We pointed out the other day that Prescott is simply too hot.
A few places provided measurements of just how hot Prescott is compared to Northwood using Intel’s standard cooling, and another test (under somewhat overvolted condition) shows that Prescott turns a watercooling into a water heating system.
What is frightening about Prescott is that these results represent winter-like temps for this processor compared to what will happen when the chip ramps and/or is overclocked.
The Intel datasheets show (rather questionably( that the 2.8 and 3.0 both chew up 89 watts of power, while the 3.2 and 3.4 chew up 103 watts of power.
This tells us that from every 400MHz ramp of Prescott, we should expect it to chew up at least another 14 watts.
Let’s assume the best for Intel and presume 89 watts for a 3.0E and 103 watts for a 3.4E.
Project that 14 extra watts per 400MHz upward, and you get:
Projected Prescott Wattages: Default Voltage
Of course, it is not likely at the higher speeds that default voltage will do, so let’s apply Ohm’s Law (wattage increases by the square of the percentage increase in voltage) and see what happens when you overvolt by a modest 10%.
Projected Prescott Wattages: 10% Overvoltage
Ugly, isn’t it?
The real hardcore folks, of course, won’t stop at a measly 10% overvolt, so let’s see what the numbers look like with a 20% overvolt.
Projected Prescott Wattages: 20% Overvoltage
Compare these numbers to the maximum 3.4 GHz Northwood/overall Hammer wattages (which very conveniently happen to be the same): 89
It’s not much of an exaggeration to say that for serious overclockers, Prescott is like two processors in one: it will put out the heat of two processors.
This is not good.
So when people tell you that Prescott should easily ramp to 5GHz or so, you might want to mention the inconvenient fact that the way it looks now, it will take about 200 watts to do so.
You’re not going to cool that with no friggin’ fan.
I don’t know if MOSFETs explode, or just die quietly, but I bet we’re going to find out soon.
The Intel datasheets indicate that Prescott is supposed to operate in a voltage range of 1.4-1.25V (I put the higher number first since all PIVs use a variable voltage scale; it starts at the high end, and drifts downward under load).
1.4V is only about 10% less than the maximum voltage for Northwood (1.55V), which is much less than you’d expect from a process shrink.
For that matter, it’s less than people designing circuitry for Prescott expected just six months ago, as this document demonstrates, among others. Prescott was supposed to come in at 1.225V.
It’s a bit unclear at the moment what the default voltage is, but if it’s 1.4V, that’s about 15% above the expected 1.225V. Quite a sizable overvolt all by itself.
Of course, when you’re Intel and you get to determine what “normal” is, there’s no such thing as “overvolting,” but if the CPU is demanding 15% more wattage than what “normal” was supposed to be six months ago, one has to wonder how much more power you can add before it becomes “too much.”
Don’t be too surprised if you hear about Sudden Prescott Death Syndrome one of these days.