The article has a few good bits about what SMP is and how an application can take advantage of it, but the benchmarks are not really relevant for people considering SMP systems.

Noone in their right mind would shell out the cash for an SMP system if all they where going to use it for was running normal desktop apps like office suites, games, mp3 players etc.

The reasons you buy SMP boxes are that you either know that your application (like 3D Studio Max, databases etc) can take advantage of multiple CPUs (multithreading) or that you need to run multiple applications at once or multiple instances of the same program (think web-servers, application-servers etc)...

SMP systems are a waste of money for most ordinary desktop users, but that's irrelevant since that's not the users they are made for. SMP boxes can provide huge performance boosts with the right set of applications - typically server related workloads, 3D rendering, compiling huge codebases, CAD/CAM applications, Databases, video/sound encoding etc...

The benchmarks show only that for most ordinary desktop uses an SMP box will be a waste of money. They fail to show some of the strengths of SMP.
I would have liked to see some tests of a multithreaded Linux kernel compile (make -j), some 3DSMax or Maya benchmarks or tests of SQL query speeds on Oracle or similar...

Jesper,

'The benchmarks show only that for most ordinary desktop uses an SMP box will be a waste of money.'

That was exactly what I was trying to get across and educate those that indeed think that an SMP system will automagically speed up even their MP3 players and what not.

Regards,

While I agree with your assertion that for most desktop app users will not see huge benefits from SMP, and I furthermore agree that SMP is not worthwhile to these users on a price/performance basis, I think that your Xeon results are a bit suspect. I would like to have the performance of a single Xeon system. It seems as if the bechmarks you chose benefitted more from the Xeon's larger cache size than SMP.

-- Odb