Sun benchmark results pose big questions

That follows a huge argy-bargy on Usenet about the individual SPECfp2k component scores Sun posted last week, and which you can find here.

At first techies showed surprise that Sun had tweaked its compiler to such a high extent, but now there are suspicions that the new compiler it is using might compromise 179.art - a program which runs around fifteen times faster than the geometric mean of the rest.

Sun so far has not explained this anomaly, but it is a little reminiscent of a problem Intel encountered with its compilers in the mid 1990s and called the eqntott bug.

This bug caused its SPECint92 scores to show around a 10 per cent boost before people realised a bug caused the difficulties.

Of course, it is just possible that Sun has come up with technology which is really a shock to those who observe these things closely.

Said Shannon Knows Compaq publisher and IT consultant Terry Shannon: "This is deja vu all over again. If the data about Sun's latest voodoo benchmarketeering is accurate, what we have here is a "Sun, er, Sun of 023.eqntott" problem."

The INQUIRER is wise enough to recall that in a previous incarnation Intel introduced the Pentium Pro as the "fastest processor on the planet".

But when the benchmarketing erratum-not-bug-not-flaw-not-defect-not-sighting was spotted, Chipzilla had to slowly munch its words.

We are attempting to get Sun's comment on the unexpected results but as it's Sunday today, this is harder than normal. µ