How Sun swerved to avoid Rambus
Roadkill on memory lane
MPF If semiconductor conferences had half a dozen Kevin Normoyles they'd be packed to rafters. Fortunately, we found one: Sun's very own project lead for its Hello-Pino USIIIi processor, and he proved fabulously good value. (Although much of what was offered can't be printed. And mostly because it defames us.)
Normoyle ran through the genesis of UltraSPARCIIi and UltraSPARCIIIi (aka Jalapeno) which he summarised like this:-
"At Sun, if a project is successful then the organization changes around you. USIIi wasn't devised by a big team, it was a bunch of us yahoos. But before USIIi you couldn't get a Sun box for under $10,000, and now you can get one for under $1,000"
So that gave the Jalapeno design team some clout. Now in most other companies, such a renegade project - one that cannibalizes the tasty margins of sibling product lines - would have been stomped into the ground in short order (cf "bradsilverberg"'s famous lowercase memo on Microsoft's corporate culture). But Sun seems to let such projects survive and prosper, which is pretty remarkable.
But one anecdote out of the unprintables is worth preserving, and it's the story of how Sun managed to sidestep the Rambus memory farrago.
One of Jalapeno's breakthroughs is asynchronous memory, which has a Moonie-like following amongst academics but which is rarely seen in the field. Reliable sources tell us that's because it's extremely difficult to instrument and test. Or as Kev puts it, "how do we know the data is there?"
Sun's challenge was that two years ago, Rambus had the field to itself. "It was brilliant technology, but from a business point of view, it had the market sewn up."
But the only alternative, DDR, didn't do what the Jalapeno team wanted it to do:
"We had this constant headache. The timings were screwed up."
The answer came from some academic work by Mark Greenstreet at the University of British Columbia, and work done in Sun Labs. Sun introduced ripple FIFOs and at the last moment the academic equations proffered by Greenstreet provided the answer. Normoyle was well aware of academic asynchronous memory work done in Manchester, England, but was able to adopt the techniques in the field.
"The cost to the industry of this whole RAMBUS-DDR war has been enormous - it's been a fiasco," he says.
Original Jalapeno's were destined for two-way systems, but the four-way was eventually accommodated. The adoption of JBUS came quite late in the day, and again was improvised from internal research, Normoyle credits Dave Bassett for the breakthrough.
(Incidentally, the name is inspired by Senor Jalapeno burrito parlour in Sunnyvale, home to the Sun chip team).
Now somewhat alarmingly, Kev expressed a murderous desire for revenge, including the suggestion of starting a website about The Register. Don't worry Kev - such reactions are normal amongst industry execs. El Reg has acquired some notoriety for revealing SPARC plans earlier than Sun likes to announce them, but this dear SPARC plugs, is the highest form of flattery:- we're paying attention, and you're worth the attention. And this, surely, is far preferable to being ignored. ®