home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!ukma!darwin.sura.net!spool.mu.edu!olivea!inews.Intel.COM!mipos2!rskinner
- From: rskinner@mipos2.intel.com (Rod Skinner)
- Newsgroups: comp.benchmarks
- Subject: Re: specint, specfloat, mips...
- Message-ID: <C1Ky2L.Dw0@inews.Intel.COM>
- Date: 28 Jan 93 19:56:44 GMT
- References: <1993Jan27.162837.23823@cenaath.cena.dgac.fr> <lmdjtrINNsc7@exodus.Eng.Sun.COM>
- Sender: news@inews.Intel.COM (USENET News System)
- Organization: Intel Corporation, Santa Clara, CA USA
- Lines: 53
- Nntp-Posting-Host: mipos2
-
- In article <lmdjtrINNsc7@exodus.Eng.Sun.COM> walter@zarquon.Eng.Sun.COM writes:
- >In article 23823@cenaath.cena.dgac.fr, brasseur@stna.dgac.fr (Christophe Le Brasseur CAP SESA TERTIAIRE AG25) writes:
- >>Here are a few questions :
- >>==========================
- >>
- >>What exactly are specint and specfloat ?
- >>How can you compare spec to mips ?
- >
- >I am emailing you the SPEC FAQ for this group, which may help you.
- >
- >>We already made simulation of the system on sparc 10/30 and on Bull DPX2/360 >Bi-processor (a french computer with two 25Mhz 68040).
- >
- >This is the best answer. Nothing represents your own code better than
- >your own code. You can use SPECint92 and SPECfp92 to get an idea of
- >CPU speed to decide which machines to test with your code.
- >
- >>We looked at Sparc 10/52 and at RS6000. ...
- >>We can also get a DPX2/380 with 4 processors (33 Mhz 68040) : 100 Mips (!?)
- >
- >Your description of the workload sounds multi-user and multi-threaded,
- >so you can use SPECrate_int92 and SPECrate_fp92 to rate overall system
- >compute capacity. (See FAQ for the difference between the speed and
- >capacity measures.)
- >
- >>So the point is : does my appplication use a lot of specints or a lot of specfloat ?
- >
- For system applications that depend on Oracle as you have
- indicated...don't waste time on SPECfp92 or SPECrate_fp92. The Oracle
- application code is 99+% integer code.
-
- The set of results that you should also consider is the set of results
- published by the TPC and/or the SPEC SDM benchmark results. These may
- more closely track performance of your application. I suspect that they
- will track more closely than any of the individual CPU benchmarks.
-
- >After running your code, the next best answer is to look at the
- >descriptions of the individual SPEC benchmarks and match them with your
- >application. Then look at the SPECrates of those benchmarks most like
- >your application. Finally you can make your own measurements or seek
- >advice on how much integer code is executed versus floating point code,
- >but also consider the source of the advice. (I won't offer any here :-)
- >
- >---
- >Speaking for myself, not for Sun nor for SPEC...
- >Walter Bays walter.bays@eng.sun.com
- >Sun Microsystems, 2550 Garcia Ave., MTV15-404, Mountain View, CA 94043
- >(415) 336-3689 SPEC Steering Committee Chairman FAX (415) 968-4873
-
-
- Rod Skinner I speak for myself only.
- Intel Corp
- 2200 Mission College Blvd RN4-21 PH (408) 765-4474
- Santa Clara, CA 95052 FAX (408) 765-4920
-