home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.ibm.pc.hardware
- Path: sparky!uunet!math.fu-berlin.de!informatik.tu-muenchen.de!LRZnews!regent!mch
- From: mch@regent.e-technik.tu-muenchen.dbp.de (Michael Hermann)
- Subject: Re: Cyrix 486DLC-33
- Message-ID: <mch.711958782@regent.e-technik.tu-muenchen.de>
- Sender: news@regent.e-technik.tu-muenchen.de (News System)
- Organization: Technical University of Munich, Germany
- References: <1992Jul23.072533.14155@usl.edu> <54893@mentor.cc.purdue.edu>
- Distribution: usa
- Date: Fri, 24 Jul 1992 06:19:42 GMT
- Lines: 98
-
- ericwolf@mentor.cc.purdue.edu (Eric Wolf) writes:
- ...
- >I, too, am very interested in this chip and would also appreciate the info if
- >someone could post it. For example, I would like to know the following (if at
- >all possible)
-
- > (1) what varieties does the Cyrix 486 come in?
- > (2) is it 100% Intel 486 compatable?
- > (3) does it offer any advantages over the Intel 486?
- > (4) How much $$$???
-
-
- As far as I know:
-
- 1) SLC (with 16D/24A bit external bus, pin-compatible to 386sx)
- DLC (like 386 dx (32D/32A))
- To my kwowledge both chips are otherwise identical.
-
- 2) At least from the user-mode software view.
- The reports I saw did not mention any problems with either DOS
- or UNIX. Presumably cache-programming is a bit different, but
- usually you set up those registers only once (startup, BIOS).
-
- 3) Not that I know. In fact, there are disadvantages:
-
- - no builtin FPU (greetings from 486sx)
- - 1K cache only
- - no bus-snooping (invalidate cache-lines on externally generated
- transactions, like DMA)
- - generally slower execution than 486 but faster than 386
-
- 4) latest info (probably wrong by the time of my posting) $119/1000
-
-
- Some remarks to 3)
-
- You can add a generic coprocessor-FPU like 387. If you use the matching
- Cyrix-Part (announced) you should get performance better than any existing
- 386/387-pairing. But (in my opinion) performance will still be worse
- than on a 486DX, because you have to establish a protocol to get both
- chips synchronized. The more complex functions (like trigono) you need,
- the better the Cyrix should perform.
-
- A think, the 1K cache yields less than 50% hit-rate, but as the 386-bus-
- protocol will always "waste" at least 2 cycles, this is definitely an
- improvement in externally cached existing 386-systems.
-
- If you have heavy external DMA-transfers (like most Ethernet or SCSI)
- the Cyrix may be the wrong part for you. As I understand, there is minimal
- support for cache-coherency. At least one (fairly independent) report
- claims, that with DMA activity the Cyrix is actually slower with enabled
- internal cache than with no internal cache at all. The part has some
- lines for cache control, but they will be useless in existing 386-mainboards.
- Even if new designs support those features I (currently) don't think
- that it will help a lot.
- However, if you have a "standard" DOS-PC-environment (like some sort
- of MFM/RLL (IDE is basically emulating those, so with IDE yoe are well, too),
- you will not suffer from this cache-control deficiency.
-
- At last, the core has been developed independently from the Intel-core,
- so the hardware resources are spent differently and this affects
- execution time for instructions.
-
- Examples:
-
- A dedicated multiplier speeds up integer multiplications to a
- constant time (I think 3 cycles) whereas the i486 uses an early-out
- algorithm which is still SERIAL in nature (~10 - ~40 cycles).
-
- The lack of some of the adress-generation hardware imposes an additional
- cycle for adressing modes requiering an add (like disp(reg)).
-
- Different pipeline designs lead to different stalls on break
- conditions like jmp or a slow memory transaction (even within
- the cache there is a difference between c486 and i486 I think).
- Generally c486 looses a bit here.
-
-
- Remarks to 4)
-
- When the c486slc was introduced, I was sceptical about the real
- advantage of beeing pin-compatible to 386sx. Almost all 386sx
- are soldered and a *normal* user could not swap CPUs. And the
- board manufacturer had to (partially) redesign his board anyway
- to make use of the external cache control. So why not change
- the bus-protocol to a faster one, too? If speed of existing
- chip-sets was a problem. you could easily force a wait-state,
- but you had at least the option to make faster systems later on.
-
- Now with the c486dlc I have revised my opinion:
- Most 386 are PGA (AMD makes pqfp, though) and *you* can swap
- the CPUs. You will get a speed improvement of about 50%, I think.
- And this may well be worth the extra $$$ as compared to throwing
- away your complete board when buying an i486.
-
- Michael
- mch@regent.e-technik.tu-muenchen.de
-
-