[rescue] Prices... Amazing

[Nathan Raymond]

On Mon, 2 May 2005, Scott Newell wrote:

> At 11:31 AM 5/2/2005 , velociraptor wrote:
>>
>> The E250 is just slightly less than the Ultra 2 in wattage (.4 KVA,
>> which I assume would be 4 A & 350 W), but the 250 is almost
>> double the BTU/hr vs. the U2 (683 BTU/hr).  The Ultra [10/30/60/80]
>
> I don't understand what you're getting at here.  As I see it, if an Ultra
> 2, an E250, and a FeeCee are each pulling, say, 300W, they're also each
> putting out the same heat.

No, that's not accurate at all to assume they'll put out the same heat 
based on power draw.  First of all, power supplies vary greatly in their 
efficiency (i.e. how much current they draw from the outlet vs. how much 
they deliver).  On top of that, different power supplies have different 
efficiencies at different loads, so that one power supply may be very 
efficient under heavy (full) load, another not so much, and vice versa 
under low load, while some can be very efficient at both, but without 
testing a specific power supply and seeing how much power it actually 
consumes and heat it actually puts out at a given load, you won't know.

Second of all, a CPU has two primary types of power loss.  The first (and 
traditionally prime) one has been the switching losses.  This is the power 
lost and converted into heat when the transistors move from a 0 to a 1 
position, and vice versa (i.e. capacitive losses).  The second type is 
known as leakage current losses.  Up until recently, leakage current 
losses have been small enough that they could be ignored in early analysis 
of a given design.  As process designs have evolved and gotten smaller, 
leakage current has been getting exponentially larger. With the shift to 
90 nm, the processor crossed that point where leakage current losses get 
very large, very quickly.  Now leakage current losses are the dominant 
cause of heat dissipation.  This is due to electrons getting out of the 
channel where they are supposed to be following, taking their own path 
somewhere else, which causes them to be transferred into heat loss. 
Therefore, CPUs with a process size larger than 90nm will actually be more 
efficient in terms of using power for switching operation vs. losing power 
as heat, and will correspondingly put out less heat per watt of power 
going in.

So you can't just look at watts consumed and correlate it to heat output 
because the efficiency of various components (power supplies and CPUs in 
the two examples I just gave) will vary from machine to machine.

- Nate