% fortune -ae paul murphy

Barcelona, T2, and Tigerton

Intel claims that its forthcoming Xeon 73XX CPUs are the fastest, bestest, cheapest, and most power efficient processors around, AMD claims the same crown thing for its "Barcelona" line, and Sun says that its T2 ultraSPARC blows both of them away.

So which two are lying?

My answer, on brief review at least, is that all three are right - which sounds absurd, but my argument is that it all depends on the software you assume is going to run on the thing.

The bottom line is simple: it's always all about the software. Remember when the PC people said the PPC Mac was slow because it ran crudely recompiled PC software slowly? while the Mac people pointed out that even cursory efforts to use the PPC as intended showed it to be significantly faster than the PC? That's what's happening here: pick the right software for any of these platforms and it beats the other two.

Notice, however, that software isn't the core issue on either price or watts used: for those numbers you need to remember first that using "tigerton" requires you to give Intel another $266 for the "Caneland" chipset without which you're not going anywhere and, secondly, that you also have to pay the power company to run the thing once you've got it - meaning that people who declare for Intel strictly on the basis of processor cost and per core power use are just lying to you.

What's critical for performance, however, is software so if you about software generations and you can see how Intel's pretence at multi-core through shared packaging is a perfect generational match for the pretend multi-threading built into most Wintel server software - and therefore why the combination of Intel's manufacturing advantage (i.e. its gigahertz leadership) with on the fly cache share realignment makes the new Xeon an easy performance winner when running mutually independent processes.

Unfortunately for AMD the software buzz these days is all about virtualization in the sense of having one computer run multiple guest OSes (ghosts), each responsible for its own application - and since that's about as 90s as you can get with Windows, Intel's multicore packaging technology is likely to be sufficiently better at this to overcome AMD's better hypervisor and multi-core memory access technologies.

Basically, Novell technology manager Lowry Snow is quite right in saying that AMD's technologies signal its rejection of mediocrity, but Intel is equally right in believing that Microsoft's software virtually requires it.

And where's the UltraSPARC T2 in all this? Ahead of AMD on technology, but irrelavant to any discussion of x86 software. Remember: CMT's magic comes mainly from concurrent muilti-threading, not from having multiple cores - meaning that its value for general purpose processing comes from throughput, not single threaded process completion. In other words, comparing Intel's Tigerton/Caneland combination to the T2 is like comparing an Arabian race horse to a team of Jutlands: the racer wins on speed with one rider but the Jutlands (or Clydesdales in the U.S.) bring home the beer.

Paul Murphy wrote and published The Unix Guide to Defenestration. Murphy is a 25-year veteran of the I.T. consulting industry, specializing in Unix and Unix-related management issues.