Client Login
Search
MDR Home

Microprocessor Watch


Issue #16 MicroDesign Resources --- September 15, 1999

Editor: Michael Slater
Contributors: Linley Gwennap, Keith Diefendorff, Peter Glaskowsky

In This Issue:

  • Rapid Pentium III Price Cuts Continue
  • S3 Preps Savage2000 for New Millennium
  • Nvidia GeForce 256 Adds Geometry Acceleration
  • MoSys Explains 1T-SRAM Technology
  • Editorial: Via Shopping Spree Reshapes x86 Market

Microprocessor Forum is less than three weeks away! Check out the updated agenda and register today at http://www.MDRonline.com/mpf. Don't miss Intel's disclosure of the Merced microarchitecture, details on Coppermine, AMD's Athlon for servers, Via's new x86 plans, the first single-chip information appliance, a radical new programmable chip for multimedia and networking, new DSPs, and much more!

Rapid Pentium III Price Cuts Continue

On 9/12, Intel cut Pentium III prices for the third time in three months. At $615, the Pentium III-600 is now 25% less expensive than at its July introduction. The price of a Pentium III-500 has fallen 41% in the same period. Prices will probably fall further when the 0.18-micron Pentium III (aka Coppermine) is introduced next month. Intel has indicated that Coppermine will debut at speeds of at least 700 MHz.

The company also cut the prices of its other processors. Celeron prices were trimmed by 8% to 15% from the August pricing; no new products are expected in this segment until next year. Mobile prices, which had not been cut since June, dropped by 30% to 40% in preparation for the Mobile Coppermine introduction next month. --L.G.

S3 Preps Savage2000 for New Millennium

With its announcement of the Savage2000, the first mainstream 3D chip with an integrated geometry engine, S3 has regained the position of technology leadership it lost in the industry transition to 3D graphics. The rendering engine on the faster Savage2000+ variant can draw two pixels per clock at 200 MHz, with one or two bilinear-filtered textures per pixel. The chip's 400-Mpixel/s dual-textured throughput is better than that of Nvidia's competing GeForce 256 (see below), but the Nvidia chip is faster in other modes.

S3 (http://www.s3.com) plans to sell the Savage2000 for just $29, while the Savage2000+ will list for $35. These prices are meaningful for only a few board makers, such as Number Nine, that use these chips in niche markets. S3's purchase of mainstream board vendor Diamond Multimedia means that the Savage2000 will not be offered to direct competitors. --P.N.G.

Nvidia GeForce 256 Adds Geometry Acceleration

Close on the heels of S3's Savage2000 comes Nvidia's own geometry-accelerated PC graphics chip, the GeForce 256. Fewer details are available about the Nvidia product, but it looks to have rendering performance comparable to that of the Savage2000. Nvidia says the slowest speed grade of the GeForce 256 will run at 120 MHz and issue up to four pixels per clock for a pixel-fill rate of 480 Mpixels/s. The company has yet to determine the maximum speed of the new design.

As its name implies, the GeForce 256 includes a 256-bit rendering engine. This engine is coupled to a 128-bit local-memory controller that supports up to 128M of DDR SDRAM (vs. 64M of SGRAM on the Savage2000). The 23-million-transistor chip is being built in a 0.22-micron five-layer-metal process. Nvidia (http://www.nvidia.com) has not announced pricing or availability for the GeForce 256, but we estimate the chip price is as much as $50 for the fastest versions. Boards should be available for the Christmas buying season. Nvidia also announced deals with six board vendors and five major PC OEMs, showing that the new chip is already a major force in the industry. --P.N.G.

MoSys Explains 1T-SRAM Technology

For the first time, MoSys has offered a complete explanation of the operation of its unique 1T-SRAM technology (see MPR 8/3/98, p. 9), used in chips from MoSys or as a core that can be integrated into ASICs from various foundries. The 1T-SRAM can be made in pure logic processes for higher speed, or with an embedded DRAM process to achieve greater density.

Though MoSys (http://www.mosys.com) calls the design an SRAM, it is based on single-transistor DRAM cells. As with any other DRAM, the data in these cells must be periodically refreshed to prevent data loss. What makes the 1T-SRAM unique is that it offers a true SRAM-style interface that hides all refresh operations from the memory controller.

This goal is achieved by a combination of a high-speed core and a unique on-chip cache. All necessary refresh operations can be performed in parallel with external accesses that are handled by the cache.

Perhaps the most valuable design win the 1T-SRAM will ever see was revealed recently when Nintendo announced it will use the MoSys design in the Flipper graphics chip for its forthcoming Dolphin video-game console (see MPR 5/31/99, p. 5), due out late next year. Nintendo expects to sell millions of the Dolphin consoles, making it a lucrative customer for MoSys.

Analog Devices, NEC, and TSMC have taken 1T-SRAM foundry licenses. MoSys says it has taped out 20 different 1T-SRAM designs at these and other foundries and expects additional announcements this year. These deals, and the Nintendo win, should give 1T-SRAM a bright future. --P.N.G.

Editorial: Via Shopping Spree Reshapes x86 Market

Via Technologies has made a bold entry into the x86 processor market by acquiring both Cyrix and IDT's Centaur group. This unusual tactic raises questions about how the Taiwanese company will combine the two design teams and two product lines. If Via plays its cards well, however, it could be a potent competitor for Intel and AMD in the low end of the PC processor market.

The first question is why Via bought both companies instead of just one. When Cyrix went up for sale, it looked like a better deal than Centaur, with a larger market share and a more powerful processor core. But upon further analysis, the advantages of Centaur's technology became clear.

Centaur's forthcoming WinChip 4 core has half as many transistors as Cyrix's next-generation Jalapeno core, making it much less expensive to build (see MPR 8/2/99, p. 1). In a 0.18-micron process, WinChip 4 will also dissipate less power than Jalapeno and is likely to achieve a higher clock speed, a metric that is more important than raw performance in the low-end PC market. Furthermore, Centaur's lean 60-person design team fit into Via's revenue projections much better than Cyrix's bloated staff of 330.

Sources indicate that Via was initially willing to pay as much as $300 million, or about one times annual revenue, for Cyrix. But once Via got a good look, it decided to acquire Centaur while renegotiating the Cyrix price down to $167 million, with much of that price dependent on future Cyrix revenues that may not be achieved. This reduction is probably more than enough to cover the cost of buying Centaur. The pair of purchases lets Via pick and combine the best assets of each.

Which assets will Via choose? Given the superior characteristics of the WinChip 4 core and its efficient design team, all indications are that Via will focus on that core in the future while discarding most of the Cyrix designers. Cyrix is ahead of Centaur in developing an Intel-compatible Socket 370 interface, however, so Via may keep some of the Cyrix team to graft that interface onto the WinChip 4 core.

Via may also choose to market future WinChips under the Cyrix brand, which has more recognition than WinChip, particularly in the U.S. market. In fact, Via could market the current WinChips under the M II brand, since they are plug-compatible with the current M II and could be PR-rated to deliver the same performance.

In the low end of the market, Intel has effectively used Celeron price cuts as a club to beat its competitors senseless. I believe the only way for a competitor to survive in this segment is to match both Celeron's clock speed and its socket with a lower-cost product. Matching the socket is necessary, because there isn't enough profit in this segment to support a non-Intel infrastructure. Competitors must have a lower manufacturing cost to generate profits from their lower prices and lower volumes.

If Via produces a 0.18-micron Socket 370 version of WinChip 4, it could follow this path to success. Centaur's compact core will cost less to build than Celeron. With a Socket 370 interface, Via can ride Intel's coattails into standard PC motherboards.

As Via pursues this strategy, its enemy is not just Intel but AMD. That vendor's K6 parts are a big seller in the Celeron space, and AMD plans to push its new Athlon processor into the low end next year. Although Via has been working on a Slot A chip set for Athlon, I doubt that chip set will see the light of day, given the company's new strategy. Why build chip sets that support only your enemy? With Via and Intel controlling more than 90% of the chip-set market, that leaves only a few also-rans (and AMD itself) likely to support Slot A.

The final question concerns patent rights. Via will use National to fab at least some of its processors, but sources indicate that National's patent license with Intel does not extend to third- party products. IBM is an Intel-licensed foundry, but Via would rather shift production to lower-cost Taiwanese fabs. Intel has already sued Via over chip-set patents, and a suit blocking Via from selling unlicensed processors appears inevitable.

The Cyrix purchase includes some patents, and Via may hope to build a patent portfolio strong enough to launch an effective countersuit against Intel. Alternatively, the company could try to market its chips only in Asian countries with weak patent laws, but that would greatly limit the potential market.

If the company can overcome its patent problems, it has all the tools necessary to successfully develop and market low-cost PC processors: an efficient design team, a low-cost CPU core, a Socket 370 interface, and a base of previous customers totaling 5% of the market. As AMD's poor example shows, strong execution is also required to successfully compete against Intel, so we must wait to see if Via can effectively use its new tools. --L.G.

LATE ADDITION TO MICROPROCESSOR FORUM AGENDA: Via will present new information about its microprocessor plans. Register today at http://www.MDRonline.com/mpf.

 


Other Microprocessor Watches
Most Recent, 2000 Articles, 1999 Articles
 

 

 

 

 

Privacy Statement Site Index Help Contact Us Subscribe
Copyright © MicroDesign Resources