[gentoo-user] Re: new machine : case + power

[james]

Alan McKinnon <alan.mckinnon <at> gmail.com> writes:


> I have yet to see a computer that draws 350W sustained
> Spikes of brief duration yes, sustained no.[1]

Maleformed arguement here, pal. It was a 'generic example. Obviously
if his rig draws 200 W on a cheap/old 400W supply, it will be fine
until the components age.

> But now I'm being pedantic right after you tried so hard not to be.
Fine, let's dance, as one of my areas of expertise is fluid flow besides
EE.


> [1] Anyone who wants to know just how much power 350W really is,
> consider that your electric kettle is about 1000W and can boil 1.7l of
> cold water in 2 minutes. I know [CG]PUs get hot, but they don't get
> *that* hot.


Huh? You really should stay in  the E&M domain, for accuracy and not
confusing the readers.


Hook up some sophiticated electrical monitoring equipment that also
logs data points, (wave-forms) at say 100+ points per second resolution
and leave in on a computer for months. If it is heavily used, or your
utility is crappy (as they all are do to costs) you will see transients,
some of which pass right through the UPS (slow to react)  and the
PS (particularly if the filter stages are poorly design (which most are
as the designers are often not allows to use expensive components)
and they hit the components on the PC side. Over time there is a 
cumulative effect. Also, components "burn in" and then mostly offer
reliable performance until they reach the end-of-life, statistically.
AT the point some last longer, other have failed and still others accentuate
the effects of transients. Most system tolerate this for quite a while,
granted. But he is now talking about purchase of some expensive
components; for example a CPU that is running at a base speed of 4.3GHz.
That load is far more demanding than the same wattage load with a 2 GHz
CPU. Higher frequency components have much faster transients (skinnier
on a wave-trace). So they just pass right through crappy (low cost)
UPS and PS. Furthermore, they by nature of their clock speeds induce
transients behind the PS. Sometimes transients cancel and sometimes
they are additive. Now all of those circuits are becoming "distributed
parameters" and cannot be accurately characterized by lump sump analysis
methods. (such as I have previously describe using cheap technician tools
like multi meters and amp meters.


Ferro resonant UPS fix most of the this, at least from the utility side, but
they are hard to find in smaller USPS. A good old fashion 1-to-1 iron core
transformer in  the mix does the same thing, but the power loss is 3-15%
depending on many factors.

Sure there are micro components on the mobo to do some of the
same thing, but in cheaper mobo they do not do the job well.  Transients
penetrate (transgress) from the circuits where they are suppose to
be "contained" into the other circuits close to those hi freq sources
(4.3 GHz is Hi freq for a digital designer) and therein cause lots
of problems. You just can't see it. Now what is the freq of the DDR3-5
ram on the GPU?  The freq of the System ram?  Get the picture? I've design
quite a bit of gear, at various frequencies:: I should be 'in the market
for something really cool, say around 2K for nerds, just like you,
that like to show off at parties. Specs are done, but it will be approaching
3000 W, have 8 channels and should radically enhance your 'dance moves'!

Me, I would *never* put an old crappy standard 400 watt power supply
on a new 4.3 GHz system. ymmv

PPS, I really enjoy (too much) the pedantic beat down, but I'm done
on this one, so rave_on, ?


James