home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!cis.ohio-state.edu!zaphod.mps.ohio-state.edu!usc!sol.ctr.columbia.edu!ira.uka.de!uka!uka!news
- From: S_JUFFA@iravcl.ira.uka.de (|S| Norbert Juffa)
- Newsgroups: comp.sys.ibm.pc.hardware
- Subject: Re: ULSI co-processor chips any good?
- Date: 18 Aug 1992 18:23:46 GMT
- Organization: University of Karlsruhe (FRG) - Informatik Rechnerabt.
- Lines: 48
- Distribution: world
- Message-ID: <16rf7iINNt38@iraul1.ira.uka.de>
- References: <1992Aug15.234529.311@magnus.acs.ohio-state.edu>
- NNTP-Posting-Host: irav1.ira.uka.de
- X-News-Reader: VMS NEWS 1.23
- In-Reply-To: rjones@magnus.acs.ohio-state.edu's message of Sat, 15 Aug 1992 23:45:29 GMT
-
- In <1992Aug15.234529.311@magnus.acs.ohio-state.edu> rjones@magnus.acs.ohio-state.edu writes:
-
- > Anyone have any experience with ULSI co-processor chips. I'm looking at
- > Cyrix, IIT, Intel and ULSI for my 386-33. I seem to recall a magazine
- > test that compared these and the ULSI came out high. It didn't suffer
- > from the accuracy problems of IIT or Cyrix, I believe. However, it
- ^^^^^^^^^^^^^^^^^^^^ ^^^^^
-
- Hold it right there! While doing some research for an article on coprocessors
- which I intend to post tomorrow, I have tested the IIT, ULSI, Intel, and Cyrix
- coprocessors as well as the Intel RapidCAD and the i486. The Cyrix 83D87 and
- 387+ are the coprocessors with the highest degree of compatibility with the
- Intel 387DX. The IIT does not support denormals for the extended precision
- format and fails quite a few tests in the IEEETEST program. The ULSI does not
- support rounding control, it always uses extended precision, and it fails a
- few of the tests in IEEETEST. The Intel and Cyrix chips pass all these test
- without a single failure. However, neither of the 387 'clones' is entirely
- compatible with the Intel 387DX. Cyrix has documented existing differences
- in a "Compatibility Report". Only the Intel 387DX can asynchronously with the
- CPU, meaning that CPU and coprocessor run at a different clock rate. The
- coprocessors from Cyrix, ULSI, and IIT always run at the full speed of the
- CPU even if you jumper your motherboard for asynchronous operation, so make
- sure to get the right speed coprocessor for your CPU. Don't try to 'push' a
- coprocessor above its stated maxmimum operating frequency, it may introduce
- computational errors into the coprocessors results (I have had this happen to
- myself). Of all coprocessors tested, the Cyrix 387+ had the most accurate
- transcendental functions, followed by the Cyrix 83D87, the Intel 486, the
- Intel RapidCAD, the IIT 3C87, and the ULSI 83C87. The maximum error for
- the transcendental function on the ULSI is within the limits specified by
- Intel though. As for performance the Cyrix offers the highest overall
- benchmark performance of all 387 'clones', followed by the IIT, ULSI, and
- Intel 387DX. However, the performance difference in real applications is
- hardly noticeable and well under 10%. Since the incompatiblities found in
- the IIT and ULSI coprocessors are hardly an issue with regard to most or
- even all applications, you might also go for the cheapest coprocessor
- available. Hope this helps!
-
-
- > --
- > < Ron Jones | E-mail: rj+@osu.edu >
- > < The Ohio State University | Office: (614) 292-1665 >
- > < 338 West Tenth Avenue | FAX:(9-5PM) (614) 292-7151 >
- > < Columbus, Ohio 43210 | "eschew obfuscation" >
-
- Norbert
- -------------------------------------------------------------------------------
- Norbert Juffa email: S_JUFFA@IRAVCL.IRA.UKA.DE Live and let live!
-
-