marconelly! said:
They're not going to let those chips sit in a warehouse for a year; they won't exist. In other words, STI is not going to mass make any Cell chips until yields are at an absolution maximum because it would be too expensive otherwise.
Well, as far as I understand, mass production
means the yields are high enough, and they indend to get there in Q3/4 2005. Therefore, they'd have to use those newly made chips for
something, no?
What I meant is that they're not going to make them prematurely. If yields are good enough by 2H 2005, then they'll start to make them. If not, they'll wait. However, I have a hunch that it would exceptionally difficult to get leakage under control, so even if yields are good, mass production will still be delayed to 2006. The same situation as the P4 Prescott as you may know about. Though this is aside from yields. Still that's 2006 either way.
Vince said:
What? No offense, but how can nobody call you on this? If you never enter mass production, the chip will allways be expensive as the physical size and basic error/area quotient won't change. The lowered cost comes from the initial lessons learned and the effects of economies of scale on the IC. Just holding back is only going to be a loss-loss scenario for them. You make it seem as if this is unprecedented for Sony...
Not to mention they blatently stated 2H 2005 for mass production. The Sony Investor call last quarter also said 2005 for the 65nm Cell IC, but that they can't exclude the possibility it will be out sooner.
:? I don't see how that would be. AFAIK, when you make a chip, you make a whole wafer-full of them. As long as the cost of each wafer stays the same, which should be the case as long as the fab is in full production, the relative cost of each chip will be completely dependent on yields and die size. If yields are bad, then they won't make any wafers, and "turn off" the fab, and save from the cost of operation of running the fab. Once yields are good, the fab will undergo full mass production and then the economy of scale will apply. Also read above. My basis on the leakage problem is partly based on the problems of the Prescott, so I'm expecting 2006 no matter what.
About XDR RAM, if the production of that isn't problematic then they can stockpile them. It's the CPU and GPU that really my main concern in terms in production.
About the 1TFLOP figure, don't want to burst you guy's bubble, but jvd is probably right. This is assuming that the Cell is what's described in the patent Panajev showed us a while ago, which, IIRC, has 4 PUs with 8 APU's each. This gives about 256 FLOPs per hertz, or that it needs to run at 4Ghz to achieve 1TFLOP. This clock speed is virtually impossible for a console or anything with a massive water-cooling solution. Even a quad core chip without any APU's going at 4Ghz is unbelievable at 65nm. It will just next year before we even see a one-core CPu going at 4Ghz. 4 of them, even with a whole process generation ahead, is just not plausable.
