1.7GHz Athlon – too hot to trot?
One step beyond?
AMD was quietly confident earlier today in London when the company confirmed its plans for a 1.7GHz Athlon sometime in the second half of 2001.
The new part will still be built using the current 180nm process as the first 0.13 micron samples from AMD's Dresden fab won't come on stream until late in the year for production in early 2002.
In the light of Athlon's already profligate heat output - a 1.2GHz TBird produces more heat than a 180nm 1.5GHz P4 (even when overclocked to 1.6GHz) - the burning (sic) question has to be asked: just how hot will a 1.7GHz Athlon run?
The general consensus, shared by The Reg, is that Intel's P6 core, first introduced with the Pentium Pro when dinosaurs ruled the world, has a usable performance ceiling of around 1.2GHz when built using a 180nm process.
With the Tualatin 130nm die shrink of the Pentium III due in the middle of next year, this upper limit should be exceeded by perhaps 30 per cent. Above that, any extra power thrown at the processor is simply converted into heat rather than producing any useful work.
The same pundits reckon that the Thunderbird Athlons, being a tad newer than PIII, should be good for perhaps 1.5GHz - a speed AMD says will ship in Q1 next year. Unless AMD has something really cool - literally - up its sleeve, it's hard to see how the thermals of a 180nm Athlon running at 1.7GHz can be reliably and economically controlled.
It should also be a concern that Athlon doesn't possess an on-chip thermal diode to protect against overheating damage. Chimpzilla is obviously worried about the dangers of running its processors at higher than rated speeds as evidenced by the incorporation of clock locking on current Athlons and Durons.
Despite Intel's well-publicised hatred of overclocking, it has long been rumoured that AMD processors are usually marketed with less headroom than their Chipzilla equivalents.
Is AMD running the risk of an embarrassing replay of the Intel 1.13GHz PIII recall by trying to run a processor too close to its performance ceiling? ®
Sponsored: Beyond the Data Frontier