Re: CPU performance/reliability

["Adam Williams" <awilliam@xxxxxxxxxxxxx>]

> I am interested in knowing how Linux performs for the members of this
> group using different CPU's such as AMD, Celeron & Cyrix versus a true
> Intel. It's one thing to read about it, and another to hear first hand
> accounts.

Almost all CPU's will function normally.  Cryrix had some buggy version, and
rumors float around about the newest AMDs bieng slightly unstable.  Basically
 I think Intel tests their CPUs much more thouroghly.

> Does CPU make a difference depending on the application? For instance,
> would I want to pick a true Intel over say a Celeron for an IP Masq box,
> or is the Celeron OK? Is a fairly fast AMD acceptable as a file or
> database server, or should I stick with a true Intel?

It depends on the application.  Most applications actually depend more on
memory, graphics, or I/O then they really do on the CPU (relative to todays CPU
speeds).  For something like an IP Masq a solid 486 is more than sufficient,
 the NICs and bus probably make more of an impact.  For a "real" file or
database server I'd go SMP,  and that means Intel (for now).  Also applications
compiled specifically for a given chip will be faster than an app compiled
genericaly.  For example,  by squid box (http cache) is a AMD K6-300 with 128Mb
of RAM.  The k6 specific squid rpm places measurably less load than the i386
rpm did.   But squid does lots of sorts/searches through hash lists and
maintains a fairly detailed set of internal statistics.  A less CPU intensive
task (say e-mail or word processing) probably wouldn't show any real benefit.
 The only time Intel really wins out is in floating point performance (and lack
of bugs),  copy cat chips often claim similair FPU perf but they're actually
refering to "MMX" like extensions which only help certain functions (say
graphics) and are only of any value if the application/video-driver are coded
to use them (read "doubtful).  The extensions often require inline assember to
get a real performance boost,  and in "corporate" applications programmers are
loathe to use such things.  So if you want to crunch lots of numbers use Intel
(or if you want SMP),  else get a chip with a fast bus, and don't worry about
it.

>
> I guess the real question is this: how do these "economy priced" CPU's
> compare to what I have always felt was a no-brainer when it comes to
> building a server for production use: a true Intel CPU. For WinNT I
> definitely will stick an Intel (had a REALLY bad experience with an AMD
> and NT 4.0...). Does the old saying "you get what you pay for" hold true
> with CPU's and Linux?
>

Yes.  For a "mission critical" server, use Intel.  At least no one can turn
around and say "I told you so".  And my experience is that fewer Intel chips
turn out to be buggy  (I've had bizarre problems with Cyrix (in the past) and
AMD (recently), that a good old Intel chip cleared up).   Also recent AMD's
seem to have a problem running at their rated speeds reliably,  I've been told
by several people you have to notch up the core voltage to get a chip that will
remain stable for a long period of time.  Thats just to disconcerting for a
machine that will have 100+ users connected to it.

> I am running Linux on an AMD PR-133 (100 MHz) and it runs great as a
> desktop machine. I really am amazed at the number of things I can have
> going on at once on this machine when using Linux.

Yep,  ran 7 Xterms off a 486DX33 and all could run Wp7.0 with out any problems,
 simply amazing.  For an "average" desktop machine a fast pentium 1 is probably
sufficient.