Bugs fail to knock PowerPC G5 off schedule
Motorola's next-generation PowerPC processor, the 8500, aka G5, may yet make it into new Macs early next year, Apple sources have claimed.
At issue are recent claims that Motorola is having problems with G5 yields, with many clocking way, way below the 1.6GHz the chip is expected to reach once it ships. We can't say we're entirely surprised - yields are always very low in the early days of a new processor, particularly those based on a new iteration of an architecture.
Yield problems will be very bad now and slowly get better through the life of the chip. Even now, we're sure Motorola is trashing a small percentage of G4s that don't make the grade. At some point a balance is reached where a sufficient number of dies that work at the right speeds to make volume shipment economic.
That point is echoed by our source. Current problems apparently include clock multiplier and external (L3) cache coherency issues. The former has been fixed, the latter still being tackled. Other bugs may yet emerge, our deep throat warns.
Again, we'd be very surprised if that wasn't the case. Apple and Motorola will be finding bugs right up until the chip ships inside Macs. What matters is that none of the glitches significantly impact performance.
If that happens - or if yields don't improve as expected - the G5's launch could slip from early 2002 to mid-2002, with Power Macs based on the part being announced at Macworld Expo New York rather than the San Francisco event. Apple CEO Steve Jobs, says the source, is shooting for a January launch.
We can see Jobs doing so, but announcing that top-end models won't ship until February or March, much as he did this year when he unveiled the 733MHz and 667MHz Power Mac G4s based on the second-generation G4-class PowerPC, the 7450.
Says our source: "At this time last year, 7450 yields were not much better than what G5 yields currently are."
Still, it's clear that the G5 won't go into volume production for a wee while yet. Previously, we heard that production would ramp rather more quickly than it now looks like it will. But that's the nature of the processor development beast and our source can't be blamed for their past enthusiasm. ®