Re: So that 6 month clock won't start ticking for quite some time. With Penryn being much closer in design to C2D based QC, would that cut down the qualifying time? You see where I'm going here.
I'm going to establish a rough time line here, and speculate on some target launch dates.
Piecing it altogether, I get: A-0 Tapeout in June A-1 Tapeout in Mid-August A-0 Silicon demoed to Charlie September 29 (3 months fab time) A-2 Tapeout in Mid-October A-1 Silicon back on October 29 (2 months fab time) A-1 Silicon demoed November 30 (1 month after silicon) A-2 Silicon back in time for Holidays, but not in time for demo (2 months fab time)
The first interesting thing here is that this tells us AMD's fab time (2 months for metal layer and 3 for full layer). That passes the sniff test.
Secondly, note that AMD taped out A-1 before even seeing A-0. This suggests they taped out A-0 hastily and found additional bugs in regressions, which caused them to tape out an A-1. And then, having found more bugs in actual A-0 silicon, they taped out A-2 a few weeks afterward. All this in parallel.
A-0 silicon was obviously not even demo worthy, and perhaps it made a pretty nice coaster. A-1 includes additional fixes found in pre-silicon, but obviously these were enough to boot Windows and run task manager. A-0 silicon probably gave them enough to probe, and maybe boot to their validation OS and run some low level tests. At any rate, it did enough for them to find more bugs and tape out A-2.
Now, there are several things here. A-2 may or may not be able enough to get more functionality than A-1. But AMD is clearly desperate here, so they are taping out as fast as they can to get more visibility into their silicon.
At the very least, they have A-1 silicon, which boots Windows and is probably good enough to get some good validation coverage. Their validation OS is probably recording most major bugs, enough for them to tape out a B-0 relatively soon (if they have not already). It would be better for them to wait for A-2 results, but they could easily jump the gun and tape out B-0 based on A-1 results.
So there are two scenarios: wait for A-2 results or tape out B-0 based on A-1.
1. They tape out B-0 this month (December) and get silicon 3 months later (March). 2. They tape out B-0 next month after they've had time to debug A-2, and get silicon back in April.
Option 1 means a very high risk that B-0 is not healthy, and AMD needs to spin a B-1, or even C-0. Option 2 is lower risk of B-0 being unhealthy, but a B-1 is still likely. Keep in mind that there are still bugs hiding in both A-1 and A-2 if AMD cannot run stress tests, and for a server part, this is even more critical.
So let's offer the next few scenarios:
1-a. B-0 silicon in March is healthy, but a B-1 is necessary to clean up the rest. Expect B-1 tapeout in April, and silicon back in June. They take one more month of validation, and launch in July/August. 1-b. B-0 silicon in March is unhealthy and needs a full metal stepping. Expect C-0 tapeout in April, and silicon back in July. C-0 may or may not be the last stepping. This needs to be explored further.
2-a. B-0 silicon in April is so healthy that AMD can finish validation in May and launch in May/June. This is their absolute best case scenario. 2-b. B-0 silicon in April is healthy, but needs one more stepping. Expect B-1 to tapeout in May, and silicon back in July. Take a month more of validation, and AMD can launch in August/September. 2-c. B-0 silicon in April is unhealthy, and AMD needs a C-0. Expect tapeout in May, and silicon back in August. C-0 may or may not be the last stepping, so we'll explore this further as well.
1-b-i. C-0 silicon in July is healthy. Add one more month for validation, and AMD can launch in August/Sept. 1-b-ii. C-0 silicon in July needs additional stepping. Assume C-1 tapeout in August, and silicon back October. Add one more month of validation, and AMD can launch in Nov/Dec.
2-c-i. C-0 silicon in August is healthy. Add one more month for validation, and AMD can launch in Sept/Oct. 2-c-ii. C-0 silicon in August needs additional stepping. Assume C-1 tapeout in Sept, and silicon back in Nov. Add one more month of validation, and AMD can launch in Dec/Jan.
Based on all of these scenarios, AMD's absolute best case is 2-a. This assumes that A-2 silicon tells all (which is unlikely), and that by some miracle, B-0 is healthy. But if this turns out to be the case, AMD can launch in Q2.
AMD's worst case is 2-c-ii, where they wait until A-2, which is still only based on A-0 findings, tape out a bad B-0 that still blocks bugs, needs to tape out a C-0, and still has to clean things up with a C-1. This could delay Barcelona all the way until Q1 2008.
Most likely, I see AMD productizing a B-1, which leaves 1-a or 2-b, both of which are Q3 2007 scenarios. I suppose it's also possible that it will be a B-2 or launch on C-0, which would probably end up being a Q4 2007 launch vehicle.
You can draw your own conclusions, but this is how the timeline works out. It's very aggressive, and let me also say, very EXPENSIVE. All those steppings cost money, but what else is AMD going to do? This is their only bread and butter part, so they will step it as many times as necessary, in parallel, with whatever data they have at their disposal.
After the first few A-x steppings, things will stabilize, and I think it will result in the scenarios I described. In the end, Q2 is extremely unlikely, but IMO, so is Q1 2008.
It will be Q3 (60% chance) or Q4 (25% chance), IMO.
Market Data 
Markets 
AI
