Ten, <Believe me, Ali, your assumption is plain wrong.>
You listen. I do not do assumptions. I speak from
facts and CORRECT PLANNING OF DATA FLOWS in
current PC architectures. What have escaped
your attention is that before some agent (PCI
of AGP or NIC) wants to access memory, the data
must be prepared (usually by CPU). You are effectively
saying that some uneducated idiot gets data from
memory, flushes processed data back, and let PCI/AGP
to get them again, or the traffic across the memory
bus is TRIPLED. There are many idiots, so what?
In this case your chipset architecture does not
matter until you get to better memory, and K7 or
P-III will have not much of a difference, you're right.
What I am saying is that with more correct
data stream allocation, the doubled CPU bandwidth
of K7 system will help in correctly written
application, while the Pentium system, with matched
bus-to-memory bandwidth, will not. (I guess we will
not fall into details of how to avoid the cache
coherency overhead and some other tiny details..)
<EOD> (End Of Discussion). - Ali
|
