home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!mcsun!uknet!siesoft!athen!weicker
- From: weicker@sinix.UUCP (Reinhold Weicker)
- Newsgroups: comp.benchmarks
- Subject: Re: SPECmarks and Multiprocessor Machines
- Message-ID: <1992Aug13.072430.3353@sinix.UUCP>
- Date: 13 Aug 92 07:24:30 GMT
- References: <1992Aug10.221306.21186@cs.sandia.gov>
- Sender: Reinhold Weicker
- Followup-To: comp.benchmarks
- Organization: SNI AG Muenchen, STO XS
- Lines: 32
-
- In article <1992Aug10.221306.21186@cs.sandia.gov> wllarso@sandia.gov (William L Larson) writes:
-
- >do SPECmarks (and possibly CPU benchmarks in general) scale with
- >the number of processors in a system?
-
- No, they don't: SPEC speed averages (BTW: SPECmarks are outdated, SPEC now
- recommends to use SPECint92 and SPECfp92, based on the newer benchmark
- suites) are based on the execution time of single-threaded programs.
- It was exactly for this reason that SPEC introduced the "Homogenuous
- Capacity Method": Several copies of the same CPU benchmarks are executed
- in parallel, to show the processing capacity of the system, often a
- multiprocessor system. The respective averages are SPECrate_int92 and
- SPECrate_fp92 (similar earlier average, based on the old benchmark suite:
- SPECthruput). For details, see
- Alexander Carlton: CINT92 and CFP92 Homogenuous Capacity Method
- Offers Fair Measure of Processing Capacity
- SPEC Newsletter vol. 4, no. 2 (June 1992), pp. 1-4
-
- >How sensitive are the SPECmark values to the general hardware
- >architecture? Would most 80486 processors return similar results?
- >(Again, assuming things like clock rate are constant.)
-
- The most important factors influencing SPEC CPU benchmark results are
- hardware architecture (and implementation), clock rate, cache size,
- compiler code optimization. The degree of the different influences
- is varying.
-
- --
- Reinhold P. Weicker, Siemens Nixdorf Information Systems, STM OS 32
- Address: Otto-Hahn-Ring 6, W-8000 Muenchen 83, Germany
- E-Mail: weicker@ztivax.zfe.siemens.de
- Disclaimer: This is just my opinion, not my company's or SPEC's
-