AGAIN :) Cheap Duali is here and in I875

[Dan@354-Pines]

As an aside,

I've never really understood, apart from being bl00dy obtuse and anoyed that they made a mistake back in the early Pentium days, as to why Intel so adamantly refuses to allow the desktop CPU's to run in MP configs. Really, it makes no sense. It means having to create a low-end Xeon in this case, which further blurs the distinction between server and desktop chips whilst simultaneously eliminating a whole section of that low-end/entry-level market.

In terms of companies like mine ...the ones who really buy Xeon-based servers en mass, they loose nothing. We wouldn't touch a low-end non-certified processor in a server, which is why we never touch AMD. Even in tests, AMD won like for like hands down on price and performance against (the early) dual-Xeon servers, and we went and got dual Xeons purely because of the Intel badge and the fact that they validate their stuff like crazy (according to one of our whitepapers, one Intel test showed the fans in the server case to be so powerful, that in advent of a cpu fire, the fans almost instantly put the fire out).

Small companies in my experience either go with no servers at all, or go to another option (i.e. use a sp desktop solution for a server, or a competitor like AMD) rather than spend big bucks on server chips and chipsets. Were the cross-over to be transparent between desktop and dual processing they'd open up a much bigger transitional market IMO. That only leaves content creators, who will pay premium's for server-calibre products if warranted (i.e. if price equals performance) and data centres who will also shirk the server cpu's where they can, instead opting for options like huge arrays of clustered small servers with lots of redundancy built-in that way.

We've used that approach with 2U dual Xeon-based rack servers, and also with larger servers (e.g. 8-way 2MB cache with Cache Coherency filters Xeons) and to be honest, the business is swaying toward the former from the latter based on cost and performance and pretty much any other argument you can find. With O/S's and applications increasingly supporting clustering better and better, this is now a real possibility. The funny thing is, in one project we spent more on this approach than a previous comparable project, because the Dual PIII based 2U option at the time, came with the Tualatin (512Kb cache) cpu's which allowed us to develop a much better solution around multiple servers, than the equivilent solution using a single four-way server set-up.

The signs are there that Intel's maybe feeling a bit more threatened in it's core server market than it ever has before, what with the 1Mb L3 Xeons and the constant Dothan rumours, but honestly, I don't see Intel really having to worry due to the reasons outlined above.

These new Opteron's do look expensive (compared to previous AMD offerings), but lets hope this is indicative of quality and performance ...and of getting proper validation to make them at least worth considering in the corporate market.