Path: utzoo!attcan!uunet!seismo!sundc!pitstop!sun!decwrl!purdue!mailrus!cornell!batcomputer!braner
From: braner@batcomputer.tn.cornell.edu (Moshe Braner)
Newsgroups: comp.sys.transputer
Subject: Re: T800 Memory Design
Summary: External RAM slows it down by a factor of 2 or so
Keywords: Transputer, Memory
Message-ID: <6497@batcomputer.tn.cornell.edu>
Date: 8 Oct 88 20:37:49 GMT
References: <502@koel.rmit.oz>
Reply-To: braner@tcgould.tn.cornell.edu (Moshe Braner)
Organization: Cornell Theory Center, Cornell University, Ithaca NY
Lines: 18

[]

The Inmos documentation (I forget which book, I think it's one of the
little red ones) says that the number of extra machine cycles needed
to access external RAM is at least 3 and more typically 5, depending
on your hardware setup.  They have a table of the effect of putting
code, data, or both in external RAM on the execution speed of a couple
of benchmark programs.  The upshot is that relative to running everything
inside the transputer's on-chip RAM, putting code outside slows things
down by a factor of about 1.3, putting data outside slows by 1.7, and
putting both outside means about a 2-fold slow-down.  Note that data
has a larger effect than code, perhaps because the transputer reads
code 4 bytes at a time and each instruction is usually only 1 byte.
So put the workspace of your most time-critical function on-chip,
and use local variables in preference to globals, and you're going to
get almost full speed.

- Moshe Braner