Performance Briefs
Intel® Pentium®4 and Xeon™ vs. Power Mac G5
Paul Martin (). Updated 6/27/2003

Summary: On June 23, 2003 Apple announced what it claimed to be the world’s fastest personal computers, based on the Power Mac G5. Apple centered their claims on SPEC benchmark results - these claims are very misleading. In fact, systems based on Intel’s current offering of 3.0Ghz Pentium 4 and 3.06Ghz Xeon processors will significantly outperform Apple G5 systems on these benchmarks, despite this deceptive positioning by Apple.

Background: The G5-based systems offer similar features to Pentium4- and Xeon-based systems that have been out for many months. To be fair, the G5 processor does offer 1GHz front side bus with a large amount of bandwidth. However, the memory subsystem cannot supply data at the rate enabled by the FSB.

Results: Apple quoted SPEC CPU2000 benchmarks run by a third party, Veritest, as “evidence” of the G5’s superiority. These numbers, however, were run under Red Hat Linux 9.0 using the GCC 3.3 compiler which produces inferior SPEC numbers for the Pentium 4 and Xeon. The SPECint numbers are about 24% below the officially published numbers on the SPEC website and the SPECfp numbers average greater than 40% below the official numbers. The official numbers were generated under WindowsXP with versions 6.0 and 7.1 of Intel’s commonly-used compilers. The chart below indicates that on the single processor benchmarks, shipping 3.0Ghz Pentium 4 systems outperform the 2.0Ghz G5 system (claimed to ship in August) by almost 50% on SPECint and SPECfp! Xeon-based systems also best the G5 significantly. On the SPECrate benchmarks, which are used to measure multiprocessor throughput, currently shipping DP 3.06 Ghz Xeon systems beat G5 by 25% on SPECint_rate and break even on SPECfp_rate (with Xeon processor upgrades around the corner). Single processor Pentium4 systems are 80-87% as fast as the Dual Processor G5 system. Clearly, Apple’s SPEC-related claims are extremely deceptive.

In addition to the SPEC results, Apple offered several very specific application performance comparisons, none of which are independently verifiable. Upon inspection, these claims are based on new versions of applications specifically tuned for G5, or unfair comparisons (Dual Processor G5 vs Uni-processor Pentium4). Mac OS X still does not support many of the workstation applications that HP customers need, including all of the MCAD and most DCC applications.

Conclusion: Apple has significantly misrepresented Pentium4 and Xeon performance in order to make the G5-based systems appear more compelling. Official, verifiable benchmarks actually indicate that current Pentium4 and Xeon significantly outperform G5, which Apple claims will ship in August 2003.

Links

SPEC CPU Benchmarks: http://www.spec.org/cpu2000/results/