AMD has revised its AMD Athlon ™ 64 Processor Power and Thermal Data Sheet to include results for its 90nm chips.
Unless you read it carefully, you’re liable to get fooled as to the meaning of the numbers, and if you think they really matter insofar as overclocking is concerned, you’re fooled already.
The AMD tech specs (page 12) say that the 1.8, 2.0 and 2.2GHz 90nm processors fall within a Thermal Design Power of 67 watts. What they don’t say is exactly how much power each chip actually chews up; it’s put into a general range.
AMD did the same thing with the 130nm Athlon 64s, giving all its chips a Thermal Design Power of 89 watts.
In this revision, though, AMD did break that 89 watt barrier by disclosing that the 130nm 2.6Ghz FX-55 will chew up 104 watts.
This tells us that a 130nm 2.4GHz Hammer (whatever flavor) probably chews up somewhere very near 89 watts (differing sizes of cache only makes a few watts difference), and chips running slower than that use somewhat less power (say, perhaps, 5-6 watts per speed grade, probably not more, more likely a bit less).
So a 130nm 2.2GHz Hammer probably chews up somewhere inbetween 82-84 watts.
Is that the figure one ought to compare to the 67 watt 90nm figure? Unfortunately not, because we don’t know yet if 2.2GHz is the end of the line for the 67-watt power parameter, and we won’t know until we see 2.4 (or better) processors coming out.
However, we do know is that the eventual power consumption of 90nm Hammers will reach the levels of current 130nm chips, which as we now know is in the neighborhood of 104 watts.
Given that the maximum released speed of a 90nm Hammer is likely to be 2.8-3GHz, looks like power consumption will rise rather rapidly per speed bump from this point on, which likely means that 67 watt figure is probably pretty close to the actual figure for the 2.2GHz, with numbers jumping up a lot from this point on.
And that is what overclockers ought to really be concerned about.
Looking at wattage comparisons at low default speed and conditions tells us nothing these days. It’s like comparing the engine conditions of a Ford Taurus and a Ferrari Tessarosa while running at 60 mph. That’s not where you’re going to find the difference between the two.
For instance, if you looked at this Intel datasheet for wattage figures, you’d see that up to 3.4GHz, the official Intel wattages are lower than the official AMD wattages. Does that mean Prescott is one cool chip? No, because of what happens after you get past 3.4GHz.
In other words, push Prescott past a certain point, and the power consumption figures go hyperlinear, power consumption increases much more than speed. It looks like something like that is likely to happen with faster Hammers, as bad as Prescott, no, but still much more than historical expectations would lead you to believe. An increase of almost 40 watts power (from 67-104 watts) for an extra 600MHz (say, from 2.2 to 2.8GHz) speed is a lot.
Want a current, real-life example of that? Look at the power consumption of the FX-55. 200 more MHz, 15 extra watts, or, 8% more frequency, 17% more power (granted, the envelope for 130nm performance is being stretched here).
That’s the situation we face with high speed 90nm Hammers. From what we’ve seen so far from overclocked Hammers, these things get toasty awfully fast.
This shouldn’t surprise you. This is what has been happening to all high-speed CPUs lately. AMD may end up being a bit better off than others, but that doesn’t mean they don’t have a problem at all. If they really had no problem at all, they’d be talking about getting 3.5GHz processors out within a year rather than just 2.8GHz, then going with dual 2GHz.
Will they end up being as toasty as Prescotts? No, but less hot than Prescott can still be too hot to handle, especially if you’re an air guy.
Less Bad Is Better, But . . . .
These are not cold chips by any means. Respect that. If you’re an air guy, you’re going to still need the best high-end cooling you can get for these chips because they’re going to get to 100 watts and beyond really fast even under moderate overclocking conditions.