Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!chilli.Berkeley.EDU!elm From: elm@chilli.Berkeley.EDU (ethan miller) Newsgroups: comp.arch Subject: Re: How Caches Work Message-ID: <31224@ucbvax.BERKELEY.EDU> Date: 11 Sep 89 08:06:00 GMT References: <21936@cup.portal.com> <1082@cernvax.UUCP> <16306@watdragon.waterloo.edu> Sender: usenet@ucbvax.BERKELEY.EDU Reply-To: elm@chilli.Berkeley.EDU (ethan miller) Organization: Berkeley--Riots 'R Us Lines: 20 In article rang@cs.wisc.edu (Anton Rang) writes: % The problem is that caching something is not without cost. Loading %128 words into the cache from main memory is generally just as %expensive as fetching 128 words from main memory directly. The cache %only helps if you're doing multiple fetches of the same memory %location before it's overwritten with other data. It is often faster to load 128 words sequentially than it is to have to fetch those words from memory. If you can use static column mode on your RAMs, for example, you can get words from memory about twice as fast as normal mode. Unless you have smart hardware that knows successive accesses refer to the same bank and treats the accesses as static column, it'll be faster to read 128 words at once. However, the cache overhead might cut into that advantage. ethan ================================= ethan miller--cs grad student elm@ginger.berkeley.edu #include {...}!ucbvax!ginger!elm "I like the Austrian way better." -- Dr. Henry Jones, Jr.