Vol 19, Issue 22

May 31, 2005

SPF05 Content Cornucopia

By Kevin Krewell

As I write this, we've just completed the Spring Processor Forum for 2005. The show produced an incredible amount of content, and Microprocessor Report's readers have already seen stories on ARC International, Marvell Semiconductor, Raza Microelectronics, and Xilinx. We have

more articles coming in the next few weeks, based on presentations from the Forum and from the multicore seminar. We would like to thank all the attendees, presenters, and sponsoring companies. This year we moved the show to the Doubletree Hotel, which turned out to be a popular decision. The hotel proved to be a more convenient and flexible locale.

This year's Spring Processor Forum (SPF) content covered a wide gamut, from a panel discussion on multicore benchmarking to intellectual-property (IP) cores for audio and video processing. We had presenters from all the major IP processor-core vendors. We also had a special presentation from IBM's Jim Kahle on programming concepts and open-source tools for the Cell processor, along with some demo video showing how ray casting can work on Cell's unique architecture.

We had the good fortune to have Mentor's chairman and CEO, Walden (Wally) Rhines, keynote on Tuesday. Picking up on all the discussions of Moore's law, Rhines talked of how the so-called law is actually an empirical observation that was based on the effects in the semiconductor industry of the more widely observed learning-curve effect. Learning curves apply to many other industries, and Rhines covered the various effects on different parts of the design, manufacturing, and testing of chips.

Interestingly, Rhines talked about the way the dominant cost of new-chip development is design verification, not mask cost (despite all the talk of million-dollar mask-set costs). The positive effect of good verification I've noted in two recent projects—Sun's Niagara and RMI's XLR processor. Both projects had a close working relationship between the design team and the verification team. As a result of the close teamwork, very early silicon worked extremely well. On many larger design projects, the design team builds the design and then "throws it over the wall" to the verification team. This bureaucratic approach often appears in entrenched organizations of larger, established companies. Smaller startups cannot afford such a luxury, and the Niagara team was formed as a startup (Afara WebSystems) and then bought by Sun Microsystems. The Afara team managed to maintain its team approach even after being swallowed up by the much larger Sun.

Multicore processor designs were a popular topic, and we expect to see more on this topic in the future. For Fall Processor Forum and Processor Forum Taiwan, we will likely expand our coverage of multicore processors and supporting IP.

Real Battlefields at E3

While we were busy in San Jose with DSPs, IP cores, and multicore processors, down in Los Angeles, Microsoft, Nintendo, and Sony were battling for the minds (and wallets) of the gamers and game developers. The common thread among all three vendors was that they all used IBM-designed Power processors for the main CPU. IBM Microelectronics has become the arms merchant of the console wars. The GPU designs are split, two designs for ATI (Nintendo and Sony) and one for Nvidia (Microsoft).

The interesting thing about the processor designs is the way Microsoft and Sony have taken very different approaches to solving the same problem (high-quality, high-definition, and highly immersive games). Microsoft's processor design for the Xbox 360 utilizes three symmetric 3.2GHz PowerPC processors with a shared 1MB L2 cache, and the processors share the main memory with the GPU (the memory controller is on the GPU) in a unified memory-architecture design. The reason the design used a very un-binary number of processors was likely driven by die-size considerations.

The Sony PlayStation 3 (PS3) uses the more radical Cell processor. After the early talk of 4GHz Cell processors and 256GFLOPS/s, the PS3 design came down to the realities of chip power and yields. The Cell processor in the PS3 will use seven of the eight available synergistic processing elements (SPE), with the eighth saved as a spare, and will run at 3.2GHz. The decision to spare one SPE is obviously a yield enhancement, allowing the use of die where not all eight SPEs are functional. The clock-frequency decision reflects the need to build a processor within a limited power budget. The PS3 design splits DRAM memory between the Cell processor and the GPU.

The generally accepted wisdom is that the Microsoft solution will be easier to program, but the Cell processor may have better performance after programmers figure out how to program it. Different solutions, same problem.

How Dumb Does Apple Think the Press Is?

Maybe that's a rhetorical question. I raise it following the story headlined "Apple Explores Use of Chips From Intel for Macintosh Line" in the May 23, 2005, edition of The Wall Street Journal. The story outlines alleged talks between Intel and Apple on using Intel processors to replace PowerPC processors in Macs. This is not the first time this rumor has surfaced, but it is the first time it's been reported by a prestigious newspaper on the basis of multiple sources.

I find it hard to believe these talks are anything but a ploy by Apple to get IBM's attention. Apple may have been feeling neglected because of all the IBM activity at E3 last week and wanted to send IBM a message. IBM has been very busy designing processors for all three major console gaming platforms and may be stretched too thin.

For Apple and its software infrastructure, the software transition would be expensive, as it would create additional software configurations to support. The Intel architecture has no inherent advantage over the PowerPC's, so I don't think there's a technical advantage to the Intel architecture that IBM cannot duplicate. IBM already has a dual-core server processor (Power5) and is working on the tri-core PowerPC for Microsoft's Xbox 360. IBM is reaching 3.2GHz with both the Cell processor and the Xbox 360 processor. The only technology IBM is not particularly good at is low-power processors; it lacks the equivalent of the Pentium M. Apple is probably very anxious to get a notebook version of the G5, and IBM has been slow in responding. IBM may be in the same situation that Intel was in: its main core is deeply pipelined for server and high-performance desktop markets, but mobile designs are more power efficient with shorter pipelines and other core modifications, like those Intel made in the Pentium M. With all the other commitments IBM has, does it have the design resources to launch another processor-core design?

In addition, if Apple switched to the Intel architecture, it would then have to launch its systems in time with Intel products and the rest of the PC business. Where is Apple's advantage and unique timing? With such a small percentage of the market, Apple would shortly find itself in line behind Dell, Gateway, HP, Lenovo, and others for Intel's attention. Apple has often driven tough deals with its processors' suppliers IBM and Freescale. I doubt Apple can drive the same bargains with Intel; it might have better luck with AMD.

If this rumor were true, I'd expect Apple to be livid with the WSJ for releasing "trade secrets." (Apple recently sued a number of websites for releasing leaked information about the new iPod.) If the leak came from Intel, that act would scuttle any talks immediately. Apple is renowned for secrecy; if Intel leaked information now, Apple would never trust that company in the future.

What is the real fallout of the story? IBM is under pressure to make amends with Apple; Apple gets IBM's attention; the WSJ get some free publicity; Intel gets to participate in the public humiliation of rival IBM. And I get about a dozen calls from various reporters.

Next rumor? Dell hints that it likes the AMD (dual-core) Opteron processor and might use it. Oh wait, it's not quite time for that one again.