Prescott: Not Fixed Enough

Intel said that they were going to work on power reduction in its next stepping of the PIV.

Well, the D0 stepping is now available for socket 775 PIVs, and so are the datasheets for them.

If you look at page 74 of the datasheet, you’ll find out that D0 stepping chips have a Thermal Design Power rating of 84 watts up to 3.2GHz, and 115 watts for CPUs up to 3.6GHz.

In comparison, the earlier Prescott stepping had a Thermal Design Power of 89 watts up to 3GHz, and 103 watts up to 3.4GHz, and Northwoods have a TDP of 82 watts at 3.2GHz and 89 watts at 3.4GHz.

What Do These Numbers Mean?

On the lower end, it means Prescott wattage numbers are only a little bit above Northwood numbers. That may sound good, and it is an improvement, but if the old rules on the effect of shrinking the CPU process still applied, the wattage on a 3.2GHz Prescott would have been more like 60 watts.

On the upper end, well, this is rather harder to interpret due to a bad habit both Intel and AMD have adopted lately.

How Much Power Does THIS Chip Chew Up?

Both Intel and AMD have stopped providing wattage ratings for individual processors. Rather, they give one maximum power/thermal rating for a range of processors.

This makes it hard for overclockers to judge how high power requirements increase when frequency is increased. It makes matters even harder when you don’t know whether or not you have all the processors in a range accounted for or not.

We can say with certainty that a 3.2GHz Prescott is going to chew up 84W, since that’s the last processor in that wattage range. However, we don’t know for sure if a 3.6GHz Prescott really chews up 115W, because we don’t know what the TDP of a 3.8 or 4.0GHz would be. Maybe 115W applies to either or both of them.

The reason why that’s important for an overclocker to know is that if it takes another 30 watts of power just to go up 400MHz, that’s absolutely terrible. If, on the other hand, it takes another 30 watts to go up 800MHz, that’s not so terrible (though still much worse than the power scaling with Northwood).

Our approach has always been, “When in doubt, don’t.” There will be a few pioneers that will effectively find out how bad the power gradient is. If there’s major improvement with the D0 stepping, we’ll know soon enough.

My suspicion is that this will make life easier for the freezers, but not much more than that. Given all the other problems socket 775 platforms have at the moment (easily worn-out socket, limits on FSB overclockability), it may not help even them very much.

Putting The Fans Out of Business

Even if the power gradient on these new Prescotts isn’t too bad, the end results are still not likely to be too appealing for heatsink/fan folks. 4GHz will probably still be the bridge these folks can’t cross.

If you see the projected AMD power requirements for 90nm Hammers, the numbers don’t look much better, which maximum thermal power for these processors coming in at 105W. Again, we don’t know what the range will be for that 105W (2.8 for sure, maybe 3.0, might be more), but even modest overclocking efforts would get one up to 130 watts or more pretty quickly.

That’s pretty hard to cool decently with air.

I think, in the short- to medium-term, the fan boys are going to be rather stunted in their overclocking attempts, no matter which company they buy from. Given that most overclockers still use air, and aren’t going to go to water or more, this is not going to be too good for this activity.

It’s going to be a choice of stepping up your game, lowering your expectations, or wait for the dual-core game to start.

Ed

Be the first to comment

Leave a Reply