AMD Tackles Intel, Intel Tackles P2P.

Intel Corp.'s recent decision to recall its 1.13GHz Pentium III processor came as little surprise to many. For a number of industry observers, the recall announcement was less an issue of whether and more one of when. For the first time in its storied history, the chip giant is playing catch-up to a competitor. The results have not been pretty.

First, a quick review of recent events. Intel announced the 1.13GHz PIII at the end of July in an effort to beat the announcement of AMD's 1.1GHz Athlon. The chip was recalled less than a month later. "It is very, very rare for Intel to miss something like this in microprocessor testing, especially on a mature design," says Gartner Group microprocessor analyst Martin Reynolds. "It is certainly an embarrassment for them."

Meanwhile, a scarcity of systems from major QEMs indicates that Intel has had problems just getting its 1GHz chip--announced ten months ago--into the hands of consumers. "Both manufacturers stretched a bit to meet the 1GHz target in the first place, and we can expect to see this clock speed leapfrogging continue," Reynolds says. "Intel has the P4 with its 1.5GHz clock speeds right around the corner," he adds, "but they will have to fix and advance the PIII chip, as it will continue to be their main revenue source through most of 2001." At press time, the P4 was expected to launch in October (Fig 1).

The company also said that it was pushing back the release of its 64-bit (Itanium) offering by six months at least. Intel was also forced to recall its 820 chip set earlier this year as well as defective mother-boards in May, and is in the process of bringing litigation against Broadcom for alleged patent infringement. Broadcom has countersued and is seeking injunctions barring Intel from selling a number of products.

Further, because of royalty payments for Rambus RDRAM--which is faster than SDRAM--and that memory technology's price premium, Intel is having trouble convincing the industry to adopt RDRAM for mainstream systems. In fact, recent tests have shown that by using DDR SDRAM, system vendors can match the current speeds of RDRAM, and for less money. Some of the performance limitations of RDRAM are due to the PIII's 133MHz bus, a problem which will presumably be eliminated with the P4's new 400MHz bus. Combine all of these problems with projections of slumping PC sales and a falling stock price and you get a giant with an image problem.

But you must contrast this picture with one of Advanced Micro Devices to get the full story. Historically the runner-up to Intel when it comes to processor speed, this year AMD surpassed Intel for the first time in core megahertz. While overall performance benchmarks indicate that the PIII is still faster than the Athlon at equivalent speeds (due primarily to the Pentium's larger on-die cache running at the processor speed), AMD is set to introduce a new chip (code-named Mustang) which will reduce or eliminate this disparity as well as add a faster front-side bus. Further, AMD has matched Intel's SpeedStep with PowerNow!; has matched or outstripped Celeron with Duron; has matched Intel's process technology (.18 micron fabrication); and had grabbed several top-tier OEMs from the Intel tent, perhaps the most telling development. And the company's stock recently split.

Future Shock

So what does the next year hold for the two companies? While Intel has been trying to move into the datacenter for five years, clearly there is some added pressure now: The company can no longer depend on its total control of the PC market to boost revenues. To this end, Intel is moving in several directions at once.

First, the company is expecting to ship its 64-bit Itanium processor sometime in 2001--predicting exactly when seems an effort in futility these days. The Itanium will move Intel (and presumably the rest of the PC industry, where Intel chips still rule) to a completely new instruction set architecture. Moving from the decade-old 32-bit, x86 instruction set to 64 bits will require an enormous investment in time and developer manpower; Intel is clearly betting that both software developers and hardware OEMs will make the commitment. AMD, by contrast, plans to ramp up the x86 instruction set with a new chip. This chip, code-named Sledgehammer (Fig 2), will prolong: the life of x86 code by adding 64-bit extensions to the existing architecture, easing the transition to a fully 64-bit world.

It's too early to tell which company's strategy will succeed. Much depends on how wedded developers and consumers are to the current base of applications. But we can expect huge marketing campaigns from both sides to begin in earnest next year, when Intel and AMD will each seek to persuade the world that their path is the road to computing Nirvana.

Intel is also now quietly supporting Linux development. In conjunction with IBM, HP, and NEC, Intel has formed the Open Source Development Lab, a non-profit lab for developers who are adding enterprise capabilities to Linux. In announcing the initiative, the four companies said they plan to provide significant (though undisclosed) equipment and funding to the lab over the next several years. Additional contributors and sponsors of the lab include Caldera, Dell, Linuxcare, LynuxWorks, Red Hat, SGI, SuSE, TurboLinux and VA Linux. The lab will be outside Portland, Oregon and is scheduled to open by the end of the year. (Unix colossus Sun Microsystems was noticeably absent from the announcement.)

Peering Into The Future

Interestingly, Intel is bullish on the much-maligned peer-to-peer (P2P) networking movement. On first pass, this support seems odd: Why would a chip maker choose to become embroiled in a technology that seems--at least thus far--to exist solely to provide users with an easy way to trade copy-righted material? But upon closer examination, Intel's position makes sense. P2P technology is both an easy way to expand Intel's goal of PCs everywhere and a great way to turn every PC into a server--and servers have heavier processing loads than PCs, of course. (By incredible coincidence, P2P technology is also a good fit with Microsoft's new.Net strategy.)

To this end, Intel has formed the Peer-to-Peer Working Group, an initiative to "create a peer-to-peer computing environment" according to Pat Gelsinger, vice president and CTO of the Intel Architecture Group. At Intel's Fall Developer Forum, Gelsinger said that "we need to initiate, lead, and as an industry come together to put these key components in place, the common protocols, driving the ease of use, addressing very valid issues in the area of security, making sure peer-to-peer environments are robust and scalable." The P2PWG includes about 20 (mostly small) companies, including Apple Soup, Applied MetaComputing, CenterSpan, Distributed Science, Dotcast, Enfish, Engenia, Entropia, Groove Networks, Life Cycle, Mangosoft, Popular Power, Static, United Devices, and Uprizer. IBM and HP have also agreed to participate.

Intel believes in the power of connected PCs; of this there is no question. As Intel sees it, P2P technology is a way to increase the gigaflops of a company's network exponentially, simply by harnessing the power of all the processors distributed throughout that company. (For a complete discussion of distributed processing technology see the April 2000 issue of CTR, available online at wwpi.com.) Of course, many IT pros will scoff at the notion of putting critical processing demands on user PCs in addition to servers, but Intel is undaunted. Gelsinger says that peer-to-peer technology allows corporations to tap into existing teraflops of performance and terabytes of storage to make today's applications more efficient and enable entirely new applications in the future.

P2P is yet another stage in the seemingly endless cycle of distribute-centralize-redistribute, the paradigm that has ruled computing's evolution for 25 years. Intel officials compare P2P (perhaps a bit hyperbolically) to the arrival of Mosaic, which was of course the basis for Netscape Navigator and the development of the commercial Internet as we know it. According to Intel, in the year before Mosaic, there were 50 Web servers. In the year after it, there were 10,000. Mosaic was simply the spark that allowed 30 years of academic and military work to be accessed by the general public. Similarly, Intel sees Napster, Gnutella, and FreeNet as today's Mosaics, the early front-ends that will allow the technology to move into the common sphere, and then the commercial corporate market.

Making the case for P2P in the consumer space is easy, since there's virtually no management overhead required. In fact, Qualcomm's latest release of the widely used email application Eudora, version 5.0, includes a proprietary P2P networking protocol--the Eudora Sharing Protocol, or ESP--that allows users to share and swap local files easily. Indeed, in announcing the protocol, the company says that the ESP feature saves time and simplifies communications by automatically sharing and synchronizing files without the need for a central server or IT support.

But can P2P technology succeed in the corporate space, where such simplicity isn't an option? The presumable management overhead, both technically and in real personnel terms, for such a computing network is immense: when every machine is a server and a backup drive, the complexity of the network increases exponentially. While companies like Intel and Platform Computing (discussed in April) are trying to add standard protocols and transparent, dynamic management rules to the infrastructure, the infrastructure itself may cause more problems than it solves. Indeed, with the prices of powerful chips and per-megabyte storage costs falling rapidly and continually, the business case for such networks may also be dubious. Nevertheless, expect new business-oriented software sometime during Q1 of next year.

Strike Up The InfiniBand

Intel may also have other motives for its peering cheerleading. P2P technology is a natural fit with any topology that increases the throughput and interconnectivity of servers. InfiniBand is just such a technology and, also at its Developer Forum, Intel unveiled the company's first InfiniBand product plans, which include chips for three new core products due next year.

The new products are planned for use both in future IA-32 and upcoming Itanium server platforms, and they are also expected within Intel's full range of server chipset products. The suite includes an InfiniBand Host Channel Adapter, which connects servers to the InfiniBand fabric within the Internet and business datacenter; an Intel InfiniBand Switch, which will connect servers to remote storage and networking and the InfiniBand Target Channel Adapter, which will connect storage and/or networking devices within the InfiniBand fabric. Look for additional InfiniBand product announcements from other industry players after the InfiniBand Developers Conference in late October.

Fig 1
System Price
w/o monitor
                               Q2'00
Performance/         Pentium [R] III processor
Professional                   1 GHz
[greater than]$2.0K             933
Mainstream 3                Pentium III
$1.5-2.0K                    Processor
                                800
Mainstream 3                Pentium III
$1.2-1.5K                    Processor
                                750
Mainstream 1                Pentium III
$1.0-1.2K                    Processor
                                700
                      Intel 820 (100/133 FSB)
                     Intel 81DE (100/133 FSB)
Value 3                Intel[R] Celeron[TM]
$850-999                     Processor
                                700
Value 2                       Celeron
$700-850                     Processor
                                566
Value 1                       Celeron
[less than]$700              Processor
                                533
                           Intel[R]810E
                            Intel[R]810
Source: Intel
                                   2H'00
Performance/                    Pentium III
Professional                     Processor
[greater than]$2.0K               1.13 GHz
Mainstream 3               Pentium III Processor
$1.5-2.0K              [greater than or equal to]900
                       [greater than or equal to]866
Mainstream 3                    Pentium III
$1.2-1.5K                        Processor
                       [greater than or equal to]800
Mainstream 1                    Pentium III
$1.0-1.2K                        Processor
                       [greater than or equal to]733
                                   Tehama
                     Intel[R]820/Intel[R]820E (100/133)
                     Intel[R]815/Intel[R]615E (100/133)
Value 3                           Celeron
$850-999                         Processor
                             [greater than]700
Value 2                           Celeron
$700-850                         Processor
                             [greater than]600
Value 1                           Celeron
[less than]$700                  Processor
                       [greater than or equal to]566
                          Intel[R]615/Intel[R]815E
                               Intel[R]810 E
Source: Intel
                                      1H'01
Performance/                        Pentium 4
Professional                        Processor
[greater than]$2.0K           [greater than]1.40 GHz
Mainstream 3                       Pentium III
$1.5-2.0K                           Processor
                        [greater than or equal to]1.13Ghz
Mainstream 3                       Pentium III
$1.2-1.5K                           Processor
                          [greater than or equal to]900
Mainstream 1                       Pentium III
$1.0-1.2K                           Processor
                          [greater than or equal to]866
Value 3                              Celeron
$850-999                            Processor
                          [greater than or equal to]800
Value 2              Celeron proc [greater than or equal to]7
$700-850                         Timna processor
Value 1                  Celeron proc. [greater than]600
[less than]$700                    Timna proc.
Source: Intel