Willamette won't launch at 1.5GHz this autumn theregister.co.uk
It seems the ever cheeky Register wasn't much more taken by the vaunted 1.5/3ghz demo than Anand or Tom. But I seem to have made a mistake, it's Albert Yu, not Joseph. Oops.
Shoot, normally I'd excerpt an article this long, but it's all amusing enough, I got to post the whole thing for posterity, with a little bold added in the extra funny parts, plus some random secondary notes from moi.
We were privileged this afternoon to have a round table briefing from Dr Albert Yu, a senior VP at the Intel Corporation, and the man who unleashed a 1.5GHz on the world stage earlier today.
We had some of those pesky questions to ask him about the product, and in the process, unearthed some interesting anomalies.
Dr Yu refused to say how much on-die cache was on the processor he introduced today, would not give a delivery date for the product and when we asked him about the die size and how many additional transistors were on the Willamette, he said the die size was "slightly bigger" than the Coppermine.
However, he did say it was unlikely that when Willamette launches on the 1st October or thereabouts, it would reach such 1.5GHz speeds. He said: "It's unlikely it will launch at that speed. This is a very first raw look at the silicon."
Which brings to mind the old Firesign theater thing, Eat it! Eat it Raw!Raw Raw , Raw, that's the spirit we have here!
Further, said Yu, from one generation to the next you can expect a 30 per cent increase. As Intel will "very shortly" sample 1000GHz Coppermines to its key customers, it won't be hard for those people with calculators to figure out what clock speed it will launch at.
Willamette, unlike Foster, its next-gen server platform based on the same core, will be based on Rambus technology. Foster, the server/workstation version of Willamette, will go out of the door with DDR memory.
The reasons for this so far remain unclear, but may be something to do with what Intel customers want.
Implying desktop customers want to pay 4x as much for their memory? This is extra confusing anyway, isn't Rambus vs. SDRAM, DDR, whatever a chipset issue, not a CPU issue?
Yu said: "The current version of the [Willamette] chipset only supports Rambus. Synchronous memory performance will be mediocre. For the forseeable future, Rambus will be on desktops."
Scratching my head again, it sounds like Intel will stick with the 820-style Rambus or really sucky SDRAM through memory hum translator solution for desktops, which sure seems like a good thing for Rambus stock, but it's totally unclear who else it's supposed to be good for.
The extra instructions in Screaming Sindie II will help give Willamette a strong position in the marketplace, said Yu.
He also confirmed that the ALU (arithmetical logical unit) in the Willamette, runs twice as fast as other slabs of the microprocessor, and that meant the CPU had two different clocks.
The Pentium brand will not go away, said Yu, but he declined to say whether Willamette will be called the Pentium IV.
Nor would Yu be drawn on exactly how Intel will manage the transition between manufacturing Willamette in 100s of thousands by year end and millions next year, and how exactly Intel will manage that transition between the Coppermine processor and Willamette, given that its sixth .18 micron fab starts up in the second half of this year, and that Chipzilla will start to manufacture in .13 micron at the end of next year. Nor how the transition of wafers from eight to 13-inch will impact this whole process.
One unrelated aside. Not to make a big thing of it, but I'd recommend the ignore button on the esteemed Mr. Engel, this thread has too much volume and noise already without the addition of totally pointless back and forth with Engel. At least the more technical Intel droids post things worth thinking about.
Cheers, Dan. |
