home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!pmafire!mica.inel.gov!ux1!news.byu.edu!hamblin.math.byu.edu!sol.ctr.columbia.edu!spool.mu.edu!olivea!inews.Intel.COM!mipos2!jwreilly
- From: jwreilly@mipos2.intel.com (Jeffrey Reilly)
- Newsgroups: comp.benchmarks
- Subject: Re: SPEC : questions, copyright, summary - a database of spec's ?
- Message-ID: <BxtHDv.Gq7@inews.Intel.COM>
- Date: 16 Nov 92 16:20:18 GMT
- References: <1992Nov16.143948.27382@crc.ac.uk>
- Sender: news@inews.Intel.COM (USENET News System)
- Organization: Intel Corporation, Santa Clara, CA USA
- Lines: 70
- Nntp-Posting-Host: mipos2
-
-
- In article <1992Nov16.143948.27382@crc.ac.uk> you write:
- >i would like to complie a list of all (!) SPECmark ratings, both INT & FP,
- >even THRUput, if possible. basically i would like to be able to produce a
- >list that i could post every-so-often when a new spec-mark for a new machine
- >appears. this would be of interest when the dec-alpha marks came out, to
- >give an example. this however, raises a coupla questions...
- >
- >is it legal ? am i infringing SPEC's copyright ?
-
- Well, I am not a lawyer :-) but from what I understand the benchmarks results are
- not copyrightable (the terminology and the reporting format from the SPEC
- Newsletter may be). For example, both companies and magazines frequently publish
- tables of SPEC data (for example, performance brief's from DEC, IBM, Sun and
- magazines such a Microprocessor Report, Andrew Allison's RISC Forum, EE Times.
-
- >
- >how should i report them ? i assume that i would include supplementary
- >info along with each benchmark result, such as cpu (with mhz rating),
- >cache size (?), memory (is it relevant?), compiler used? i assume that
- >it is necessary to chop the results up into specint92, specfp92, and
- >specthruput, (and also spec89marks, if relevant). i would probably also
- >separate the multi-cpu from the single-cpu results.
-
- Typically, SPEC suggests that enough information be reported to allow someone to
- easily reproduce the results. Cache size and memory are important (though
- admittedly if you throw enough hardware at the problem, eventually you reach a
- point where adding more doesn't help). The choice of a compiler makes a big
- difference (as an extreme, look at the difference between the results published
- by SUN, HP, etc. before and after the introduction of the KAP Preprocessor).
-
- The OS may make a difference in SPEC SDM results.
-
- Also, note that configurations may vary between a reported SPECint92, SPECfp92
- and SPECrate_int92, SPECrate_fp92.
-
- Another key point to make is you should make sure you mention the source of the
- information.
-
- All and all, it depends on how much work you want to put into it. It is a lot of
- information that changes and is added to quite rapidly from many sources (SPEC
- Newletter, company hardware and software announcements, etc.)
-
- SPEC is currently planning to include a year end summary of performance for it's
- December Newsletter.
- >
- >i have read the FAQ pertaining to spec-stuff. perhaps i am trying to create
- >a store of spec-marks that could be attached to a FAQ, although this is not
- >my intention. it is intented to be stand-alone (!) from a faq. do you think
- >anyone from spec would be upset. and no, i can't be bothered to go and
- >ask them myself. any comments welcomed (flame-proof suit on...) mainly be
- >postings, email also.
- >
-
- Well, if you ever want to bring anything up with SPEC, we're always willing to
- listen... :-)
-
- Personally, I think its a useful idea and I think most other people would too
- (particularly if it is kept current and correctly).
- >thanks...
-
- Good Luck!
-
- Jeff
- Intel's Representitive To SPEC
- Jeff Reilly | "There is something fascinating about
- Intel Corporation | science. One gets such wholesale returns
- jwreilly@mipos2.intel.com | of conjecture out of such a trifling
- (408) 765 - 5909 | investment of fact" - M. Twain
- All opinions are my own....
-