Yeah, but they'd need to stock those extra parts everytime they add
another Intel product line too, for example, Pentium M, Pentium 4 Socket
478, Pentium 4 Socket 775, P4EE, Celeron, Itanium, and their various
speed grades. Actually in the case of most of these mail-order PC
places, I've never seen them stocking much more than two speeds of the
same processor, usually the top two most common speed grades.
Yes, but they can share those chips between multiple product lines for
the most part. The same socket 478 P4 chips and Celeron chips were
used in all of their Dimension and Optiplex lines with only small
variations. But you do have a point here about the socket change to
Socket 775 being a bad thing for them, and I suspect that they'll try
to switch over as many products at once to minimize the impact of
this.
As for the other chips, the P4EE is a bit of an odd-ball, but they
sell in sufficiently expensive machines that it's probably worth it
for bragging rights if nothing else. They don't need to stock any
Pentium-M chips since they don't make laptops (they just sell the
final product with their name branded on top). As for the Itanium...
err.. well there's no real explaining for that one. My only guess
here is that Intel is providing a SIGNIFICANT portion of the funding
(like all of it) for Dell's Itanium systems, because they just do NOT
fit the Dell model at all.
In fact, I think AMD produced the Socket 754 Sempron just for this case,
to minimize the number of parts needed to be stocked, at the insistence
of HP.
Probably. AMD is definitely learning and it's paying off for them
with their HPaq deal. HP sells quite a large number of AMD-based
systems, including some in their commercial-grade systems.
In fact, one could probably argue that Dell would theoretically be
better off going exclusively with AMD chips than they are with only
Intel chips. However this really just isn't an option for a variety
of reasons.
I don't think there's much doubt anymore that Dell will have to start
selling Opteron very soon. Intel screwed up its 90nm manufacturing
process by not going with SOI, and Prescott was supposed to mask this
misstep by going fast even without it (eg. 50% extra pipeline stages).
This worked for Intel about 4 years ago when it introduced Williamette,
it masked their last screwed up manufacturing process which was 180nm
without copper. But even Prescott couldn't bypass the laws of physics
this time, and Intel's bacon is now cooked, because Intel won't be able
to design a completely new processor for at least 3 years. Opteron is
the only option for the foreseeable future. For all of Intel's vaunted
manufacturing skills, it really bets on the wrong horse quite often.
It certainly seems that way for the moment, though these things can
change. I suspect that once we start seeing dual-core chips coming to
market, the advantage for the Opteron is going to become even more
apparent. Combine that with the fact that AMD might have up to a
6-month head-start on dual-core chips and Dell could be in a rather
poor position if they don't sell Opteron systems.
On the flip side though, AMD's 90nm process isn't exactly making
waves. Sure they got the power consumption down nice, but so did
Intel if you look at it from a per-transistor perspective (which
doesn't help much when you more than double the number of
transistors). However the clock speed of AMD's 90nm parts is still
LOWER than that of their 130nm parts. I figured that this would
change fairly quickly as they ramped up production, but thus far it
hasn't.
But strangely, IBM seems to think that the Xeon is a better
cost/performance part than Opteron.
FT.com / Industries / IT - Companies get bit between their teeth
http://news.ft.com/cms/s/79cb46d0-575d-11d9-a8db-00000e2511c8.html
I thought that Opterons cost less than Xeons?
Remember that the price you or I would pay for the chip might not have
any relation to what IBM or Dell pays for them.
FWIW I just configured out some sample servers to compare the
near-identical Xeon-based e336 and Opteron-based e326. I configured
out the systems as a bare-bones setup with 1GB of RAM and either dual
3.6GHz Xeons or dual Opteron 250 (2.4GHz) chips and no hard drive or
extra hardware. The result was that the dual Xeons server cost $4478
and the dual Opteron server cost $4418, not exactly a significant
difference. Of course, when you add a hard drive in the Xeon system
becomes more expensive, but that is because it uses up to 4
hot-swapable drives of the laptop-style 2.5" form factor while the
Opteron uses only 2 drives in a more standard 3.5" form factor.
Anyway, long story short, the price looks to be pretty much the same.
Performance will vary from one application to the next, but they are
likely to be fairly close. As mentioned above though, I think that
the dual-core Opteron could throw a real monkey-wrench into this
argument, especially given that dual-core Xeons might end up having to
clock WELL down in order to high a viable power consumption number to
fit into a 2-socket 1U system.
