Originally posted by G-News
Over several years, which is nowhere near P4 sales in comparison.
Yes, but its beyond Apple's sales by quite a bit. And I'm sure MS is gunning for PS2 territory (or better) with 70+ million units. That is surely enough to justify the R&D costs of the chip and bring its actual cost down to the cost of actually producing it.
PS: The Gamecube is based on a 400 core derivate, last thing I knew. Why not assume a dual core version of that core family, maybe even a triple core version for some strange reason. That would make more sense in a lot of ways.
No, it is a 750 core derivative. Its only 32-bit, doesn't have vector processing units, and doesn't clock nearly as high.
I'm still trying to see what your objection is (and the following doesn't reflect anything I know from non-public sources)...
Do you just not believe that a near-200 million transistor chip can be a consumer commodity level part by the end of 2005? ATI and nVidia are shipping 130 nm and 150 nm 110-125 million transistor GPU parts right now at the add-in board level. 90 nm SOI is shipping from IBM right now, and they are spending very heavily to get to 65 nm quickly (so is everybody else). Naive arithmetic says that'll quadruple the transistor counts... meaning 400M+ transistors at the add-in board price point, and near-200M at a considerably lower price point. We already know they can ship a leading edge GPU in a game console (they did with XBox), so ATI can do it this time around. So why is it hard to believe in a customized CPU from IBM using the core technology from the established lineage POWER4, POWER4+, POWER5, 970, 970@90 (plus whatever we see between now and XBox2 launch)?
The 200M transistor figure might be too high. The 970 without its L2 cache and bus interface unit is probably only 30-40 transistors, so for 2 extra cores that's only 60 - 80M added to a 970 with a (presumably) boosted cache for a total of 120-150M. That's in the same ballpark as Intel's Prescott (125M at 90nm and shipping this week), and it may not clock as high. Power is not linear with clock rate so running at only 2-3 GHz the rumoured 3 core PPC would be much much cooler than Prescott (not to mention IBM's SOI technology which seems to limit power loss better than Intel has managed).
A couple of years ago I read a published paper from IBM (written circa 2000, IIRC) that predicted transistor counts in commodity level chips would hit 1 billion
by 2010. So far things seem on track to hit that, and 2005/2006 is easily in the 200M range. The paper was about how to use this kind of transistor count, and the conclusion of the author was that massive multi-core processors ("Cells") were inevitable. Nobody has figured out how to effectively use more than about 50 million transistors in a core, that much cache is just silly, so the only direction to go is multiple cores.