Price is the key selling point for most PC buyers, but for the speed-hungry, only computers with the fastest microprocessor will do. Processor performance is measured using benchmarks, of which there are many, used to gauge how a chip will perform when running operating systems, games or other programs. Chip giant Intel, which settled a class-action lawsuit last year after posting 'defective' benchmarks, is trying to push the message that newer classes of processors need to be measured differently. A benchmark frequently used as the basis for computer magazine reviews is Winstone. It is an application benchmark, meaning the chip's performance is tested while the PC is running a common software application such as Microsoft Word or Excel. Some others, such as the CPUmark and Intel's own Media Benchmark, are synthetic benchmarks that measure performance using simulated application program functions. Application benchmarks provide a more accurate measurement, but synthetic benchmarks are easier to run and can be distributed to testers on floppy disks. This made it attractive for home computer users, said Jim Nucci, program manager for Intel's microprocessor marketing and business planning unit. Chip benchmarks measure the performance on three vectors - integers (used in word processing), multimedia (for sound, video and photos) and floating point (for graphics and 3-D images). Some benchmarks are optimised to measure specific tasks, such as mechanical computer-aided design or graphics. Others are used specifically for server chips. Intel has devised an iCOMP index that it uses to compare the performance of its own various chip families, including its low-cost Celeron and top-range Pentium II processors for PCs. The index is compiled using a mix of synthetic and application performance benchmarks - Intel Multimedia, NortonSI32, SPECint95 and SPECfp95. The iCOMP allows consumers to see the difference in performance between various lines of chips. A 300 MHz Celeron, for instance, has a lower performance than a 300 MHz Pentium II. Mr Nucci warned that benchmarks used by Intel rivals 'compared apples to oranges', although he did not name any companies. Fellow chipmakers Advanced Micro Devices and National Semiconductor division Cyrix use the Winstone benchmark frequently when comparing processors with Intel's. They also are using function-specific benchmarks increasing to measure new features, such as 3-D. Lily Wu, semiconductor analyst for Salomon Smith Barney, said she held no opinions on any benchmark. 'I just assume or rely on [industry] magazines to pick reasonable benchmarks' when reviewing chips, she said. Most publications used up to 10 benchmarks to compile a composite score on a chip's performance, she said. Intel has been holding worldwide benchmark tutorials for reporters over the past year. Its product literature commonly compares the performance difference only between Intel-brand chips while its rivals compare their products with Intel's. Coincidentally, Intel last year admitted that it had overstated by about 10 per cent the performance of its 120 MHz and 133 MHz Pentium chips. It quoted results from the SPECint92 benchmark, which the company later described as 'defective'. A class-action lawsuit was filed on behalf of Intel customers by a United States lawyer. Intel settled the lawsuit, offering to pay a US$50 rebate to customers for a chip upgrade, reportedly paying the lawyer US$1.5 million. Analyst group Dataquest estimated about 1.2 million computers sold worldwide contained the chips. Mr Nucci said Intel had quoted benchmark figures from a unbiased third party but declined to comment further.
