home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.unix.sysv386
- Path: sparky!uunet!cs.utexas.edu!chinacat!chip
- From: chip@chinacat.unicom.com (Chip Rosenthal)
- Subject: Re: 60nSec memory - Will 486/33 be faster?
- Organization: Unicom Systems Development, Austin, TX
- Date: Thu, 20 Aug 1992 06:13:22 GMT
- Message-ID: <1992Aug20.061322.15796@chinacat.unicom.com>
- References: <Bt91F2.2KJ@encore.com>
- Lines: 56
-
- In article <Bt91F2.2KJ@encore.com> mpalmer@encore.com (Mike Palmer) writes:
- >Will a 486/33 be faster with 60nSec memory over 70 or 80nSec memory, assuming
- >the data required is not in cache memory.
-
- When the system needs to fetch information from main memory, it sends
- out the memory address, tells the memory that it needs to read, and
- at some time later it snarfs up the information provided by the memory.
-
- That `some time later' is the interval in which the memory must respond.
- Simply throwing in different RAM chips does not change that time.
- You must assure that the RAM is fast enough to respond by that time.
- If you buy RAM which is faster and responds before that time, you have
- bought nothing.
-
- >How does the CPU wait state tie into this?
-
- This is how you define how long `some time later' is. The more wait
- states you insert, the more time you allow the memory to respond.
-
- [warning - lots of simplification and handwaving follows]
-
- A wait state is achieved by asserting the READY/ pin on the processor.
- When asserted, the CPU simply idles along waiting for the signal to
- go inactive. This signal is used to tell the CPU to `wait up a moment'
- to give slower memory a chance to respond.
-
- When the system is configured for two wait states, the time for memory
- access is stretched by two bus cycles. That is, with zero wait states,
- `some time later' is equal to one bus cycle. With two wait states it
- becomes three bus cycles.
-
- I don't recall offhand whether a 486 bus cycle corresponds to one or
- two clock cycles -- I believe it is one. This means that with a 33MHz
- clock, the bus cycle would be about 30ns. A 486/33 system typically
- runs at two wait states. Two wait states would stretch the access
- time from 30ns to 90ns.
-
- If you cut this down to one wait state you get 60ns. This means that
- you probably cannot run the machine reliably at one wait state unless
- you get 50ns or better DRAM. Good luck! I'd suggest sticking with
- 70ns DRAM and two wait states.
-
- Note that a good motherboard design can use banked interleaving, page
- mode access, and caching to mitigate the delays due to wait states.
- These things don't reduce the amount of wait delays, but instead work
- to reduce the number of times you invoke the penalty of wait state
- delays. Unfortunately, this is the stuff which is never documented,
- and is one of the best reasons to go with some of the name brand
- motherboards (e.g. Micronics) which seem to do some serious engineering
- in their memory subsystems. From what I hear, the folks at Dell also
- do a good job at peecee memory architectures.
-
- --
- Chip Rosenthal 512-482-8260 | The secret to a long life is
- Unicom Systems Development | knowing when it's time to go.
- <chip@chinacat.Unicom.COM> | - Michelle Shocked
-