Tony said:
Well, on the latter case they seemed to have done pretty well (though
AMD64 was definitely not the only reason for IA64's rather limited
success), but they aren't exactly taking a huge amount of market share
away from Xeon. There was something like 1.4M Xeon servers sold in Q2
vs. 60,000 Opteron servers. This gives the Opteron only about 4%
market share. I guess this is a lot better than 0%, though at it's
height the AthlonMP managed something like 5 or 6% of the global
server market, so the Opteron hasn't even reached that stage yet,
despite signing up some big OEMs.
I found this new article which gives the actual number of server chips sold:
http://www.thestreet.com/_yahoo/tech/semis/10181117.html?cm_ven=YAHOO&cm_cat=FREE&cm_ite=NA
http://tinyurl.com/3mfo4
In the second quarter of 2003, AMD shipped a mere 110,000 server chips
compared with Intel's 4.6 million shipments, according to Gartner. Since
then Opteron scored some major design wins, helping it nearly double
shipments to 205,000 as of the second quarter this year.
But it still lagged far behind Intel's 5.4 million shipments.
Err, uhh, Yousuf, either you're cut 'n paste is a little wonky or they
changed the article since you read it. Now it reads:
"In the second quarter of 2003, AMD shipped a mere $110 million in
server chips compared with Intel's $4.6 billion in shipments,
according to Gartner."
Note the dollar values instead of units shipped. Lat though they do
seem to be switching back to units shipped:
"Since then, Opteron scored some major design wins, helping it nearly
double shipments to 205,000 as of the second quarter this year.
But it still lagged far behind Intel's 5.4 million shipments."
Hmm... strange.
Now 205,000 chips into the 60,000 servers (previously stated) equates to
about on average 3.4 processors per server. Considering that the vast
majority of Opteron servers are usually either 2P or 4P, that makes complete
sense. And since the number is closer to 4P than to 2P, that would indicate
that more 4P Opteron servers got sold than 2P ones.
I'm not sure that this is accurate as it might also include some
AthlonMP chips where the 60,000 server number might just be for
Opterons.
Also the two numbers came from two different companies, so I wouldn't
be surprised if they are not measuring quite the same thing, it
certainly would not be the first time that Gartner and IDC came up
with conflicting reports.
So it would seem, that Opteron's multiprocessing capacities are being
exploited to their utmost. Once 8P Opterons come into more common usage, it
would be interesting to see if corporations are utilizing their capacity
will be utilized too?
I really doubt that the 3.4 processors/server number is accurate, it
seems just way too high considering that Sun only just recently
started selling 4P Opterons, IBM never sold them and many small OEMs
also stick to only 1 and 2P Opteron servers. I would be VERY
surprised if AMD is really selling more 4P Opteron servers than 2P
ones, it just doesn't fit the market dynamics at all.
Of course, part of the confusion might be related to dollar value vs.
unit shipment confusion mentioned above.
Wonder how many Xeon servers were sold that same quarter? That way we can do
the same math and find out what the average number of processors there are
in a Xeon.
Roughly 1.4M Xeon servers were sold in Q2 of 2004. I don't know the
exact number, but it was somewhere around 1.6M total servers and about
90% of them are x86.