Sun to Take New T1 Sparc Chip Open Source
by Timothy Prickett Morgan
Sun Microsystems hosted its Network Computing Q4 2005 announcement extravaganza in New York today, and as we report elsewhere, the company rolled out the "Niagara" T1 eight-core processors and made its case for eco-friendly, cost-effective computing with the new T1000 and T2000 servers that make use of the chip. And in a surprising move, after several hours of positioning the boxes, Sun chairman and CEO, Scott McNealy, announced that Sun would be taking the T1 processor open source.
The ears of both the open source community--surely best characterized as Webs of individual communities (and Wall Street analysts) better described as a weary band of people who hope Sun starts making money someday so perhaps they can, too--pricked up. Open sourcing a chip? What on earth does that mean? And why on earth would anyone do that after having spent untold millions of dollars to invest in a differentiating architecture?
Good questions, and Sun did not provide a lot of detailed answers because, as McNealy explained, Sun has just made a commitment to the OpenSparc project and is still working out the details about how it will do this. "This is going to bend a lot of brains in the marketplace," McNealy conceded, with his characteristic wry smile. When asked what Sun would do when and if rival Intel decided to grab the T1 intellectual property that Sun is going to let loose and then make a clone chip, McNealy quickly countered, "Wouldn't that be cool?" The audience was not exactly with him yet, so he explained a little further. "If it works in software, why wouldn't it work for processors?" He went on to say that by open sourcing the T1 chips, Sun could envision other operating system suppliers--including the open source Linux and FreeBSDs as well as suppliers of embedded operating systems--having an easier time tuning their software for the chips. And McNealy said that open sourcing of the T1 specs would allow makers of other devices such as set top boxes and embedded controllers to build variants of the chips, or if someone wanted to make an even larger, more powerful chip, they could do that, too. "You don't know exactly where it is going to go, and that is the beauty of it," he said.
Sun is no stranger to licensing its processor technology. Fujitsu has been a licensee of the guts of the Sparc chip for a very long time, and has its own clone Sparc64 chip business, one with very different design points than what Sun is pursuing with the T1 and future "Rock" processors. Sun and Fujitsu are co-selling machines bearing the Fujitsu-Siemens brand, and have plans to sell the dual-core Sparc64-VI when they are available in late 2006, considerably later than Fujitsu had originally planned to get them out the door. Ross Technology, which was also a licensee of the Sparc technology from Sparc International, designed its own chips until 1998, when it ceased operations. There is a German company called Gaisler Research that makes embedded processors based on the 32-bit Sparc V8 specification. (The current spec that Sun and Fujitsu adhere to is called Sparc V9, which is the 64-bit specification.)
Sun said in a statement that the OpenSparc community will be up some time in the first quarter of 2006, and said it made the move to help spur participation in the processor and system design aspects of computer systems, much as it has taken its Solaris Unix variant to an open source community, OpenSolaris. Sun plans to publish specifications for the T1 chip, including the source of the design expressed in Verilog, the verification suite for the chip, and the simulation models for creating and testing it. Sun will also publish the instruction set architecture specification for the T1, which is unique from the other UltraSparc chips Sun has created even if it is binary compatible. Sun will also deliver a Solaris port tuned for the chip. "The goal is to enable community members to build on proven technology at a markedly lower cost and to innovate freely. The source code will be released under an Open Source Initiative (OSI)-approved open source license," the statement read.
While participation in a chip development community may be an admirable thing to try to foment, by the very nature of IT systems, this OpenSparc community will be much smaller than the OpenSolaris community or other communities, such as those behind Linux or FreeBSD. With software-based communities, you can build code, use it yourself, and set it free. All that you are really out is your time. With a chip community, using your brains to design a chip is the easy part; fabricating a chip you design--presumably in relatively low volumes--is another thing entirely. The amount of capital equipment you need to run a compiler and host a Web site is tiny, but getting manufacturing capacity from a chip fab is not going to be so easy to come by.
So what is Sun really after here? That's a good question, and Java is part of the answer. Sun launched Java because it wanted to Webify C++ and make a lot of PR noise for itself at the same time. Java has done wonderful things for Sun--including allowing it to get its hands on some of Microsoft's monopoly money. And, to its great credit, Java has helped created a more unified ISV and internal application development community and platform. (Microsoft is no slouch with Visual Basic and .NET, either.) What Sun didn't do is make a lot of money on Java, but it has made some and it got a lot more press and marketing for free than it could ever--even during the boom boom times of the dot-com bubble--have afforded to pay for with real money. Sun's Sparc servers made them the darlings of the data center in the late 1990s, but Java made Sun a household name--well, a businesshold name, at least. Java made Sun a player in the software and systems business. Perhaps Sun thinks that by taking the T1s open source, it can get a lot of PR and community mileage out of it. Sun is correct when it talks up the eco-responsibility and bang for the buck per watts aspects of the T1 chips, because these things matter. Sun did not, you will notice, open source all prior and current UltraSparc designs--just the T1s. Open source chips are only going so far, apparently.
What seems unclear is how giving away the intellectual property behind the T1s helps Sun make money in the short or long term. It might simply mean that Sun is letting go of something that it cannot hold onto very tightly any more. If Sparc is to live on as a platform, it is going to have to have more support than just in low-end and high-end systems from Sun and big boxes from Fujitsu-Siemens. That much is for sure. The value inherent in the intellectual property rights inside each T1 chip might be large, and it might be small. And you can bet that all chip makers will be doing the math and asking themselves why Sun has done this, and what will happen to them if they have to follow suit. It is fairly safe to say that the major chip makers--including staunch Sun supporter Advanced Micro Devices--think that this open source T1 chip initiative is crazy. Then again, every time one chip maker sues another, they end up settling and cross-licensing intellectual property anyway, so it is hard to say if keeping a tight hold on it is really all that valuable in the long run.