Psst. Mister. Wanna buy a broken chip? Sun may be making you that offer soon. It's preparing to introduce its eight-core Niagara top-end processor, but has been thinking aloud of shipping parts that don't quite make the grade as seven or six core options.
It might seem perverse to sell faulty units of a luxury product, but Sun is keeping a very old industry tradition alive. In the early days of a new chip, production errors can knock out a high percentage of parts while they're still on the wafer. Each error may affect only a tiny fraction of the chip area, but that's enough to ruin the whole circuit. If you make eight small cores instead of one big one, though, you'll only lose an eighth as much: in Niagara's case, a chip that will run 30 threads instead of 32. You can sell that, albeit at a discount.
And when you do sell that, something wonderful happens. Your full-function chips still command top dollar, but people who wouldn't consider paying that much start to buy the cheaper option. You get a larger market, more developers, more of a buzz — and all from silicon you'd otherwise have to throw away. Without effort, you have a basic and a premium product: a classic marketing posture.
The trouble begins when you iron out your production flaws. Suddenly, you don't have enough second-grade parts to keep the basic market supplied. That's no problem; you've still got lots of chips, so you can just mark a batch of good ones as faulty and sell them accordingly. Unless you're operating at capacity, you're still not hurting your premium sales, and good chips cost no more than bad ones to make, so nobody's the wiser and you're none the poorer.
Except the market soon learns that it's getting a bargain. People find that their cheap systems can be made to run as fast as the expensive ones — but because you don't guarantee that, mission critical tasks still have to pay full price. By the time consumer awareness has spread, though, you should be onto the next revision of the product and can start all over again.
As chip industry sleight of hands go, this is remarkably benign — it almost amounts to progressive taxation. In fact, the economics are so compelling you should expect to see the same from all massively multicore makers — meaning bargains for the sharp and profits from the rest. Not such a bad deal, after all.