home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.sys.ibm.pc.hardware:22439 comp.sys.intel:1556
- Newsgroups: comp.sys.ibm.pc.hardware,comp.sys.intel
- Path: sparky!uunet!psinntp!tandon!tdbear
- From: tdbear@tandon.com (Tom Barrett)
- Subject: Re: Help - Cyrix processors, anyone know for sure?
- Message-ID: <1992Aug21.170154.23076@tandon.com>
- Keywords: Intel, P5, Cyrix, 486DLC, i386, Gomez
- Organization: Tandon Corporation, Moorpark, CA
- References: <1992Aug13.215958.4016@bcars64a.bnr.ca> <Bt91A8.212@nntp-sc.Intel.COM>
- Date: Fri, 21 Aug 1992 17:01:54 GMT
- Lines: 51
-
- In article <Bt91A8.212@nntp-sc.Intel.COM> cpurkis@gomez.intel.com (Clif Purkiser) writes:
- >I think you would have cache coherency
- >problems between the internal cache and external cache, because
- >the cache controller on the i386 motherboard wouldn't be aware
- >of the internal cache.
-
- [note who that quote came from... particularly the server
- address]
-
- The external cache controller doesn't need to be aware of the
- Cyrix internal cache because they did something smart... as
- long as the 386 socket has a real HOLD signal going to it, the
- Cyrix can be programmed to flush it's internal write-through
- cache on each DMA. Furthermore, they provided a special
- non-cacheability setting for the 1st 64K of each 1M to
- eliminate the A20 problem. And, still furthermore, they did
- something extra smart by providing an internal cache-ability
- map (something Intel should have done, especially with the
- P5... it is a total pain to provide KEN support which is cheap
- and can easily be programmed by BIOS). Now, if they would
- have only provided WriteProtect mapping, I would have been
- totally happy!
-
- With the external cache, the 486DLC operates up there with the
- AMD/Intel 486 (just wishful thinking on the AMD :), and the
- performance hit from the DMA flush isn't noticable unless you
- have a alots of DMA activity (ie. Floppy, DMA LAN card, DMA
- SCSI, DMA video, etc.) or if your cache controller doesn't
- hide the refresh (even without hidden refresh it is still
- pretty darn swift). And, with external cache I have seen no
- real performance hit using the Cyrix A20 feature.
-
- NOW, if the motherboard is Cyrix ready then chances are you
- won't see any problems with refresh and no low performance due
- to DMA read ops or because of the A20.
-
- >Cyrix performance claims are generally based on modified i386
- >processor motherboard. So without enabling the cache I doubt
- >you'd see a big performance boost.
-
- Right on the first part and right on the second part, but just
- like a political spin doctor you left out the fact that
- typical unmodified i386 processor m'boards (sans the i386)
- will would tons faster than when the i386 was used!
-
-
- --
- Tom Barrett (TDBear) tdbear@tandon.com voice 805-378-6207
- Tandon Corporation tdbear@p6.f1006.n102.z1.fidonet.org fax 805-529-8895
- Sr. HW Design Engineer "War is Peace, No is Yes, And We're All Free!"
- [The views expressed herein may not be shared by the organization of origin]
-