Advanced Micro Devices Inc (AMD)

[alan81]

Racksaver TPC data--
The Racksaver system is configured almost identically to the
HP, Dell, and IBM systems that get very similar TPM-c Results.
If you look at the detailed reports submitted by the vendors
you will notice that racksaver charges much less than Dell or
IBM for both memory and disk subsystems, for virtually
identical hardware.
Note that racksaver does not break out the cost of the
memory, but for $35K you get the 32G of memory plus the 4P
opteron system.

company    TPMC   cost/tpmc  users  memory  mem cost  disk       type      storage cost 
 racksaver  82226  2.72       66k    32G     35K*      294 disks  15K scsi  92K 
 Dell       84595  3.58       67k    32G     27899     296 disks  15K scsi  140000 
 HP         84712  3.83       67k    32G     34396     239 disks  15K scsi  120,000? 
 IBM        90271  3.97       72k    32G     46384     238 disks  15K scsi  182,000</PRE> 
 Racksaver 4P opteron system 82,226: 
 http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=103071002 
 Dell 4P Xeon system 84,595: 
 http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=103063003 
 HP 4P Xeon System 84712: 
 http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=103090802 
 IBM 4P Xeon System 90272: 
 http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=103101701 
 I am not arguing that Xeon scaling in spec is not much worse  
 than Opteron, but in this real world application that values  
 SMP, Xeon does very well. If you look at TPC-H where clusters  
 are used, Opteron does very well...along with clusters of 2P  
 Xeon nodes. 
 I am sure Intel has some plans to improve the pathetic 4P  
 specfp performance of Xeon.  Exactly what is not clear yet,  
 but the 4M L3 cache and faster FSB will help.  I suspect they  
 will also change the cache policy a bit and perhaps adjust  
 prefetching some.  We will know more when the prescott  
 version of these products become available. 
 --Alan