Benchmark Update

The new SPEC95 benchmark suites have replaced the old standby SPECint92 and SPECfp92 benchmarks, but not without a minor parting controversy. Critics contend that the SPECmark 92 suite had become obsolete and was too easily influenced by heavily optimized compilers (see "Bringing Benchmarks Up to SPEC," March BYTE).

In January, Intel said a bug in its compiler used to generate SPEC benchmark results led to an overstatement--15 to 18 percent--of SPECint92 results for Pentium processors running at 100 MHz or fas ter. The error did not affect performance results for SPECfp92 or SPEC95 results, the company says.

At about the same time that Intel clarified its Pentium Pro performance, BYTE learned it understated the performance of Sun's UltraSparc I. We erred in a January review of Sun's 167-MHz Ultr aSparc-based Ultra I workstation. Due to the use of a beta compiler in generating the BYTEmark tests, performance in the Integer and especially the FPU test suite was understated.

Keeping in mind that benchmark results provide only ballpark estimates of actual performance when running applications, here are the latest SPEC95 benchmark numbers of coming high-end processors, plus the restated UltraSparc BYTEmark numbers from Sun. BYTE will retest the UltraSparc system and post updated numbers on the BYTE Web site ( http://www.byte.com ).

---

New BYTEmarks Show Stronger UltraSparc FPU

illustration_link (6 Kbytes)

---

The Latest SPEC Estimates

illustration_link (9 Kbytes)