home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!elroy.jpl.nasa.gov!usc!rpi!bu.edu!dartvax!mars.caps.maine.edu!maine.maine.edu!ree700a
- From: REE700A@MAINE.MAINE.EDU
- Newsgroups: comp.sys.ibm.pc.hardware
- Subject: Re: 60nSec memory - Will 486/33 be faster?
- Message-ID: <92233.100402REE700A@MAINE.MAINE.EDU>
- Date: 20 Aug 92 14:04:02 GMT
- References: <Bt90qD.26p@encore.com>
- Organization: University of Maine System
- Lines: 17
-
-
- The change from 70 or 80 nS RAM to 60 nS RAM is only necessary if :
- a) Your board doesn't properly perform interleaved cache fills
- b) It does, but all four banks of SIMMS are full (16 SIMMS!)
- or
- c) You are truely concerned about to 10-20 nS hit the first time a cache
- miss occurs. I assume your 486/33 has 256K cache (if not - upgrade
- that before all of the main memory). For a modest 90% cache hit rate,
- a 25% reduction in SIMM cycle time corresponds to about a 2.5% memory
- system speed-up on an effective zero wait state system.
-
- Therefore, my recommendation: Fill all four banks (16MB of 1MB simms!)
- Shut off any virtual memory bullshit. Get a 256K cache with less than 30
- nanosecond cycle time. If you are currently running 4MB in 256K simms,
- don't replace them with 4MB of faster 256K simms! Get 8MB of 1MB simms,
- take the temporary wash-out in performance (two way interleave of 60's
- compared to 4-way 80's) and add the other 8MB when you can afford it.
-