@WBMW - K10 Ramp
Good analysis which should be well known for all board readers
, but for me one point is clearly forgetten, which you in-
directly mentioned, costs vs. price vs. segmentation.
As seen by Intel itself, QC isn't ramping to "gigantic"
volumes that fast, even if you have huge amounts of FAB-space.
When we further speak about Intel, we have to take into
account, that they could use 2 relativly small DIEs for the
Quad Core. What has AMD to do? They have to build a near 300
mm^2 monolitic DIE. I'm really wondering, how much desktop
and servervolume we could reasonable expect? I'm sure that
AMD will do all to support all server-needs, but what comes
after that - how much desktop will be available? They could
do pricing in a way, that "noone" commands these modells,
thats clear, but this makes no sense. Thats why, I always
give a much more focus on the DC DIE, which will be in good
volume over time and which will be THE main product for the
years to come. I think it should be clear, that QC from AMD
is no big volume in desktop untill 45nm finally appears. Its
up to everyone, when he expects that to happen. I think it
should be a safe bet, that first 45nm production volume will
at best appear at the end of Q3-2008, which would mean
production starts before mid 2008. I don't think thats
likely and I expect more mid Q4 for real 45nm volume, but
that has to be seen. That means, that after K10 appears, real
QC will not happen and all will depend on the DC ramp, which
will be THE main product by far - no question around.
What are the implications? I couldn't say it often enough,
when we now look at the C2D 1333MHz 4MB DC Cores, a 3G modell
goes away NOW! for 266$. Thats NOW! When we could trust some
AMD execs, real K10 desktop volume will appear in Q1-2008. So, is
anyone out here who predicts, that Intel will not decrease
prices in 3 quarters? I'm seeing a 3G "old" C2D core at
around 200$ in Q1-2008. Could be a few dollars more, could be
a few dollars less - a 183$ tag is not out of reach. When we
further assume, that K10 DC is roughly the same IPC, which
is good, a virtual (will be non-existant) 2,9-3G modell has
to be in the same $ tag, which means 200$. We know, that no
3G will be available at this point, but I wanted to show,
what that means. The next implication is the implication for
your "old" K8 modells, which will be at the start THE volume
driver - no way around. Lets imagine, how much you can get
for 3G K8, which is now around 170-180$, when your virtual
K10 3G DC CPU goes away for 200$? Keep in mind, what that
means for the whole K8 line. When we think a second about that
, we could also look a bit back to your 2006 stated high
ASP times, where AMD hold the performance lead. Ok, with what
product they achieved that? Right, with small DIE (84/86mm^2)
Athlon64 cores on 90nm. Now we look a bit down the line, how
the picture will look like with 65nm in the next time with
K10. We have near 300mm^2 Quad Core DIEs -> over 3x in DIE
size compared to 84mm^2 and nearly 3x, when we use the bigger
1MB Opteron (just over 100mm^2) at that time. And we could
compare the mainstream parts, for which the new DIE size is
still unknown. Because the L3 will be also in place, we
couldn't half the QC-DIE size and some connect logic has to be
in place too. So a K10 DC DIE on 65nm should be in the 155-
160mm^2 region, no way around. Could be even a few mm^2 more,
but I don't want to split hairs at this point. Is anyone here
how sees that picture - past vs. near future - 84/86mm^2
cores vs. 160mm^2 in the same segment - desktop volume. The
first one on 90nm, the second on 65nm already. Thats the
environment, when we compare past Margins and past ASPs with
the possible future and as you well done mentioned, Intels
lineup made more than a 180° turn ...
When we further look at the old price structure and the
possible new one, you could also feel the difference. Its fun
to look at the old 2005/2006 prices - thanks to JJ, which has all
the details in place. Just look here to get some impressions.
28/09/2005
epscontest.com
06/03/2006
epscontest.com
06/09/2006
epscontest.com
Look at the X2 and Server prices and of course at the Sempron
line too. With Intels 3G C2D DC modell now @266$ -> is anyone
here who could reasonable explain to me, why AMDs ASPs should
go up to old regions? Thats the problem I have to assume,
that GMs should go to the 40+ region with the new ATI integr.
which has historically much smaller margins.
BUGGI |
