Crusoe: Transmeta Comes Out of the Closet - page 2
RISC to the Extreme
The Crusoe processors' key advantage, as described by Transmeta, is their ability to run standard programs while consuming a fraction of the power of a conventional microprocessor. Transmeta's LongRun technology, which slows the processor to save power when it's not active, is strikingly similar to the SpeedStep technology in Intel's latest crop of mobile Pentiums. But since Crusoe draws very little power from the get-go, slowing the CPU allows it to run even longer before a recharge is required.
Other small RISC processors, such as StrongARM and MIPS, also use very little power, and portable doodads with these chips often use similar strategies to prolong battery life. Transmeta hopes that the Crusoe chips' ability to emulate other microprocessors will give it an edge over these other low-power products.
The Linux Connection
Many Linux fans speculated that, because Transmeta had hired Linus Torvalds, Transmeta's products would be specifically oriented toward Linux. But this doesn't seem to be so. While the Crusoe chips run Linux, and the company will be offering a version of Linux which they call Mobile Linux (a version of Linux that can run from ROM and does not require a hard drive), the chips were shown running many operating systems, including Windows. (During the presentation, Dave Taylor, formerly of id Software, defeated Linus Torvalds in a networked game of Quake between two Crusoe systems. Taylor played on Windows, while Torvalds used Linux.)
Linux will surely be a good fit for some mobile applications of Crusoe. However, there are good reasons for vendors to consider other embedded operating systems, such as QNX, Wind River Systems' VXWorks, and PicoBSD (a close relative of FreeBSD). Manufacturers who embed Linux must reveal to the world any enhancements or customizations they make to the operating system. Companies that do not wish to show competitors exactly how they design their hardware will likely opt for another OS, though they may still use the Crusoe chips.
How Fast Is It?
During a question and answer session, Transmeta was evasive when asked to quantify the new chips' benchmark performance. When pressed, executives said that they did not believe that conventional benchmarks (e.g. WinMark) were the correct way to evaluate performance. This stance is explained at http://www.transmeta.com/crusoe/download/pdf/BenchmarkWhitePaper_1-1 8-00.pdf, which says,
Mobile computer benchmarks should measure performance in combination with the energy consumption and battery life penalty for that performance, and should address the issue of performance sufficiency as well as peak performance.
In short, because they must translate computer programs as they run them, the Crusoe processors will probably not be the fastest ones available to run Windows, Linux, or any other operating system. Rather, the company hopes that vendors will use the chips because they are "fast enough" and use less power.
Today, creators of handheld computing devices tend to write their own code from scratch, or license programs written in high level languages such as C, rather than using binary programs that have already been developed for other microprocessors. Therefore, it is not clear whether the Crusoe chips will be more attractive than other RISC chips to manufacturers who are designing products such as cell phones and DVD players. At the same time, the fact that Crusoe will not perform as well as a Pentium may limit its use in machines which must run Windows--which is sluggish, at best, on even the fastest processors. Has Transmeta made the right tradeoffs? Will mobile computing become ubiquitous enough to fuel Transmeta's success? Only time, and the market, will tell.