home *** CD-ROM | disk | FTP | other *** search
- Path: vixen.cso.uiuc.edu!msaladin
- From: msaladin@students.uiuc.edu (Saladino Michael D)
- Newsgroups: comp.sys.amiga.programmer
- Subject: Re: One hardware-basher's manifesto
- Date: 27 Feb 1996 08:18:01 GMT
- Organization: University of Illinois at Urbana
- Message-ID: <4guenp$1a1@vixen.cso.uiuc.edu>
- References: <4ge8na$vhe@ar.ar.com.au> <08000305729070351637@BIRDLAND>
- NNTP-Posting-Host: ux5.cso.uiuc.edu
-
- >> : > Please use OwnBlitter/DisOwn... and use CacheControl to turn on only the caches
- >> : > you need (If data burst (030/040/060) modes is on the CPU will take 4x as much
- >> : > time when accessing fastram randomly.
-
- I don't think I can express in words just how wrong this is, but I'm going to do my
- best. First of all, even if you were accessing RAM "randomly", you will not suffer
- a 4x speed drop. Checking a cache (L1 or L2) you only suffer a few clock cycles and
- this is kept to a minimum thanks to pipelining the cache lookup hardware. However,
- this is a moot point because programs don't access RAM randomly. The whole point
- of the cache is to take advantage of the temporal and spatial locality that occurs
- in ALL programs. You would actually have to go WAY out of your way to make a program
- that runs faster with caches turned off and I bet it wouldn't do much. Are you aware
- that modern day processors (and I'm even including the 680x0 line) get over 90% hit
- rates on their data caches and even higher for the instruction cache. Turning a
- cache off is like flipping a switch to drop your clock from 50Mhz to 5Mhz. It's
- stupid... don't do it!
-
- Caches should always remain on. ALWAYS.
-