Transmeta Crusoe CPU

Transmeta may be most notable today for its $273 million IPO on Nov 7, 2000 being the last successful tech IPO until the Google IPO in 2004. But on January 19, 2000, Transmeta's big news was its new Crusoe CPU.

How Transmeta CPUs worked

Transmeta wanted to take a different approach than more traditional Intel competitors like AMD or Cyrix or Rise, the technology behind the Vortex86. They were going to produce an x86 compatible CPU, but they were going to use a translation layer to do it. They would design a very efficient CPU, and in theory, they could place any translation layer in front of it that they wanted. Emulating PowerPC or ARM would have been possible if they saw the need to do it, and, indeed, they floated the idea of emulating PowerPC, MIPS, and Alpha in addition to x86. But by the time they had a product to sell, popular yet inefficient x86 architecture was the target that made the most sense.

AMD was doing something conceptually similar with the technology it acquired from NexGen, essentially translating x86 instructions into RISC instructions for efficiency. But they did it in hardware rather than software like Transmeta did. AMD never had any designs on putting any different translation layer in front of it. And while the AMD approach did all of the translation in hardware, Transmeta was doing most of the work in software.

The main idea of doing the translation in software was the first time you ran x86 code, it would run slowly because of the translation. But on subsequent runs, the translation was cached and the CPU knew what it could run in parallel, so it would be much faster.

The Transmeta Crusoe CPU

Transmeta's first CPU, Crusoe, could run at Pentium III-like speeds but was about 30% less efficient, so a 700 MHz Crusoe ran like a 500 MHz Pentium III. The performance was barely better than a VIA Cyrix III. But unlike VIA's solution, It wasn't pin-compatible with any Intel CPUs, so it needed its own special motherboard design. Transmeta didn't have fabrication plants of its own, so IBM handled manufacturing using its .18 micron process with copper interconnects.

Internally it had a 128-bit core and it executed bundles of instructions. The CPU would chain 32 bits of instructions at a time into either a 64- or 128-bit bundle, padding any unusable space with NOPs. Having two bundles of different lengths helped to reduce waste.

The embedded version ran at speeds of 333, 366, or 400 MHz and had 96K of L1 cache, with no L2 cache. The TM5400 for PCs had 128K or L1 and 256K of L2 cache and ran at speeds between 500 MHz and 1 GHz depending on the generation. The Crusoe supported MMX instructions but lacked SSE support, which the Pentium III had.

Transmeta's Crusoe CPUs saw use in low-power laptops, thin clients, and embedded applications. Sony used them in its third- and fourth-generation Vaio Picture Book laptops. The Bluecoat web filtering appliance used them for a while. But if you owned a computer at the turn of the century, it's much more likely to have had an AMD or Intel CPU in it. If you used one at work, it was even more likely to have an AMD or Intel CPU in it unless it was a thin client.

Transmeta ultimately lost $600 million trying to bring Crusoe to market.

Critical reception

Since Transmeta lost $600 million bringing the Crusoe to market, you would think it wasn't terribly successful. You would be right. In the May 2000 issue of Maximum PC, Tom Halfhill said it was cool technology, but overblown. And he panned the performance, noting a 700 MHz Crusoe ran at about the speed of a 500 MHz Pentium III. It's a shame he put that line in the middle of the article, because most of Maximum PC's audience would turn the page right after reading that. He needed to save that tidbit for the end to keep them reading.

Halfhill did say that to match Crusoe's power consumption, Intel and AMD would have to design new CPUs in order to had comparable battery life in a laptop. Unfortunately for Transmeta, that's exactly what Intel did. Intel responded with its power-efficient Pentium M in 2003. Transmeta sued Intel for violating its patents, ultimately settling for $250 million.

Also, the software emulation had overhead. It needed 16 MB of RAM to function, so your Crusoe-powered laptop had to give up 16 MB of RAM for the processor.

What happened to Transmeta

Even though Transmeta was the last big dotcom era IPO, issuing the IPO six months after the market had peaked, they were not among the survivors. In 2005, Transmeta shifted to licensing intellectual property rather than selling CPUs. And in January 2009, Transmeta sold itself to Novafora, who in turn sold the patent portfolio to Intellectual Ventures, a private equity company. Novafora ceased operations in August 2009, just seven months later. Intellectual Ventures licenses the Transmeta intellectual property to other companies on a non-exclusive basis. Companies like AMD and Nvidia use that intellectual property to make their chips scale their clock speed and voltage on demand, making them more power efficient.

Today, Transmeta hardware is rare enough to be interesting as a collectible. But there aren't a lot of people nostalgic for it, and that probably keeps prices low.