[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: server hardware
tuvia beker wrote:
<...>
> Note thatthat the 300A celerons are not only the most overclockable chips
> ever made - the fact that their cache is on-dye compensates for its
> reduced amount, and this compensation is even more noticeable at 450MHz.
> In fact they are only slightly weaker than regular P450.
> I seriously consider doing it, mostly for the fun of it...
<...>
Yeah, sounds like fun indeed. Me and a couple friends of mine are going
to SMP-ize the Celerons in early December. However, in my opinion it's
better to use regular Celerons, not the A-type: they're cheaper, you can
get a used ones and the multiplier's not locked, like on the newer Intel
chips.
The real question is whether or not the penalty for absence of the cache
will be really that great? Obviously, it will depend on application. For
instance, there are reports that show that the execution speed of some
benchmarks on SMP system with disabled cache is "only" 30% lower then
with the cache enabled. I can imagine, that non-threaded applications
(Netscape Proxy Server comes to mind) will suffer less, whereas the
highly-threaded ones (DB's) -- more. How the overall speed of the system
(kernel-wise, especially under heavy load) will be impacted by the
absence of the cache? Can it be helped by tweaking of the penalty for
context switching kernel parameter? I'm interested to know your opinion
on the subject.
Best regards,
--
Nahum Greenberg
PF1 Systems Ltd.
http://www.pf1.co.il