Pete, <That is because, 4 Xeons have to share each bus and the associated memory, while each Tbird has its own dedicated memory.>
You are attempting a strawman here. In a server cluster, a 4-way Xeon box is considered a single unit. Then several of these 4-way SMP boxes will be clustered together into a system. SMP is only confined within each box. It is NOT meant to be a replacement for clustering.
<When you can get 72 1G Tbirds plus Beowulf Cluster infrastructure for less than a 8 way Xeon SMP server goes for, there will be much less demand for those servers.>
You should tell that to IBM. They decided to use 4-way Xeon boxes in their record-setting TPC-C benchmark:
zdnet.com
DB2 EEE 7.1, running on a cluster of 32 IBM Netfinity servers and on Windows 2000 Advanced Server, showed a sustained throughput of 440,879.95 transactions per minute at a five-year cost of ownership of $32.28 per transaction in audited TPC-C results released July 3.
Wow, 32 4-way Xeon servers. That's an astounding number of servers, but that number is merely there to demonstrate the scalability of 4-way Xeons in cluster configurations. I'll guarantee you that uniprocessor Tbirds will not match the scalability nor the price/performance, no matter how many you use.
But hey, if you're right and I'm wrong, then somebody ought to tell IBM that they could have gotten even better TPC-C results had they gone with uniprocessor Tbird boxes over 4-way Xeon servers. It doesn't work that way.
So don't pretend that AMD doesn't need SMP. They do, and they know it.
Tenchusatsu |
