🚀 go-pugleaf

In article <4l6rnv$753@netone.netonecom.net>, bbcat@netonecom.net   
(Michel Catudal) wrote:

> In <4l6kb3$p7n@news.aros.net>, John Brush@aros.net writes:
> >>
> >>>... because OS/2 tests memory in the lower address space to determine its
> >>>speed, then runs ALL memory at that speed
> >>
> >>Unbelieveable! OS/2 *cannot* "determine the speed of a bank of memory".
> >>Period.
> >>
> >
> >I won't be quite as emphatic, but this does sound a bit dubious.
> >
> >John
> 
> 
> I've heard that crap from several people. Actually you could find out how
> fast a simm can run but you have no guaranty that it could run that
> fast continuously (assuming you pick the high speed).
> Anyone done that is playing with fire, the 70nS part may work fantastic
> as a 60nS for 99.9% of the time then oooppss!
> As far as the manufacturers, too funny to comment.
> 
> Michel Catudal
> Proud user of OS/2 Warp

Considering the general audience for Warp, what benefit to the user would
there be for the OS to clock the memory subsystem?  And, at what cost?  I
could see putting monitor and control on the memory subsystem for a
real-time embedded application where timing is essential to the proper
function of the system, but I just don't see how that would provide
significant advantage on the desktop or on a server.  The demands just
aren't so critical that everything has to be in perfect sync and tune to
perform well.  Your much better off leaving the chores such as memory
timing up to those subsytems that are more concerned with it.

Now, the OS would be really concerned whether the memory gives it data
that is valid, so performing a couple checks to verify that the memory
basically checks out, and even to check out which CPU you're on, might
prove beneficial, but for desktop use, there is very little difference
between 50 and 80 ns once you get to end user perception.  At least, I
know I can't tell the difference between getting smashed in the face for
50ns v. 80ns.  It still happens pretty darned fast.

It would be really interesting to find out which OS task is so important
that it has to start at one edge of a memory cycles and end at the other
end.  It just doesn't seem that there would be anything so critical to the
OS that it would need to do this.

                 Kevin

--