home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!cs.utexas.edu!qt.cs.utexas.edu!yale.edu!ira.uka.de!uka!uka!news
- From: S_JUFFA@iravcl.ira.uka.de (|S| Norbert Juffa)
- Newsgroups: comp.sys.intel
- Subject: Re: Help - Cyrix processors, anyone know for sure?
- Date: 18 Aug 1992 17:14:56 GMT
- Organization: University of Karlsruhe (FRG) - Informatik Rechnerabt.
- Lines: 45
- Distribution: world
- Message-ID: <16rb6gINNt38@iraul1.ira.uka.de>
- References: <15120@suns4.crosfield.co.uk> <15152@suns5.crosfield.co.uk>
- NNTP-Posting-Host: irav1.ira.uka.de
- X-News-Reader: VMS NEWS 1.23
- In-Reply-To: pdg@crosfield.co.uk's message of 17 Aug 92 09:13:10 GMT
-
- In <15152@suns5.crosfield.co.uk> pdg@crosfield.co.uk writes:
- >
- > Well I got a pointer to an answer about these new
- > Cyrix chips, the Cx486SLC and the Cx486DLC
- > (thanks Henrik Vallgren).
- >
- > It seems that the chief performance gain (over a 80386)
- > comes from the cache. So, unless your motherboard has
- > been designed for these chips, there's not much of a gain
- > at all. There is a software fix (for DOS) which causes
- > the cache to be flushed every time some external
- > device DMAs to the motherboard, but that reduces the
- > performance gain.
-
- The cache is definitely the main source of the 486DLC's performance
- gain over the 80386. This can be easily deduced from the performance
- of Intel's RapidCAD chip which is basically a 486 without the cache
- and with a 386 pin out. While the RapidCAD offers roughly twice the
- floating point performance of the Intel 387DX, my benchmarks showed
- it to be at most 25% faster than the 80386. Typical performance gain
- for integer applications is more like 15% though. This definitely
- shows that the 386 bus interface (2 clock cycles per bus cycle, no
- burst mode) is a bottleneck for the RapidCAD that can only be overcome
- with an internal cache (the above mentioned benchmarks were run on a
- machine with a fast 128 kB direct-mapped write-thru cache with one
- write buffer). When executing register-to-register instructions, the
- RapidCAD will execute them faster than they can be delivered thru the
- slow 386 bus interface. When executing instructions that access memory
- the most time is spent in the memory access so one doesn't see a big
- improvement over the 386 either.
-
- The German computer magazine c't has shown that for heavy DMA traffic
- (e.g. playing a WAV file thru the Soundblaster), the 486DLC with the
- cache switched on is slower than with the cache switched off due to
- the frequent cache flushes.
-
-
- > Paul
- > --
-
- Norbert
- -------------------------------------------------------------------------------
- Norbert Juffa email: S_JUFFA@IRAVCL.IRA.UKA.DE Live and let live!
-
-
-