Something you might find interesting

[CTho9305]

I've been meaning to do this since I was in college... finally got a friend to run the experiment on a non-NDA process technology (one of the ones used in academia). I don't know exactly how he sized the gates, but for these purposes, whatever he did worked well enough.

http://ctho.ath.cx/pics/ring_oscillator_delay_vs_temperature.png
This shows the performance of a simple circuit at various temperatures. The circuit doesn't include long wires, so scaling of wire resistance won't show up here.

http://ctho.ath.cx/pics/ring_oscillator_delay_vs_temperature_chart.png

Temp (C)__Delay (picoseconds)
100______795.6 ps
50_______676.6 ps
0________566.2 ps
-50______464.2 ps

Alternately, it could run at:
100: 1.25GHz
50: 1.45GHz
0: 1.75GHz
-50: 2.15GHz

I believe these temperatures are the temperatures at the transistors (rather than on the top of the package), so they're highly unlikely to correlate with the temperatures your CPUs report.

Real chips have additional complexities that tend to make the scaling less than ideal (for example, some circuit styles are sensitive to the ratios of the speeds of different transistor types which can change with temperature; other styles are sensitive to the relative delays of gates).

 

[RESmonkey]

What college are you in/what year?

That experiment looks interesting. How did you do it?

 

[CTho9305]

I got my MS from Carnegie Mellon last year. Since I no longer have access to their systems, and the processes I do have access to are all under NDA, I asked a friend who's still in school to build a ring oscillator (a loop of gates set up so that they constantly switch between 0 and 1) and simulate it at different temperatures. This kind of information is probably available in other places (and I'm sure something like this was part of a homework assignment at some point), but I haven't seen it on forums.

 

[myocardia]

That's a cool experiment, but if you want to know about processors, all you really need to do is look around in a few forums. For instance, nobody has ever gotten a QX6700 over 4 Ghz on air; not anywhere close. But yet, nearly everyone who owns one, plus a decent phase change setup, can go over 4 Ghz. It's kind of cool to see it a bit more scientifically, though.