Intel iCOMP revision draws fire

Intel will shortly take the wraps off an updated version of its controversial iCOMP CPU benchmark for comparing performance of its own chips. However, according to US reports, the processor behemoth could be running into another storm by making the ratings exclusively based on 32bit software code.

While the original iCOMP released in 1992 was largely ignored by the PC industry because it didn't reflect performance of non-Intel chips, iCOMP Index 2.0 could cause a stir by ignoring what is going on in the real world, say critics. As well as concentrating on 32bit performance, the benchmark measures multimedia performance based on Intel's own compilers that emphasise Pentium Pro performance, they claim.

"Windows 95 still has a bunch of 16-bit code," said Linley Gwennap, editor of newsletter The Microprocessor Report, told US trade newspaper PC Week. "Even if you're running 32-bit apps on Windows 95, [iCOMP 2.0] overstates the performance."

Richard Baker, regional marketing manager for PC products at Intel rival AMD, said iCOMP was "spurious" and "completely synthetic". "The thing with iCOMP is it's impossible to reproduce it," said Baker. "You can't compare performance like with real benchmarks where you can run software on a Compaq and an IBM and look at the difference. They aim it at the sales guy in Dixons to say how much faster a 166MHz Pentium is than a 133MHz."