The last few days weíve been putting the Athlon 64 FX through quite some tests which resulted in a number of columns
with new info about AMD's latest. And as a result of this we were obviously contacted by AMD. AMD asked us to hold off on posting any benchmark data or results until the launch of the Athlon 64 series on the 22nd of September, about two weeks from today. Which would give other editors and publications enough time to properly evaluate their sample and would also mean that when the NDA expires there would be dozens of website releasing info about the Athlon 64. They also mentioned that if we did go ahead and posted a full featured article we could forget about ever getting samples from them or in any other way be a guest at any AMD launch event. Which comes as no suprise, as a lot is riding on this for AMD and they want to keep information from leaking prior to the launch on the 22nd of September.
So instead of posting a full featured article as we promised I hope you understand weíll stick with updating the column without going into performance data too much. That however doesnít mean there isnít much to tell, for example how about our attempts to get an impression how much headroom the Athlon 64 FX has got? We use a modified ChipCon Prometeia Mach II vapor phase cooling system to be able to keep the heatload of the processor within bounds. Our Athlon 64 FX sample managed to clock in 2.48GHz stable at -10C and 2.56GHz was the highest we could manage without it freezing up at bootup, no benchmarks at that speed though.
Fig 1. The modified ChipCon Prometeia Mach II cooling head attached to the processor.
Fig 2. The general tab of system properties showing the 2.48GHz clockspeed of the processor.
Obviously we tried to push it a little further, but we were simply held back by the processor. Upping the Vcore to 1.65-volts only caused the system to become more unstable as the heatload would increase by about 40-watts, more than the cooling system could handle. On average we measured a heat output of 116-watt at 2.48GHz at full load. That value was calculated from the capacity of the cooling system (in BTU/hr) and the difference between the room temperature and the die temperature. Using the supplied Cooler Master heatsink we pushed it as far as 2.3GHz, which is a mere 100MHz over stock, which is nice, but not a huge increase in clockspeed.
If we however put things into perspective, this is a first generation processor, Iím not surprised there initially isnít much headroom, this is normal. The integrated memory controller and the massive 1MB of cache both will quickly put the brakes on higher clockspeeds on a 0.13-micron process, as their signal timing is critical. But things should improve when AMD moves to a 0.09-micron process, clockspeeds usually go up byt 20 to 25% due to a die shrink which would yield about 3GHz. At stock speeds the Cooler Master heatsink did a pretty good job at keeping the processor cool, we measured temperatures around 55C under 100% load and about 43C when idling, which is about average for a modern processor fitted with a capable cooling solution.