monopolization, unlawful attempts to monopolize, and unfair methods of competition". Specifically, the Complaint refers to Intel's refusal to do business with Digital, Intergraph, and Compaq. Intel would have to prove that its actions resulted from "legitimate business considerations" (such as bad credit or misusing the technology -- not because Intel preferred no competitors) to win the case, which was presided over by Chief Administrative Law Judge James Timony. The hearing was scheduled for 12 January 1999, then later delayed to 23 February because of the complexity of the case, and then finally to 9 March because Intel had failed to produce documents to the FTC on time. In 1996, Intel wanted to obtain a royalty-free licence to Intergraph's Clipper processors, which Intergraph refused, so Intel withdrew essential technical information and prototypes from Intergraph. Intergraph won a preliminary injunction against Intel in April 1998 in the District Court in Birmingham, Alabama, but that trial is not scheduled until 14 February 2000, when the judge has set aside five weeks for it. Digital sued Intel in May 1997 for patent infringement, alleging that the Pentium infringed ten Digital patents. Intel cut off the supply of technical information to Digital, and threatened to stop supplying processors. Only when the lawsuit was settled did Intel recommence the flow of information. Compaq sued Packard Bell for using Compaq-patented technology in its motherboards, and since these were manufactured by Intel, Compaq did not receive technical information from Intel until Compaq agreed to cross-license its patents with Intel. Intel denied it had monopolised any market, claimed that the disputes arose from intellectual property issues, and managed to obtain a substantial protective order to keep its information confidential. Intel's counsel initially failed to obtain the court's permission to see confidential information from the other parties, and was denied permission to appeal. Later, the judge agreed that Joseph Kattan, who advised Intel on business matters as well as the present case, could see "a reasonable number" of Digital and Compaq supersensitive documents. Transmeta Corporation [employer of Linus Torvalds] entered the frame when it unsuccessfully tried to quash a subpoena to produce documents. The suspicion grew that Intel was using the case as a fishing expedition to find out as much as possible about rivals and potential rivals. Dell managed to keep in camera some information demanded by Intel. Texas Instruments and STMicroelectronics initially refused to provide subpoenaed documents until threatened with enforced compulsion. There were many objections to the public status of documents, and the appropriateness of witnesses. In January and early February this year, 13 Motions and Oppositions were filed under seal concerning witnesses and document production. Intel tried to get Richard Parker, the FTC's Counsel, removed from the case because when he was in private practice, he had represented AMD in litigation against Intel. Since Intel had known of his participation for four months, this was seen as an eve-of-trial tactic. Parker's participation had been cleared by the FTC's ethics officer and the Office of Government Ethics. Curiously, Intel claimed that an American Bar Association model rule forbade his participation, but the DC bar, of which Parker is a member, had not adopted the rule (which is concerned with lawyers moving from private to government practice). Intel proved itself to be tough when it subjected Wade Patterson to three days of deposition, and still wanted more information from him, which the judge would not allow. In an expert report, FTC witness Frederic Scherer said that the effect of Intel producing around three quarters of new-generation chip sets was to render a computer assembler a "stylist" and "undifferentiated box assembler". As the trial date approached, Intel tried hard to have evidence excluded. A week before the trial date, Intel moved to strike the designation of Dean Klein, Donald Lewine, Lee Hoevel, Gordon Campbell, Vinod Dham and David Hixson as witnesses for the FTC. Intel also had a go at precluding any hearsay testimony by FTC witnesses Atiq Raza and Robert Herb. In the end, the FTC succeeded in getting a witness list that would have helped it to prove that Intel's "broad pattern of behaviour" was illegal. It seems that Intel approached Michael Sohn of the DC law firm Arnold & Porter (where Pitovsky and Baer had both previously worked) and he brokered the proposed settlement agreement with the FTC. The FTC was keen to find a solution that would achieve its objectives but not curtail Intel's ability to compete. There is no room for negotiation, because Barrett signed for Intel Sunday evening, in Washington, following telephone negotiations. On Monday 8 March, there was a joint Motion to withdraw the matter from adjudication "for the purpose of considering an executed proposed consent agreement". The FTC secretary announced that the request was granted. The consent agreement will now be considered by the FTC commissioners, and it is extremely unlikely that they will dissent. Generally the FTC likes consent agreements - and the commissioners are unlikely to want to get their hands dirty with a very technical and contentious case. The coincidence that the Intel cave-in followed so closely after Microsoft had concluded giving evidence in its trial is just that. The legal procedures are entirely separate and unrelated. Nor is there the same groundswell of feeling against Intel as has happened with Microsoft. However, Intel would have noted how the once mighty Microsoft had been pilloried in court, and would not have wished to bare its soul before the FTC. As The Register exclusively reported on 22 February, a settlement was in the wind, although the final capitulation decision was not made until the last moment as the FTC was still taking witness testimony on Friday, and Intel's legal team at Gibson, Dunn & Crutcher was polishing the defence case. Although the text of the proposed agreement was not disclosed on Monday, it would be surprising if the terms differed substantially from the FTC Notice of Contemplated Relief. Some cosmetic changes from the FTC's Notice may have been allowed to help Intel to save face, but the result should not be seen as anything but a win for the FTC. The main terms are therefore likely to be a "cease and desist" agreement that Intel will not in future discriminate in its processor sales, its NDAs, the supply of relevant information, the supply of prototypes, and technical assistance in areas where Intel is dominant. Intel's CEO will probably be required to send all Intel's customers and employees a copy of the Complaint and the Consent Agreement, stating that Intel will abide by the terms of the Agreement. The Intel response was rather predictable: CEO Craig Barrett claimed a "win-win" and that it "gives us value for our intellectual property rights" which is a very woolly remark. Intel spokesman Chuck Malloy's line was that Intel was pleased because it had avoided "a long and expensive trial". He too was woolly on the subject of the outcome, saying that Intel was pleased that the "issues associated with our intellectual property have been addressed" which seems to be a pretty clear admission of defeat, in spokesman's lingo. Victoria Streitfield, the FTC spokeswoman, said the FTC "got the relief it wanted". William Baer, Director of the FTC Bureau of Competition, said that there were "remaining issues under investigation by the Commission. The Commission's staff is committed to working expeditiously to resolve those concerns." The Complaint did extend beyond issues concerned with the three third parties, but details were not fleshed out. This is the first time that the FTC has publicly acknowledged it was pursuing other issues against Intel: they probably include the consequences of Intel expanding its processor and so displacing other chips [a rerun of the incorporation of utilities into the operating system, this]; concern about the Intel Inside campaign, with its up-to six per cent rebates for OEMs that played the game; motherboards; and chipsets. Reading between the lines, it seems that what Intel has succeeded in doing is splitting the case into two: giving in to the part that in effect it had admitted but claimed was not unlawful - its actions towards Digital, Intergraph and Compaq. There remains the more serious accusation of monopolisation -- using its market position to achieve dominance in other areas -- which could deter acceptance of the agreement by the commissioners when they consider it during the next few weeks. Settlements do not create precedents, since they cannot be cited. Insiders have suggested that the settlement avoids dealing with the controversial issues, such as the proof of Intel's monopoly, and Intel's culpability. It may well turn out that Intel will try to conduct business as usual, but give in more easily if challenged lest it finds itself back in court. Intel is full of engineers, so a cost-benefit analysis in the light of present realities could well have decided the issue. There is also the issue as to what extent an extended case would defocus Intel as it faces strengthening competition, and create adverse PR. Furthermore, Intel is spending $300 million on its current ad campaign for the Pentium III, so any distraction could be costly. The outcome for Intel is certainty rather than uncertainty, and perhaps a two-year breathing space before it might be in court again on additional charges. The case the FTC had to prove was not a certain one, since it was clear that Intel's monopoly was weakening. What remains unresolved is the interaction between antitrust law and intellectual property law. ®
Graham Lea, 10 Mar 1999

IBM gives Linux drivers helping hand

Big Blue said today that it is designing a set of drivers to speed Linux along on its Netfinity range of servers. The announcement puts flesh on the bones of the deal it recently struck with Red Hat to bundle the latter's version of Linux with certain Netfinitys. At the time, IBM said the agreement would involve involve work to ensure OS and hardware were fully compatible. Tikiri Wanduragala, IBM EMEA Netfinity server consultant, said: "Customers want us to bring Linux to the same level as other operating systems. A lot of the work is upgrading other software that Linux requires. "We're working on specific areas in the kernel that have to be addressed for Linux. RAID systems for servers is one. Comms adaptors need very specialised operations. You've got microprocessors behind every adaptor. That's where the work has got to go in. "Linux currently is a two CPU model. Initially, we will support Linux on our 3000, 5000 and 5500 Netfinitys and not on the high end 7000. The reason for that is Linux is only two way. We're working with the Linux community to fix that. "We have to bring in clustering too. The work for 16-way systems is in place, it's not there yet. Application is software is not yet out there to support the high end. Customers are happy to get their two way systems up and running. "Systems management is an absolute must. Our hardware generates alerts so we have to work on Linux-based code for that. Service and support has to be patched in so we can offer exceptional support. We're going to offer 90 days support for Linux as well." Wanduragala said: "Linux is a different world. It's a question of identifying a problem and putting the problem in front of people to lots of people can fix it." He said that the drivers IBM is working on will be made open source code, meaning that other vendors can adapt it to their needs. "We're looking into doing co-marketing with Linux. But we're OS transparent. The drivers are the only things that tune a machine. Adverts will be OS generic." A team in North Carolina will also do chip optimisation for Linux, he said. ®
Mike Magee, 10 Mar 1999

Big Blue schizzy over AIX and Linux

A senior executive at IBM said this morning that Linux may challenge AIX at the high end. Jon Prial, who runs IBM's software EMEA group, said: "No one can predict where Linux is going to go. Linux is looking to move to the high end. "AIX continues to be a secure powerful system and we're very supportive of it. At this point we're happily supporting many operating systems." Prial said: "All the boats are rising and we haven't seen any of the operating systems declining [in market share]." But Prial was reluctant to say what IBM would do if Linux ever competed with AIX. He did say: "We're focused on Netfinity on Linux. Right now our focus is on the server side. The systems are more ready on the server side. We're not here to bully anyone around. We're going to confirm our development on NT." ®
Mike Magee, 10 Mar 1999

Free ISPs safe from UK watchdog

The future of subscription-free Internet access in the UK is safe -- for the time being at least. After weeks of intense speculation that any decision by OFTEL could signal the end of the 'FreeServe phenomenon', the telecoms watchdog has decided to sit on its hands and do nothing. OFTEL is proposing that there should be no immediate change to current arrangements that allow ISPs to offer free access in return for a slice for the interconnect charge. OFTEL is also suggesting that companies should be able to offer customers different call charges to reflect the type of service on offer. This, OFTEL predicts, would help to stimulate competition and could lead to reduced charges for customers who want basic Internet access. "I believe our proposals strike the right balance between supporting the growth of Internet users and meeting the needs of the telephone operators," said David Edmonds, OFTEL's director general. "Internet service providers [should be able] to generate the funds necessary to allow additional investment in their respective networks to cope with growth. "ISPs and their supporting telephone networks should be able to vary the price they charge for a call to reflect the level of service they provide to their customers." ®
Tim Richardson, 10 Mar 1999

US halts “dumping” of Japanese supercomputers

Asian reports said that the US International Trade Commission (ITC) has ruled that Japanese sales of supercomputers into its home market will damage American-built boxes Last month, NEC asked the US Supreme Court to stop the Commerce Department from dumping SX-4 machines into the home market. Now, if Japanese companies attempt to sell their super-tin to US government agencies, such sales will be prima facie, construed as dumping. That will mean punitive anti-dumping duties will be levied. ®
A staffer, 10 Mar 1999

UK start-up to take surplus stock out of channel

XS Technology, a new UK company re-selling surplus channel stock, will start next month. XS said it will take unsold and unwanted kit from vendors, distributors and resellers and sell it in immature markets. Thereby keeping everyone's channels clean, it claims. Based in Milton Keynes, XS will start with five staff, and plans to take just about any PC or component. Stock will be sourced from Europe, and sent out to Africa, Asia or America. Its three directors will be Martyn Parsons, ex-IBM at Greenock, Nigel Lippard, currently MD of manufacturer and distributor Pine Technology, and Keith Williams, also from Pine, where he was head of European sales. Rival surplus stock dealer PST was unfazed. Peter King, PST sales director, said: "We've been in business for 25 years, and have a very sound relationship with our suppliers. Manufacturers must be confident about what's going to happen with their surplus stock and I think they will be wary about a company setting up in this sector." King said there were many small-time brokers in the UK, which were not seen by PST as competitors. Williams stated: "XS will not be a broker, but a trader. We will re-work, or re-sell, new or second-hand surplus machines." Lippard added: "There is definitely room in this market for another company. As we see increased financial pressure on inventory, vendors will be under increasing pressure to hold inventory rather than put it through the distribution channel and companies like XS will give them way out for discreet disposal." ®
Linda Harrison, 10 Mar 1999

S3 takes time to check The Register out

After we wrote a story about S3 being up for sale a couple of weeks ago, the guys came into the office to give us the SP on its future developments. Paul Ayscough, senior manager of S3's European division, said that we'd see its next generation part, codenamed GX4, in four months' time, with unspecified "other things", which he could not disclose, which will move its part into the high end. He said: "Eighty million [desktop] PCs will be sold this year. The top, one million, end of the market is not where we're at." He said the large slice below that was, in S3's sights. "That large bulk area is where Savage4 is aimed at," he said. "We'll have three parts for that, the 143 Pro, the 125GT, and the 110LT." The 143 starts at $25 with the others stepped below that. "Trio3D is low range at the moment, at around $10. Next year, that will be a combined element in the North Bridge," he said. The Trio3D is S3's best version of its old Virge technology but with about four times the performance. He said that Savage4 will cost S3 much less to make and the yields "will be very good". Its design wins include Diamond, Creative, ELSA, Hercules, Number Nine and Creative, he said. The company has already won four out of the top ten OEMs and he expected to sign another two in the next few weeks. S3 already has samples of 0.22 and 0.18 micron chips, said Ayscough. "A mobile part based on Savage4 will be out in a few months' time," he said. He said: "We expect to ship huge quantities of this chip and get a very significant share of the marketplace." ®
Mike Magee, 10 Mar 1999

Seagate counts cost of closing Scottish factory

Seagate is taking a restructuring charge of around $60 million in its current third quarter, including the scrapping of its Scottish factory. The vendor said the one-off $50 million to $60 million charge would pay for the closure of the Livingston plant, consolidation of its customers service facilities and other action it was unwilling to detail. The future of the semiconductor plant had been uncertain since last year. In October, Seagate announced it was looking for a buyer, then later that month decided to sell the plant. At the time the factory employed around 262 people. This number is now down to about 50, production has stopped, and the closure will go ahead in July. The restructure will also include the closure of the two tape repair centres in Irvine, Scotland and in California. This facility will be consolidated in Mexico. The distribution centre, also in Irvine, will remain open according to Ian O'Leary, Seagate corporate communications manager for European operations. O'Leary added that the company would continue in its main business of storage: "Seagate is not a micro-electronics company. This area employed around 260 people, out of 83,000 in Seagate as a whole." He admitted: "The disk drive sector is a very competitive market and the price of storage is continually going down." O'Leary said Seagate would tackle this by concentrating on reducing costs and improving time to market. The US giant denied it would be hit by the IBM/Dell deal announced last week. Seagate previously supplied direct vendor Dell with disk drives, but Dell has announced it buy such items from IBM. O'Leary said: "We retain a very good relationship with Dell. Every indication is that Dell will continue a balanced, multiple-source strategy for their components." ®
Linda Harrison, 10 Mar 1999

Corel signs Cygnus to simplify Windows-to-Linux ports

Corel has signed Cygnus Solutions to modify the compiler developer's Linux programming tools to make it easier for software companies to port their Windows applications to the open source OS. The deal seems to have been done to make it easier for Corel to deliver on its commitment to port its productivity software to Linux. Having translated the WordPerfect 8 word processor to Linux, Corel said it would port over the other applications in the WordPerfect Office suite too. However, since Cygnus will make changes to its GNUPro professional Linux development system, other companies should benefit from Corel's largesse. The move follows Intel's deal with Cygnus, in which the Great Stan of Chips paid the developer to upgrade GNUPro to cater for its more recent chip developments, including the Celeron architecture and the Pentium III's Streaming SIMD Extensions. It also suggests Corel is moving away from its support for Wine, the Windows-on-Linux emulator. Corel executives have been talking up Wine for some time as the main way of getting Windows apps over to Linux. However, the company's executive VP of engineering, Derek Burney, recently admitted that Wine was only successful with a handful of applications, "mostly games". The process of extending Wine, currently handled by members of the open source community (ie. people with a slightly less commercial interest than Corel), may therefore be too slow for Corel's liking, hence the desire to improve the compatibility between GNUPro and Windows development tools. ®
Tony Smith, 10 Mar 1999

Linux community demurs over Big Blue statements

We have had several letters from Linux readers which begged to differ with IBM over SMP plans for the OS. But, on the other hand, nearly all were thankful IBM was doing the necessary driver porting work. Earlier today, we posted a story saying that IBM had undertaken a massive drivers project for the Linux OS on Netfinity servers. (Story: IBM gives Linux helping hand) One Linux fan said: "IBM's statement about linux being only 'two way' are incorrect. According to the Linux SMP FAQ, Alan Cox states that even Linux 2.0 has been tested with four CPU systems. Linux 2.2 supports up to 16 processors (the same as 2.0), but due to kernel changes, can handle more than two more efficiently than 2.0". However, the fan added: "It Does seem like IBM will do some nice stuff for Linux. Look out billyboi!" Another reader wrote: "The Linux that IBM is supporting on its Netfinity servers is Red Hat 5.2, which is indeed only two-way since it is based on the older 2.0.3x Linuxk kernel. Red Hat does not yet support the newer 2.2 kernel released mid-last month which has significantly improved SMP capability. "The newer Linux 2.2 kernel does support up to 16-way, but keep in mind, this only means that it boots on a 16-way. Actual scalability has yet to be demonstrated. That means there is little or no evidence that performance actually increases when you add more processors. There has been no public demonstration of scalability (e.g. published benchmarks from trade press or independent benchmark organisations) of improved performance to even four-way as far as I am aware of, as I've been keeping an eye out for that." ®
Mike Magee, 10 Mar 1999

Silicon Graphics looks to MIPS to cover Merced delays

Silicon Graphics (SGI) yesterday went on record to commit itself to the MIPS Risc processor for at least two more generations of the current core and possibly a third incarnation beyond those two. Doubts were cast on SGI's support for MIPS last year when the company not only spun the chip developer -- which it bought out in the late 80s in a move designed to counter Sun's aggressive Sparc development plans -- as a separate business unit but stated its decision to migrate to Intel's 64-bit Merced chip for future high-end machines and the IA-32 architecture for low-end workstations. Now SGI will offer machines based on MIPS' R14000 CPU, which is due next year, and later on the R16000, scheduled to ship in 2001. Both processors will be updates of the current R12000 core. The R14000 will support double-speed backside cache memory and a 200MHz external data bus, twice that of the R12000. It will also offer a clock speed increase, taking the design to 450MHz. The R16000 will boost the clock speed to the 600-800MHz band. The current processor's 32K data and instruction L1 caches will be both the doubled to 64K. The chip will support up to 8MB of external L2 cache. The decision to continue with MIPS chips was based on customers demand for continuity over SGI's transition to Merced, claimed John Mashey, the company's chief scientist, in an interview with the Reuters news agency. SGI has said it will port its Irix Unix-derived operating system to Merced. However, the plan to stick with MIPS, at least for the immediate future, will guard against delays not only in the shipment of Merced CPUs but in SGI's effort to move the OS from one processor platform to another. The fact the R14000 and R16000 are little more than speed bumps of the R12000 suggests that SGI is doing little more than cover its rear end while MIPS pursues its designs on the embedded market. The possible fourth-generation R12000, presumably to be called the R18000, appears to be simply a more compact version of the R16000, again suggesting it's little more than a stop-gap to fill the hole left by a possibly delayed Merced and/or IA-64 Irix. ®
Tony Smith, 10 Mar 1999

Dixon Intel mobile PIIs have serial numbers – it's a bug, official

UpdatedA Register reader is claiming that Pentium II/mobiles using the Dixon cores embed unique serial numbers and has posted information on his site to back up his claim. And now Intel has confirmed it has a manufacturing problem. Pierre Chassaing's Web site produces evidence for his claim. He used a utility called WCPUID to interrogate a Dell Inspiron 7000 laptop. An Intel representative confirmed there was a problem. He said: "There is an erratum in a limited number of the mobile processors. Some prototype circuitry has gone through. We'll do a BIOS patch in the next day. This was not intentional." He added: "This erratum only affects those mobile Celeron and Pentium II Processor parts with on-die L2 cache that use the mobile module package. These products were introduced on Jan 25th of this year, and have only been sold in systems since then." Intel describes bugs as errata, not bugs. ® Intel moves to re-assure industry on erratanotbugs Intel Developer Forum Feb 99
Mike Magee, 10 Mar 1999