In theory, it should indeed be possible to measure the actual "temperature" inside the "computer" degrees of freedom. What would we need? 1. We'd need an exquisitely sensitive thermometer with an extremely small heat capacity. This is because almost every real thermometer would be Avogadro-sized in comparison to the relatively small number of bits in our computer. Thus, the heat from the thermometer itself would overwhelm the relatively small heat capacity of our "computer" degrees of freedom. 2. We'd need a way to put the thermometer "in contact" with the internal "computer" degrees of freedom. This will be difficult, since computers are well-designed to keep these internal degrees of freedom well-insulated from the "outside". 3. Our thermometer itself would have to be very well insulated from the rest of the world, since we are trying to measure a temperature lower than the rest of the world; we don't want the thermometer itself to be a conduit for heat flow into the "computer" degrees of freedom. Due to these difficulties, it is simpler to calculate/estimate the actual "temperature". At 01:22 PM 12/10/2013, Eugene Salamin wrote:

You seem to be using the term "temperature" in a different sense from "that which is measured with a thermometer". All computers that I'm aware of generate heat, and in fact heat removal is the major obstacle to faster CPU frequency. In your sense, the coldest place on Earth is a petroglyph, a stone carving.