Intrinsity Licenses Fast14 to ATI
Peter Glaskowsky - Editor-in-Chief  {02/23/2004}

Intrinsity has licensed its Fast14 technology to ATI, the maker of PC graphics chips and related products. According to the press release, Fast 14 will be used “in future consumer products.” The announcement does not specify the sort of products Fast14 will be used in, but we believe it reasonable to suppose ATI will use the technology in developing the graphics core for Microsoft’s Xbox Next video-game console, now under development.

Fast14 has been proved in Intrinsity’s own FastMIPS and FastMath processors, which are now sampling at clock frequencies up to 2.5GHz. This speed is a remarkable achievement for a chip designed by a fairly small team and manufactured in a commodity 130nm process at TSMC.

The deal is not exclusive; Intrinsity is free to pursue similar relationships with other companies. We will be surprised if Intrinsity can’t sign up other partners. The advantages of Fast14 technology are substantial, and the cost of licensing this technology must surely be lower than the cost of competing against it.

Microprocessor Report readers can access the full story here (1 page): www.mdronline.com/mpr/h/2004/0223/180801.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

IA-32 Plus 64 Equals What?
Peter Glaskowsky - Editor-in-Chief  {02/23/2004}

After two years of speculation, the truth finally emerged last week—Intel has been working on 64-bit extensions for its x86 processors, and they’ll appear in new Xeon server processors by June. The extensions create what Intel calls the IA-32e operating mode, a name that somewhat understates the significance of the change—but IA-64 was already taken. There’s already a perfectly good name for the new instruction set, but Intel will never use it: AMD64.

The surprising truth is that Intel is already within a few months of releasing x86 processors that are software compatible with AMD’s Opteron and Athlon 64. Intel representatives have assured me that all the software Microsoft is now developing will run on both AMD and Intel platforms.

The 64-bit extensions are present in Prescott, the recently introduced 90nm Pentium 4, although they will not be enabled until the next stepping of this die is available. Nocona, a new 90nm die based on the Prescott core, but with an added L3 cache, also has 64-bit support. Both these chips have features not found in Opteron, such as Hyper-Threading and the SSE3 extensions. These features, plus the undeniable might of Intel’s marketing department, will put immediate pressure on AMD, probably forcing Opteron price cuts.

Intel may be playing catch-up with the AMD64 instruction set, but it has far better OEM relationships and platform designs that will need little work to take full advantage of 64-bit computing. I think it reasonable to assume that by the end of this year, Intel’s 64-bit x86 processors will be outselling AMD’s.

It’s an even safer bet that Xeon will outsell Itanium in 2004. This observation begs the 64-bit question: Can Intel afford to let Itanium compete with Xeon?

Today, Itanium has many features that Xeon lacks: more bus bandwidth, better floating-point support, and comprehensive internal error detection and correction circuitry. The last feature gives Itanium a tremendous advantage in reliability, availability, and serviceability (RAS), considered critical for enterprise server systems.

As long as Xeon continues to be developed as a derivative of Intel’s desktop processors, these server-specific features will not be added to it. AMD, lacking an equivalent of Itanium to protect, will have a greater incentive to differentiate its Opteron and Athlon 64 product lines. If AMD adds RAS features to Opteron, Intel’s Xeon group will have no choice but to respond, and Intel’s management will not object. Intel will not sacrifice its high-volume server processors to protect a low-volume line.

This scenario will take years to play itself out; in the meantime, Intel will maintain or even increase its pace of investment in Itanium. Just as Intel has waited until the last possible minute before admitting it has adopted AMD64, the company’s public commitment to Itanium will not waver until the day it decides to abandon the dual-architecture strategy.

That day may never come. Itanium is the fastest, most scalable server-processor family now on the market, the result of billions of development dollars. Xeon can’t possibly be ready to take over Itanium’s role in the server market before 2007. By that time, Intel will probably have sold over a million Itanium processors, making Itanium a solid commercial success in the market for high-end servers.

The ultimate fate of Itanium depends on a long series of decisions by Intel and other companies, including AMD, HP, IBM, and Sun. If Intel can keep Itanium sufficiently ahead of Opteron to prevent server OEMs and buyers from being tempted to view Opteron as an Itanium alternative, there will be no need to allow Xeon to encroach on Itanium’s turf, assuring Itanium’s long-term survival.

To find out more about Microprocessor Report, please visit: www.mdronline.com.

Putting Prescott in Perspective
Peter Glaskowsky - Editor-in-Chief  {02/17/2004}

As part of the launch of its new 90nm Pentium 4 processor, formerly code-named Prescott, Intel has released pricing information and a high-quality die photo of the new part. The new Prescott offerings come in at midrange prices from $163 to $417. These prices exactly match those of Northwood chips at the same speeds. Prescott offers better performance than Northwood on many applications, but this advantage is offset by its greater power consumption. Some OEM systems can’t be upgraded to Prescott because they were not designed for Prescott’s 103W thermal design power (TDP) specification.

Four letter-suffixes are now used to distinguish between Northwood-based and Prescott-based chips—and between the 533MHz and 800MHz front-side bus speeds found on these new products. When Intel publishes a Pentium 4 clock speed without a letter suffix, it means a Northwood die with a 533MHz front-side bus and Hyper-Threading disabled. The A suffix refers to a Prescott die having the same bus and HT configuration. The C and E suffixes refer to Northwood and Prescott die, respectively, with an 800MHz front-side bus and Hyper-Threading enabled.

As the new die photo shows, Prescott’s core occupies almost as much die area as Northwood’s does, despite a more efficient physical layout, reflecting its much higher transistor count. We’ll have more Prescott analysis following the International Solid-State Circuits Conference (ISSCC) and the Intel Developer Forum (IDF). Stay tuned.

Microprocessor Report readers can access the full story here (1+ pages/1 graphic): www.mdronline.com/mpr/h/2004/0217/180701.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Intel Steps Aside
Peter Glaskowsky - Editor-in-Chief  {02/09/2004}

For Intel’s desktop-processor marketing group, it’s almost as if 2003 never happened. By the end of 2002, Intel had begun shipping the 3.06GHz Pentium 4, clearly the best processor on the market for desktop PCs; we gave that part our Analysts’ Choice Award a year ago. Twelve months later—two-thirds of a Moore’s Law interval—the clock speed of the Pentium 4 had increased to only 3.2GHz. Even so, Intel’s Pentium 4 is still the most successful PC processor on the market, and it is one of our nominees for the Analysts’ Choice Award this year.

Intel’s competitors, however, progressed considerably in 2003. AMD rolled out its Athlon 64 family, a derivative of the Opteron server processors that the company also released during the year. A special version of this design, the Athlon 64 FX-51, combined the best features from the PC and server products, earning its Analysts’ Choice nomination.

Surprisingly, perhaps, Apple’s Macintosh platform saw the greatest advance, courtesy of IBM’s PowerPC 970 processor, which Apple calls the G5. The most recent evolution of this design, the 90nm PowerPC 970FX, also receives an Analysts’ Choice nomination.

Intel’s CPU designers must be frustrated that the 90nm Prescott version of the Pentium 4 wasn’t ready sooner, but Intel’s salespeople must be pleased with the company’s bottom line: Intel set a sales record in the fourth quarter of the year, and annual sales grew almost 13% over the preceding year.

AMD, on the other hand, must be frustrated that it continues to lose money despite launching such technically successful products. Apple is slightly profitable, but its new G5 systems can’t take credit for that. The old Motorola G4-based systems, such as the iMac, PowerBook, and iBook, account for almost half of Apple’s total revenue; the G5 is well under the 20% mark.

We expect 2004 to be a very different story, with activity on all fronts. PCI Express will trigger the first redesign of the PC system architecture in years. Intel’s OEM customers will roll out many Prescott-based systems; AMD will also move to 90nm products; and IBM is expected to deliver 3GHz PowerPCs to Apple. VIA will have its own 90nm chips, its first from IBM, and with these, it hopes to expand its share of the PC market. During the coming year, we expect to keep busy covering these activities.

Microprocessor Report readers can access the full story here (4 pages/2 graphics): www.mdronline.com/mpr/h/2004/0209/180601.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

IBM Takes the Lead
Peter Glaskowsky - Editor-in-Chief  {02/09/2004}

The Microprocessor Report Analysts’ Choice Award for Best Desktop Processor of 2003 goes to IBM’s PowerPC 970FX, which shipped to Apple late in the year for use in the Xserve G5 and possibly other still-unannounced systems. IBM has yet to reveal all the details of the 970FX, but we know the chip is made in IBM’s 90nm CMOS 10S process, yielding a chip that is smaller, faster, and more power efficient than the original 130nm PowerPC 970.

We believe the 970FX has a large cost advantage over Intel’s 90nm Prescott. Because it has more than twice as many transistors as Northwood, Prescott is not much smaller—112mm2 instead of 130mm2. The 970FX is a simple shrink, without substantial changes to cache or core transistor counts, so we expect it to be a bit more than half the size of the 118mm2 PowerPC 970.

The 970FX should yield well at 2.5GHz, up from the 2GHz speed of the 970 used in Apple’s Power Mac G5. This 25% increase in clock rate will not soon be matched by Intel or AMD. Prescott is struggling to eke out minor clock-rate improvements, and AMD will need to wait for its own 90nm products—due in 2H04—to achieve substantial speedups for Athlon 64.

Apple has promised to deliver 3GHz G5 systems this summer, implying the arrival of further speedups for the 970FX. That’s an aggressive target, but it should be achievable. The 970FX has the necessary architectural sophistication in its deep pipelines to make this speed possible, and IBM has the necessary technology in its CMOS 10S process. At 3GHz, the 970FX should outrun the chips we expect from AMD and Intel in the same time frame. Even if IBM falls short of this goal, the 970FX has put Apple back among the leaders of the PC speed race and given it a chance to win greater sales in the PC market.

Microprocessor Report readers can access the full story here (2 pages/1 graphic): www.mdronline.com/mpr/h/2004/0209/180602.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Mobile Processors Move Forward
Peter Glaskowsky - Editor-in-Chief  {02/09/2004}

Sales of mobile PCs are growing faster than those of desktops. Large, high-resolution LCDs—up to 17 inches in size, with as many as 1,920 ´ 1,200 pixels—and high-capacity hard disks holding up to 80G of data make today’s laptops reasonable replacements for both corporate and personal desktops.

Selecting a CPU can make or break any mobile-system design. A wide range of size and power budgets exists in the mobile PC market today, from wafer-thin models like Sharp’s Actius MM10 (0.54-inch thick and just 2.1 pounds), built around Transmeta’s Crusoe 5800, to the m:855 gaming laptop from Voodoo PC, which weighs eight pounds but is available with AMD’s Mobile Athlon 64 3400+ processor.

All our nominees for this 2003 Analysts’ Choice Award were substantially new in 2003. AMD’s Mobile Athlon 64 is the fastest processor obtainable for a notebook, but it’s primarily relevant to the desktop-replacement category. Intel’s Pentium M became the mainstream mobile processor of choice almost instantly after its release in March 2003. Transmeta’s Efficeon TM8000 processor earned a spot as our third nominee by virtue of its greatly improved performance and the inclusion of LongRun2 technology.

The coming year will see smaller and larger laptops, and models having both longer and shorter battery life. Performance, like sales, will only go up. In-Stat/MDR estimates that notebook sales will grow about 24% in 2004, to some 44 million units. This growth owes much to the new technology being developed for mobile PC processors.

Microprocessor Report readers can access the full story here (2 pages): www.mdronline.com/mpr/h/2004/0209/180603.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Intel Beats Intel for Mobile Crown
Peter Glaskowsky - Editor-in-Chief  {02/09/2004}

We didn’t give an Analysts’ Choice Award for Mobile PC processors last year, choosing instead to combine desktop and mobile chips into a single category. To our way of thinking, Intel’s Pentium M would have been the only mobile processor worth nominating. This year, Intel faced serious competition once again, with a faster mobile CPU (AMD Mobile Athlon 64) and an ultralow-power device (Transmeta’s Efficeon TM8000) staking claims at opposite ends of the power spectrum.

After due deliberation, we believe the Pentium M—especially the new 90nm version code-named Dothan—is so good that those other parts will never be more than niche players. Accordingly, we give the 2003 Microprocessor Report Analysts’ Choice Award for Best Mobile PC Processor to Intel’s Pentium M processor. Intel has clearly mastered the right technology for mobile PCs, and the Pentium M processor rightfully dominates this market.

Microprocessor Report readers can access the full story here (2 pages/1 graphic): www.mdronline.com/mpr/h/2004/0209/180604.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Servers Take Giant Steps
Peter Glaskowsky - Editor-in-Chief  {02/09/2004}

The largest server-processor vendor in the world was also the quietest in 2003. Intel introduced no new cores, and its only new server processors were shrinks of previous designs with more (or less) cache. A midyear refresh of its Itanium and Xeon MP product families included the “Madison” Itanium 2 with 6M of L3 cache and a 2.8GHz Xeon MP. Not much excitement there, unless you’re in sales—both products immediately appeared in various OEM server systems. By year end, Intel could claim that Intel-architecture servers had surpassed RISC-based machines in both performance and overall revenue. Itanium 2 and Xeon MP are excellent products, and both earn Analysts’ Choice nominations this year.

The big news came from AMD, which rolled out its Opteron family in April and the derivative Athlon 64 series for PCs in September. As if to close an opening for Opteron, Intel in September released two new low-cost Itanium 2 processors with 1.5M of cache, built on the 6M die—but AMD’s foot was already solidly in the door. The Opteron 848 is also among our nominees.

Other companies—HP with the Alpha EV7, Sun with its Throughput Computing initiative, and IBM with its new Power5 processor—also showed more progress in 2003 than Intel did, but none of these introductions threatens Intel’s dominance of the server space. Power5 is especially notable for its sophisticated design and high performance; it is our fourth Analysts’ Choice nominee for 2003.

Although 2003 seemed quiet for Intel, and 2004 may not be much noiser, these two years are just the pre-storm calm. Intel’s Itanium roadmap includes the dual-core Montecito in 2005, built in 90nm technology with 24M of L3 cache, and the multicore Tukwila processor after that.

Intel also plans to bring the cost of Itanium below that of Xeon while extending the performance advantage, converting ugly-duckling Itanium into the company’s most attractive product. Intel has never tried to hide the fact that Itanium is a long-term investment. Although the payoff may come later than the company had hoped, Intel continues to bet on Itanium. Intel’s competitors are making their own bets, but they’ll need more than chips to win.

Editor’s note: Shortly before we went to press, Intel president Paul Otellini gave the first public indication that Intel is prepared to introduce a 64-bit x86 processor, saying “You can be fairly confident that when there is software from an application and operating system standpoint that we'll be there.” In future issues of MPR, we’ll evaluate the effect of this statement on Itanium and Intel’s other products.

Microprocessor Report readers can access the full story here (4 pages/4 graphics): www.mdronline.com/mpr/h/2004/0209/180605.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Opteron Hits the Mark
Peter Glaskowsky - Editor-in-Chief  {02/09/2004}

Building and selling server processors is very different than participating in the PC market. It takes years of consistent, intelligent effort to earn the trust of buyers in IT departments, and, even then, they test the waters carefully. AMD’s server group has been building its reputation for many years, originally working with whatever it could adapt from AMD’s PC processors.

AMD’s Opteron, with its local-memory controller and high-performance HyperTransport interconnects, represents a clean-slate approach to system architecture. Opteron was an expensive, complex development project for AMD, but it has paid off with excellent performance and scalability at competitive prices.

Opteron consistently outscores Intel’s Xeon family on the SPEC CPU measurements, even on the single-processor SPECint2000 tests. On SPECfp_rate2000, a multiprocessor benchmark most relevant to scientific computing but showing the benefit of Opteron’s high-bandwidth, dual-processor Opteron systems outscored four-processor Xeon machines. (See www.spec.org/osg/cpu2000 for more information on these scores, including system-configuration details.)

AMD can’t expect these results to immediately drive large volumes of server sales. The benefits of 64-bit addressing will take years to be fully realized on the Opteron platform. In the meantime, AMD has our vote. We present the 2003 Microprocessor Report Analysts’ Choice Award for Best Server Processor to AMD for the Opteron processor, Model 848—the top of the Opteron line today and the best value in its class.

Microprocessor Report readers can access the full story here (2 pages/1 graphic): www.mdronline.com/mpr/h/2004/0209/180606.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Better, Faster, Cheaper: Take All Three
Peter Glaskowsky - Editor-in-Chief  {02/09/2004}

Now that we’re safely past the Internet bubble, its collapse, and the ensuing industry recession, it’s as if all the industry’s researchers have returned to more-productive work. The same talent that was once directed to create e-commerce technology for pet-food purchasing is now giving us secure operating systems and processor-design tools that automatically adapt instruction-set extensions to the specific needs of a target application. Hardware engineers, no longer creating Internet-connected microwave ovens, are now working on microprocessors that provide extraordinarily high performance and exceptionally low power consumption.

And, of course, the process-technology gurus keep marching forward. They have much more work to do in each new-technology generation, but they stay on target—every two years or so, we get new foundries making new chips containing about twice as many transistors per square millimeter. New 90nm products are just starting to appear, most providing huge improvements over their 130nm predecessors. We couldn’t very well have “Process Technology” as a nominee for our Analysts’ Choice Awards—it’s too broad a concept—but let’s give credit where it’s due. Process technology is the greatest contributor to progress in the microprocessor industry.

Advances in Reconfigurable Computing by companies such as Ascenium and academic research efforts such as the TRIPS project at the University of Texas at Austin earned this field a nomination for the Microprocessor Report Analysts’ Choice Best Technology award. It isn’t a new idea, but there’s a lot more technology behind the idea than there was in years past.

Tensilica is planning what should be a significant step forward in configurable-processor design with a new automatic processor-extension generator for TIE, the Tensilica instruction extension language. This technology earns a Best Technology nomination because it has the potential to help designers create better extensions with less manual labor. The generator may make it possible for even the smallest companies to create their own application-specific processors.

For the problem of transistor leakage current, Transmeta—a company without a fab of its own—has developed what appears to be an effective solution that can be added to standard CMOS processes. This technology, dubbed LongRun 2, can vary the threshold voltage (Vt) of transistors hundreds of times per second in a running microprocessor. LongRun 2 is present in Transmeta’s Efficeon processor, making it the most commercial of our Best Technology nominees. The full details of the mechanism behind LongRun 2 have not yet been disclosed, but its practical value is clear.

Until recently, most software developers focused on increasing sales by supporting new applications. Microsoft, for example, spent about 20 years adding features to Windows before a companywide email memo from chairman Bill Gates launched Microsoft’s Trustworthy Computing initiative in 2002.

The most significant element of Trustworthy Computing so far is the Next Generation Secure Computing Base (NGSCB) project, announced at the Windows Hardware Engineering Conference in May 2003. NGSCB, due to ship with PCs based on Microsoft’s next major operating-system release (code-named Longhorn) by 2006, requires hardware changes to the CPU, core logic, and I/O devices.

Microsoft released a developer preview of the NGSCB code at its Professional Developers Conference in October 2003. At the same time, the company said the first release of NGSCB in Longhorn will be aimed solely at business customers, especially government users and those in the financial and health-care industries. Consumer-oriented features must wait until version 2. We expect it to take several years beyond 2006 before NGSCB-compliant hardware is included in all PCs sold, and many years could elapse beyond that point before PCs are truly secure.

The sooner, we believe, the better. We give the 2003 Microprocessor Report Analysts’ Choice Award for Best Technology to Microsoft’s Next-Generation Secure Computing Base initiative.

Microprocessor Report readers can access the full story here (3 pages; 1 graphic): www.mdronline.com/mpr/h/2004/0209/180619.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Prescott Pushes Pipelining Limits
Peter Glaskowsky - Editor-in-Chief  {02/02/2004}

Intel launches its new 90nm Pentium 4 processor today, but there’s no party—no great cause for celebration. After two years of promises, hints, and rumors, the facts are somewhat disappointing. Prescott is being announced at speeds from 2.8GHz to 3.4GHz, but it simply isn’t good enough to take a place at the top of Intel’s x86 product line. Intel is also boosting the speeds of its 130nm chips, the Northwood Pentium 4 and Gallatin Pentium 4 Extreme Edition, to 3.4GHz. The latter product remains the highest-performance x86 processor Intel offers.

Consider the technology Intel put into Prescott. The core was completely redesigned, giving it 31 pipeline stages for simple integer operations—up from 20 in Northwood. Intel is using one of the industry’s first 90nm processes, with strained-silicon transistors and seven layers of copper interconnect with low-k dielectric.

All these changes turn out to produce no net clock-speed increase at launch and only minor improvements in architectural performance. According to Intel, the 3.4GHz Prescott’s SPECint2000 scores are only about 5.7% better than those of the 3.4GHz Northwood. SPECfp2000 scores go up about 11.5%. We attribute these increases mostly to the cache enhancements—not to the changes in the core.

The ultimate question must be this: Why did Intel bother? The facts we have don’t seem to justify the costs of the project, even given the 4GHz performance promised by the end of 2004. We believe the Northwood core, retargeted for a 90nm process and paired with new, larger caches, would have produced a smaller, faster chip.

There may be some other reason for Prescott’s existence. We can only speculate, but the Prescott situation may be the result of a combination of factors—a redesign needed to integrate the rumored Yamhill 64-bit extensions, a process that doesn’t yield the transistor or interconnect speeds Intel expected, plus the company’s commitment to 90nm fabs, which must now be filled.

We’ll learn more about Prescott at the International Solid-State Circuits Conference (ISSCC) and Intel Developer Forum (IDF) in February. Following those presentations, we’ll be able to take a more detailed look at Prescott, and perhaps then it will make more sense.

Microprocessor Report readers can access the full story here (2 pages): www.mdronline.com/mpr/h/2004/0202/180501.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

• Most recent Processor Watch articles
• Processor Watch archives