Chipmakers Admit Your Power May Vary

Dylan Knight Rogers writes to mention a News.com story discussing the realities of chip power consumption. From the article: "Assessing only pure performance is passe. The debate these days is about performance-per-watt, which seems like it should be a simple miles-per-gallon type of calculation. However, miles are miles, and gallons are gallons. There's no one simple way to measure processor performance, and measuring the amount of power output by today's chips is proving just as difficult."

News?

[tomstdenis]

I'd think both AMD and Intel are well aware of the MIPS/Watt challenge. It's not new. Problem is CUSTOMERS still want a bazillion Ghz attached to the processor because they think it will make it faster or better or something.

I've got two x85 class Opterons sitting here at 1Ghz most of the time. That's ~35W vs. ~95W. AMD seems to care about power. Intel is no worse off with the Pentium M and "core" series (netburst was a mistake).

Tom

Re:It's not just the CPU

[NutscrapeSucks]

My 86 Camry will beat your 2007 Camry in a drag race

I recently read an article noting that the 2007 Camry (with 250HP or whatever) will out drag race most sports cars from the 1980s, much less the sedans. Although, greater point taken about bloat.

Re:How is this news?

[Firethorn]

they only tolerate a fraction of a milliohm of added resistance

Say what? They're not [i]that[/i] intolerant. Otherwise the overclockers wouldn't be playing around with increasing the voltage. Normal power supplies would have to be far better, and motherboard power compensators far more expensive. Besides, if your measurement device adds that much resistance, you simply increase the voltage of the rail a smidgen to compensate.

Now, I am talking about doing all this in a lab, for best results.

The true difficulty comes in that the measurements are highly dependent upon what you're doing with the CPU. Acting as a webserver, managing database access is a totally different from calculating weather patterns, for example. Different CPU's have different performance for their floating point systems, integer, memory operations, etc...

Re:I would like to know...

[wkitchen]

Why, for a given chip, power consumption raises with clock speed? I know there's corelation, but I'd like to know the physical relation between the two variables.

When not changing state, a CMOS device dissipates almost no power. But each CMOS gate has a tiny capacitance that must be charged or discharged each time it changes state. This requires energy. The energy dissipated for each transition is essentially constant, but the number of transitions in a given time can vary. Since power=energy/time, the more transitions per unit of time, the more power is required and the more heat is generated.

Re:I would like to know...

[Anonymous Coward]

When a gate output switches from one level to the other, there is a brief amount of time of which a low resistance connetion is made from the power supply directly to ground. Since ALL switching is based on the clock, this is when most of the power is dissapated.
    This happens because of the way CMOS logic works. Being Compilimentary Metal-Oxide Semiconductor logic, every gate has p-type and n-type transistors. n-type transistors can only drive a '0' but p-types can only drive a '1' so both are needed to drive both values. I drew an inverter below so you can see a simple case. Normally both are on during a transistion so the output is always at a defined value. Shutting off one before enabling the other is a much worse condition as that would cause the gate output to go undefined for a short while. This would cause the next gate to switch uncontrollably depending on the heat/radiation/nearby magnetic fields acting on the interconnect wire.

Vdd
|
p-type
|
Output
|
n-type
|
Ground

Re:The Solution

[Anonymous Coward]

Well, that's lovely, except that in the real world performance-requiring apps ARE optimised. And the level to which software can be optimised to a chip is a selling factor.

I don't care if some useless piece of code completes in 30 seconds on Pentium64 and 45 seconds on CeleronFX, if I get smoother Quake 4 play on the CeleronFX because Quake 4 has optional CeleronFX specific code paths, and linux boots faster because gcc optimises better for CeleronFX.

What you're describing is called a synthetic benchmark, and they are pointless for anything except bragging rights.

Re:Well...

[smoker2]

That's why manufacturers base their MPG figures on something called the Urban Cycle. [vcacarfueldata.org.uk]

This takes in slow city traffic, faster freeway traffic and top speed travelling, approximate to an average consumers car usage.