Wasn't that the shootout where a pre-release IBM Xeon system with the X3
chipset beat out a production HP Opteron system in the TPC benchmark? It
seemed like for awhile that IBM had done something extraordinary with
the X3 chipset, but later it was found that the real difference in the
performance was HP used a Microsoft SQLServer database, while IBM used a
DB2 database. HP turned it around on IBM by redoing the test using DB2
as well on the Opteron system and ended up beating the IBM system with
its own software product.
That's been an on-going back and forth battle between the two. Right
now IBM's Xeon-based system has top rung, though HP has a speed grade
of dual-core Opteron chips in hand.
Current top-dog is the IBM xSeries 460 using 4 dual-core Xeon 7060
processors (Paxville 3.0GHz, 2x2MB L2, 667MT/s bus speed), with a
score of 273,520 tpmC and a cost of $1,273,691 USD. Executive summary
here:
http://www.tpc.org/results/individual_results/IBM/ibm.x460-DB2.4P.c5.6.030706.es.pdf
HP's best offering is a Proliant DL585-G1 using 4 dual-core Opteron
880 processors (2.4GHz, 2x1MB L2, PC2700 memory bus), with a score of
236,054 tpmC and a cost of $476,378. Executive summary for this one
is here:
http://www.tpc.org/results/individual_results/HP/HP_DL585_4P_2.4DC_DB2_ES.PDF
A few points of note here:
1. The cost of the IBM system is MUCH higher, due almost entirely to
the disk array being used. This disk array plays a HUGELY important
role in determining performance in tpmC. It's rather unclear whether
or not HP could have gained more performance using a higher-end/more
expensive disk array, though presumably they wouldn't gain much. HP
certainly hasn't shied away from using extremely expensive arrays for
their tpmC scores in the past.
2. As mentioned above, HP has a processor speed grade in hand. AMD
now sells the Opteron 885 processor and HP has validated it for the
Proliant DL585 server, ie it's a drop-in replacement.
3. IBM has a bit of extra speed on it's way too. Intel has a new Xeon
7041 processor on the way which bumps the bus speed up to 800MT/s.
IBM does not currently list support for these chips, but that may be
simply because they aren't particularly available.
4. The HP system is available now (actually as of the 5th of Dec.,
2005) while anyone wanting this IBM setup has to wait just over a week
(available May 1st, 2006).
5. These systems are a little ways behind the fastest Itanium
4-processor (8-core) system which managed a score of 290,644 tpmC at a
cost of $788,155 USD. On quick glance both Opteron and Xeon appear to
be faster than IBM's Power5 based systems, but that's due to
differences in naming convention for number of processors. The
Opteron and Xeon systems are both 4-processor/8-core systems, which
are sold as "4 CPU" systems. IBM, on the other hand, used to refer to
their 4-processor/8-core systems as "8 CPU" systems. They have since
changed this to the same naming convention as AMD and Intel (and
nearly everyone else), and as soon as they submit some new 4P/8-core
results for their new Power5 systems they should be WELL ahead of
everyone else. Based on their new 8P/16-core results, they'll
probably manage a score in the 500,000 tpmC range for the smaller 4P
system.
6. Perhaps most importantly, I think a lot of people have come to
realize that tpmC is just as limited as any other benchmark and
becoming more dated then most. The results in this test do not
necessarily reflect any sort of real-world performance.
