wbmw,
On 90nm in H1 2006, you would be looking at a ~400mm^2 die running at around 1.6GHz to hit 100W power envelope.
That sounds like the right ballpark.
What a pitiful launch that would be. Almost zero volume and just barely more performance than the 275.
I disagree. The performance would be great. Some bandwidth demanding apps would be starved, but that's not exactly the rule for all apps to be bandwidth starved. K8 does not have a hyper-active prefetcher like P4, which causes a lot of bandwidth use, some of it may end up being unnecessary. So for things like database apps, which are moderate users of bandwidth, the performance would be very good. Significantly higher than 275.
Just think of the 4 way Opteron system and the traffic between the cores. Now picture single socket system, with lower clock speed, but latency of coherency checks basically eliminated to close to zero.
There were some people on these threads who claimed that the 4 way Opteron systems would be severely limited because of this traffic, so it was a negative. Now that this traffic (at least a big portion of it) disapppears, it is not a positive...
What a pitiful launch that would be. Almost zero volume and just barely more performance than the 275.
Well, Rink has been talking about a fake quad, like the fake dual that intel has, basically 2 dies in the same package. Maybe you could claim that to be pitiful, but a real quad Opteron would be nothing short of stunning achievement. As far as volume, your previous predictions of volume / availability of AMD DC have not proven to be right. QC volume would not have to be that high anyway.
On first gen 65nm in H2 2006, the die size ought to be inline with today's dual cores. A 100W power restraint might enable 2.4GHz; although, dual core will have moved to 2.8GHz by then.
I don't know about timing of the 65nm, but the clock speeds you are talking about would certainly make it a performance leader.
Again, without some additional features, it seems like a lame duck, and a huge risk to coincide with initial 65nm volumes (i.e. delays likely with this approach).
I don't know where you see Intel, performance wise in that time frame, and what could possibly compete with hypothetical 2.4 GHz QC. I don't really see anything on the horizon. Merom derived core may be very competitive with a hypothetical 2.8 GHz DC, but would be blown away by 2.4 GHz QC.
As far as delays, if you are about to introduce an industry leading product, and it still is an industry leading product 2Q delayed, I don't see the risk. QC would certainly not be delaying other product. The process technology quality and ramp are the bottleneck.
"F"-stepping, maybe?
Hmm... could be.
My expectation is 0% chance in H1 2006, 10% chance in H2, and only then as a rushed low volume product, and a high chance that it coincides with AMD's new micro-architecture in 2007, pending no major delays, of course.
Prior to Groo's rumors, my guess would be similar (slightly higher probability for H2 with new socket, new process.
Joe
Market Data 
Markets 
AI
