Differences between AMD and Intel CPUs [was: Re: Dragonfly and Hyperthreading....]

EM1897 at aol.com EM1897 at aol.com
Wed Mar 9 15:01:57 PST 2005


In a message dated 3/9/2005 3:11:55 PM Eastern Standard Time, Matthew Dillon <dillon at xxxxxxxxxxxxxxxxxxxx> writes:

>    Well, all I can say EM (not knowing what you real name is since you
>    haven't bothered to tell anyone) that I think you need to go back and
>    actually do some research on equivalent systems, because nobody else
>    is seeing the price differentials you are.  Perhaps the problem is that
>    you seem to have a requirement for PCI-X, which is a very new standard,
>    which is severely limiting your choices.

You usually are up on things,Matt, but you seem to be a few
years behind lately, so maybe you need to come up for
air? Not knowing about the Pentium-M was surprising, 
but now you seem stuck in 2001. I expect that from the
Europeons, but I'm surprised at you. You seem to 
have made a major life's commitment to making a 
faster O/S; do you really propose running these
multicore speed demons with a 1Ghz  bus? What (the
hell) do I need a 3+ Ghz processor for if I'm running
on a bus thats 8X slower? What's the point of Hyper
tunneling if you have such a slow conduit to the outside
world? Where cares how fast the processor is if you can't
pass any data? And PCI-X is not any "newer" than the 
Opteron or any of the other processors we've mentioned 
here, so what exactly are you talking about when you
say its "new"? 

A 1Ghz PIII with a PCI-X bus will outperform a 3.2Ghz
P4 with a regular PCI bus for a network application, 
so yes, it is a basic criteria in any performance
oriented formula. Maybe you use your system to render
3D images, but I use mine for networking.

>    Your comments on the Xeon systems are also incorrect... you have to
>    realize that Xeons are a good fit for the economics that many commercial
>    installations deal with.   Xeon's are very good at handling a certain
>    class of problems, primarily large parallel database applications, and
>    because the commercial software in these markets is a *LOT* more
>    expensive then the actual machine hardware, and usually taylored to
>    have the best price point when running on big honking boxes, companies
>    tend to buy big honking Xeon systems for that sort of thing.

We're not talking about "big honking parallel database
applications", so I don't see your point. I don't think
anyone said that Xeons had no value. I only said that
if you are building a box, there is no point in
considering a Xeon because they cost more for no benefit.
With new P4 chipsets and the lack of a decent MP O/S 
there is no distinction between P4s and Xeons performance
wise. Xeons used to have larger caches and only Xeon 
chipsets had PCI-X busses, but thats no longer 
the case. So from a price/performance perspective for
the vast marjority of applications, Intel
wins by a large margin, which was my original point.

It seems that all anyone here cares about is how fast
you can do a 'make buildworld'. It makes one wonder if
there is any point to the project at all.





More information about the Users mailing list