HEWLETT PACKARD'S quick spec page
for its nForce2-based d325 business class PC shows an AMD Athlon XP
3200+ processor that supports a 333 FSB (166 MHz DDR). To compensate
for this slower bus, instead of the 400 version that normally
accompanies this model, the frequency has been increased to 2.33 GHz,
which would make it the highest frequency Athlon product that AMD has
shipped to date.
This processor is not currently available as a
configurable option. So is it for real? Will HP make it available at
some future date? This could of course be a typo, but the frequency
nicely matches a multiplier setting of 14 (166.666 x 14 = 2333). For a
400 FSB, the multiplier setting would be an inconvenient 11.67 (2333 \
200 = 11.67). Multiplier settings normally end as whole or half
numbers, so this adds credibility that this part could be for real.
Also, for logistics and support reasons it would be far easier for HP
to ship all of its d325 business PCs with DDR333 memory, as HP's quick
spec page shows, instead of having two or more memory types to worry
about - that is, DDR266 and DDR400.
Will AMD launch an Athlon MP 3000+?
AMD's flagship Athlon MP
2800+ dual processor ships with a 266 FSB and a frequency of 2.13 GHz.
Using this FSB and a multiplier of 17.5, it would be possible for AMD
to deliver an Athlon MP 3000+ at 2.33 GHz, the same frequency as the
touted HP device, which would give it a 9.4% faster clock speed than
the current top-of-the-line offering.
The maximum thermal power that has been specified for an Athlon MP
processor is 66 W. Today's top model only generates 60 W. So it looks
like AMD has sufficient thermal room to make a 3000+ model doable.
Why waste time on Athlon MP?
People may be wondering why AMD
would expend precious engineering resources on a product that for all
intents and purposes could be considered as dead-end. There are several
reasons why AMD might do this.
An Athlon MP with a 3000+ model rating would give the platform some much
needed exposure. It would also nicely round out AMD's current Athlon MP
offerings.
If one was able to look at Athlon MP's financial performance in
isolation, it would probably be one of AMD's few product lines that
currently returns a profit, which is a bottom line reason why an Athlon
MP 3000+ would make good economic sense. Also, AMD could sell this
processor for $100 cheaper than the 3.06 GHz Xeon and still make a very
tidy sum.
An Athlon MP 3000+ with a 9.4% faster clock rate than the current
flagship would deliver very respectable performance. GamePC has
published a dual workstation group test,
which included Opteron, Xeon, and Athlon MP. It is interesting to note
that out of the seven workstation benchmarks used - excluding SiSoft
Sandra - the Athlon MP 2800+ platform was able to better the Xeon 2.8
GHz system in five of the eight results.
Another point to
consider is current Opteron pricing. It might just be the case that IT
decision makers are very impressed with Opteron's performance but just
can't stomach its cost, especially when they take into account Intel's
current fire sale pricing for Xeon.
Ignoring Intel's top-of-the-line 1 MB Xeon, every other up to two-way
Xeon is from $20 to $340 cheaper than corresponding Opterons. So buyers
looking for a cheaper AMD alternative to Opteron might consider the
Athlon MP platform, especially if a 3000+ processor fronts it.
Why the Athlon MP platform
is still important to AMD
The
enterprise community tends to be conservative when it comes to buying
IT infrastructure, which means they generally prefer to buy technology
that has been tried and thoroughly tested. The AMD Athlon MP platform
has been with us for over two years and has gained a reputation for stability, performance, and value.
On the other hand, the HPC community has shown that it does not always
follow this conservatism, which has been manifested by a string of AMD
Opteron design wins.
Cost is another factor for those who prefer to buy AMD. It should not
have escaped those who follow this industry that dual Opteron server
motherboards begin at a price point of around $290. Athlon MP server
boards can be purchased for around $100 less. The Athlon MP 2000+ is
half the price of the cheapest Opteron processor. If one's budget is
really tight the 1.2 GHz Athlon MP can still be purchased for around
$50. The price disparity is clear for those who buy this technology in
volume. So the Athlon MP or even the Athlon XP platform will still be
the preferred choice for certain applications.
A recent cluster purchase by the University of Kentucky chose the Athlon XP platform instead of the Opteron one because, it claimed, Athlon delivered better price/performance for the optimized code that it runs:
"Why not use Opterons? Well, we are very impressed by Opterons
(especially the new memory pipeline), but being impressed doesn't mean
Opterons yield the best price/performance. An Opteron and an Athlon
have essentially the same core arithmetic performance at the same clock
rate, and Athlons still have a faster clock and a lower price. Running
code that is not well-tuned for cache usage, the Opteron would clearly
win... but our codes are very carefully tuned to get near peak
performance, which includes very careful cache tuning. We expect the
Athlon 64 processors to have much more favorable price/performance for
use in clusters."
A 333 FSB?
Further increases in frequency can still clearly
help the Athlon MP platform. But adopting a 333 FSB would really give
it a serious pedal to the metal performance boost that would also
extend its life. If that decision was made, AMD would have to tweak the
northbridge and re-qualify the platform. I don't expect that to happen.
What I would expect AMD to do is to launch a new entry level dual
platform using its 754-pin Athlon 64, which would leave Athlon MP to
continue as the bottom dollar platform.
Athlon MP on 90 nm?
AMD has said that it will likely move Athlon XP onto 90 nm,
which would make Athlon MP devices doable on that process as well. This
would give AMD the option to deliver lower power devices for blade use,
high performance solutions beyond the 3000+ model rating, and possibly
a version with 1MB of level two cache. µ