Subject: Re: NoName and cache
To: None <mouse@Rodents.Montreal.QC.CA>
From: Joerg Czeranski <email@example.com>
Date: 03/18/2001 12:37:41
der Mouse wrote:
> I've got a NoName, 16MB RAM (4x4M), and its speed is comparable to or
> perhaps a bit slower than my SPARCs' (doing a "make build" with
> comparable disk drives).
> How much difference does adding cache actually make? Does anyone have
> before-and-after numbers, or even "feel" comments? (I'm just thinking
> that there's no point sinking effort into finding cache chips if it
> really doesn't make any practical difference, for example.) Does 15ns
> versus 20ns cache make any practical difference? Would adding more
> ordinary RAM help? (If so, anybody got a source for RAM for the thing?
> All my 16MB sticks are 32-bit or 33-bit, not 36-bit, and I think
> they're all slower than 70ns anyway.)
I have a NoName and once measured the effect of adding cache.
The situation is a bit different for me, though:
It's a 233MHz 21066A with 128MB RAM and Dec Unix 4.
I bought 256KB 12ns cache (that's an unsupported option, ask me
if you want to know the "secret" jumper setting).
As a benchmark I compiled O'Caml. To reduce the influence of the
hard disk, I put everything into /tmp mounted as memory filesystem.
The speedup was a factor of 1.6. I think it should've been even
better with 1MB cache, but I didn't find a source for the wide-
Of course with a mere 16MB RAM it might be quite different,
especially as gcc really wants a lot of RAM.
I think you should first try to install enough RAM to prevent paging
before you think about cache.
Unfortunately, 36-bit PS/2 SIMMs are rare and therefore rather expensive.