Replies to post #16269 on Advanced Micro Devices Inc (AMD)

[chipguy]

Can you run the numbers for us and show how this means pretty decent yields?

Half of wafer starts are given away as keepsakes
to German schoolchildren taking tours of the fab.
Yeah, that's the ticket!

[HailMary]

I'll run some numbers, making some assumptions:
1. all production is making bartons, which according to this http://www6.tomshardware.com/cpu/20030210/barton-08.html shows the maximum possible die per wafer is 273 for bartons. I'm sure this is not exact, but it is close enough.
2. in Q2 AMD sold 7.5M processors (educated guess)

Case 1 (4000 wafers/week to production, 1000 wafers/week to process development, future products, etc.):

4000 wafers/week x 13 weeks/quarter = 52000 wafers/quarter
7.5M/52000 = 144 die per wafer
144/273 = 53%

Case 2 (4500 wafers/week to production, 500 wafers/week to process development, future products, etc.):

4500 wafers/week x 13 weeks/quarter = 58500 wafers/quarter
7.5M/58500 = 128 die per wafer
128/273 = 47%

Not so hot, but in reality, AMD is likely producing more than they are selling. Some of this may be going to invertory to be sold in Q4 (we'll probably see 8M or more units shipped this quarter). It is also likely a portion of the product is simply unsellable. The lowest speed bins have to be sold quickly, or they become worthless and a writeoff. Do you count this against yields? In my calculation above, I am, because I used units shipped, not units produced. I have also left out the portion of production going to larger sized products that yield much lower.

This is all just a game. Without really knowing exactly what mix of products, how much inventory is being written off, how many wafers are going to production vs. development/samples, none of us can truly say what the yields are. I would have to agree, however, that it looks like AMD should be able to ship closer to 10M units of a barton sized product per quarter. Perhaps they could if they went full steam ahead on just this product alone. Intel has this luxury, but AMD does not. They only have one fab to run all the possible experiments they can. Most of these probably yield a big pile of coasters.

The striking thing to me is, even with this low output and low asp, the chip segment of AMD is making money.

HailMary

[Petz]

<<XP yields will have to be pretty decent to maintain or increase Athlon XP units while increasing Athlon 64 units by 4x to 10x>>

<MCan you run the numbers for us and show how this means pretty decent yields?>

No, we don't know how many addional XPs AMD made in Q3 and put in inventory.

Petz