Processor Performance 2006

New article by Joe Chang posted:

Processor Performance 2006
http://www.sql-server-performance.com/jc_processor_performance_06.asp


Shane Tasker
--
Contributing Editor, SQL-Server-Performance.Com
Independent Consultant

the new Montecito core Itanium is launched,
instead of asking shane to reformat the entire 6000 word article, i will post new info here.

HP Integrity rx6600
4 x 1.6GHz Dual Core Itanium 9050, 1x24M L3 cache
192 GB memory
5 QLogic QLA 2342 FC controllers
756 36GB 15K U320 drives
8 300GB 10K U320 for logs
18 MSA 1000 + 36 MSA30 for data
1MSA 1000 for log
344,928 tpm-C

this is better than the expected performance for 4 socket Tulsa processor (320K ?)
and much better than the 4 socket DC Opteron 2.6GHz (214K)
for results on SQL Server 2005 and Windows Server 2003

given that SPEC CPU 2000 integer results to date show only slight gains for Montecito over Madison, the possible explanations for the very Montecito TPC-C result is either the very large 24M L3 cache or hyper-threading
HT was very problematic with the Xeon line.
My tests showed that HT only helped in handling the network round-trip,
no SQL Query showed gain, some negative results.
SQL also had problems with sequential disk IO when using all logical procs.
HT showed big performance gains with the Quest/Imceda LiteSpeed compression engine, which has no contention issues.
if i had to guess, the Itanium team improved HT performance for SQL queries on the Montecito

The first benchmark result (TPC-H) for Tulsa (2x65nm Prescott cores, 16M shared L3 cache) is out

Dell PowerEdge 6800
4 x Dual Core Intel Xeon MP 7140 3.4GHz, 16M L3 $22K ?
64GB $53K
3 PERC/5E SAS PCI-E adapters
270 36GB SAS hard drives

16,320 QphH @ 100GB

this is just moderately better than HP's result for the 7041 (DC 3.0GHz) at 14,242 QphH @ 100GB
and much better than Dell's 7041 result of 11,529 QppH

there is evidence to suggest that the big cache is important for transaction processing applications but not as important for DW apps.