Transmeta, the exotic x86 house, is back for a second try at the brass ring having flubbed it big-time the first time through. This time it's taking on both Intel and Moore's (industry-held-sacred) Law. It claims Moore's Law will ultimately be overturned by leakage, the power that processors lose to transistor bloat, something to really look forward to when the industry goes to gazillions of itty- bitty 90nm and 65nm transistors and beyond. As a matter of fact, rumor has it that voltage problems have been bedeviling and delaying Intel's excursion into 90nm. See, the higher the speed, the smaller the surface, the more transistors, the more power is dissipated. Transmeta says, with a dire tone in its voice, that "Leakage power could easily dominate total chip power and prevent low-power standby operation if not controlled." Naturally, Transmeta says it's got a solution, but it isn't out of development yet. However, even with the number of transistors doubling every two years, there's no great rush since the voltage crisis, despite Intel's current problems, apparently won't crest until, oh, 2010-2015. Anyhow the secret sauce is in Transmeta's next-generation power management software scheme LongRun2, which - together with special hardware circuits - will let it dynamically control threshold leakage, it claims. It says software control is important in order to adjust leakage due to changes in run-time conditions such as voltage and temperature that are not pre-determined when the chip is manufactured. LongRun 2 will go into future versions of Transmeta's brand new this-better-work-or-we're-toast chip Astro, which has now officially been dubbed Efficeon, which is supposed to suggest the opening of an era of efficiency. When exactly this wonderfulness will occur is unclear. A prototype demonstration showed the widgetry cutting Efficeon's core leakage power by 70 times in standby mode from 144mW to 2mW and hundreds of times a second while playing a video game and DVD movie. The new chip family, whose development is supposed to go back to before Transmeta brought out Crusoe in January of 2000 and sampled this summer, starts with the 1GHz-1.3GHz TM8000 chip, which is supposed to outdo Intel when operating within the 7W thermal limits of a fan-less notebook design. (This is called the performance-per- watt-per-dollar contest and Transmeta's got a bunch of benchmarks saying it's better than Intel's Centrino-Pentium M.) Transmeta founder and CTO Dave Ditzel says the company pondered the shortcomings of its first chip, the Crusoe, and built the Efficeon accordingly. It involves a new silicon microarchitecture and a redesigned version of Transmeta's Code Morphing software. The microarchitecture is based on a 256-bit Very Long Instruction Word (VLIW) approach that can issue up to eight instructions per clock cycle, as against Intel's four, which means twice the throughput per clock of the Pentium 4. The Northbridge chip is integrated and this time through Transmeta provided for a high-speed integrated AGP-4X graphics interface rather than just a lonely PCI interface. It's got DDR-400 SDRAM with ECC memory as an option, something it wishes it had on the Crusoe. Efficeon uses a HyperTransport interconnect that can send data at up to 1.6 Gbytes/sec aggregate, which is 12 times faster than the I/O throughput of Crusoe's PCI interface. The chip is being made by TSMC in Taiwan on a 130nm process. Transmeta plans to switch to Fujitsu's 90nm CMOS process in the second half of next year to fetch speeds of up to 2GHz. Initially, there will be three Efficeon chips: one with a meg of on- chip L2 cache, one with a 512KB cache and a little guy that fits into a smaller package. They will compete with Centrino on space, cost, power and energy, Ditzel said. Efficeon is aimed mostly at the mainstream notebook market rather than the company's usual lower-volume ultra-light haunts although Transmeta is also hoping to pick up its share of business in blade servers, tablets, thin and light notebooks, embedded and - playing to the burgeoning digital media craze - silent desktops too. It mumbled an ambition to lead in thin clients next year. A day after it got its second-generation Efficeon launched, Transmeta posted terrible numbers for the third quarter. It lost $23.7 million, or 17 cents a share, on revenues that followed Crusoe demand down hill and dropped to a pitiful $2.7 million from 6.4 million last year when it lost $22 million. Analysts were expecting, oh, $5 million. Its losses, which included non-cash charges of $4.9 million, were in line with expectations. CEO Matt Perry said Crusoe demand fell off quicker than anticipated in mobile notebooks though the 5800 still has momentum in embedded designs and ultra PCs (UPCs). Perry claimed 5800 volumes would increase going forward as vendors launch their UPCs and thin clients like HP's models. CFO Olav Carlsen projected that Q4 revenues would be flat to up 50%, depending on how the Efficeon ramps, a forecast that evidently routed the stock. Losses should range around 13 or 14 cents. Obviously if Efficeon doesn't cut it, Transmeta, which will have $52 million in bank at the end of December, is through. Although Efficeon systems will start shipping this quarter, Perry said the company has "strong interest in Efficeon processors for designs beginning in the spring of 2004 in the mainstream consumer notebook segment. We are also receiving very strong feedback from leading notebook manufacturers for Efficeon processors based on 90 nanometer technology, which is expected to be in volume production in the second half of next year." Oh, yeah, by the way, being hungry for revenues, Transmeta is looking to license its power management technologies. The Efficeon is reportedly selling for around 100 bucks to the Crusoe's 20.

• The Christmas Day Bomber, Moore’s Law, and Enterprise IT