SPEC CPU92 Benchmarks

Status

Note: The CPU92 benchmarks have been obsoleted by the CPU95 benchmarks. See the article entitled "Example of Benchmark Obsolescence" for some of the reasons why...

This material for the CPU92 benchmarks is provided for historical purposes only.

Results

Because SPEC's processes were paper-based and not electronic back when CPU92 was the current benchmark, SPEC does not have any electronic storage of CPU92 benchmark results.

The people at the Performance Database Server have made the effort to bring online some of the CPU92 results.

Changes

CPU92 did evolve over its 3 year lifespan (see CPU95 for information about its successor). Changes were necessary to keep up with rising expectations and to keep pace with technical developments.

New Baseline Rules

SPEC has implemented baseline reporting rules. These rules are an attempt to limit vendors from applying all sorts of esoteric compiler technology to what is supposed to be a generally applicable benchmark.

072.sc and Curses

As performance increased, one of the 6 CINT92 benchmarks, 072.sc (based on the public domain spreadsheet calculator) proved to be very dependant upon ill-defined behavior in a set of library routines: curses(3). In September, 1994 SPEC moved to define and sanction a "common curses package" to address this problem.

For more information, read either a hindsight commentary, or the official rule-change notice.

Rules

For the dedicated reader, the original SPEC Run and Reporting Rules are available here. The Run Rules define what is required for a test to be run in a compliant manner. The Reporting Rules define what is necessary to disclose when reporting a result.

• CINT92 Run Rules (~10K text).
• CINT92 Reporting Rules (~15K text).
• CFP92 Run Rules (~10K text).
• CFP92 Reporting Rules (~15K text).

Benchmark Descriptions

Information about each of the benchmarks in the SPEC Suites.

• CINT92 Rel. 1.1
• CFP92 Rel. 1.1

SPECrates

A description of the multi-processor metric from these SPEC Suites.