Sun gambles its shirt on new threads

Sun has embraced multicore, multithread, open computing in a dramatic way. If these things matter, the company will do well
Written by Leader , Contributor

Twenty years ago, three products first put beak through eggshell: Windows, Tetris and SPARC. Of these, only Tetris has unambiguously lived up to its promise. Windows — we know what happened there. The SPARC processor design has had its successes and its failures: it has flown the flag for RISC designs through micro, hyper, super and turbo revisions.

Now we have Niagara, more properly UltraSparc T1. The multicore technology within Sun's new servers, it has if nothing else boldness on its side. Eight cores each running four threads make Intel's two core/two thread designs look, well, threadbare. Those 32 threads take around 70W in total or roughly 2W apiece: that's unbeatable in enterprise systems And Sun has made Niagara's design open. Hardware engineers can get the Verilog description — the coded description of the chip's internals — and do with it what they will. If any of these things matter, then Sun has given itself a huge advantage over AMD and Intel. If.

There's no if about performance — that matters. Yet performance here is almost impossible to usefully quantify. It's possible that Niagara will be the clear cost-performance leader for applications with 16 to 32 threads while simultaneously poor for more lightly threaded systems. Sun can trail a healthy gaggle of orders at launch: ether it does have a decent spread of applicability or some very generous early bird discounts are on offer.

There's no if about compatibility either. Older software must run. Sun says it does — but as always, that's only half the answer. 32 bit Intel software runs on Itaniums too, but with all the grace and speed of a one-legged hen. Of course, says Sun, some optimisation will be required to make best use of the new architecture. What that means, only time will tell.

Then there's the open source hardware. This is a long-term move by Sun: the only thing that will make highly parallel processing take off is lots of people who know how to effectively program for it. Such a community is best seeded by universities, start-ups and other blank-slate projects, all of whom have the chance to pick the best ideas on offer unencumbered by legacy. By making the Niagara hardware design freely available, Sun is making an investment in future markets: an idea so impeccably capitalist as to make fools of those who claim open source is in some way anti-commercial.

In all these ways, Sun is betting that cleverness and innovation will beat off the mindless monsters of mass-market design. That's always a risky strategy. We hope it pays off — the vitality of the industry depends on new ideas winning through. If in twenty years time, it's impossible to pick three ideas from 2005 that have survived and prospered, we'll have lost something very important.

Editorial standards