The main order of business in this underclocking experiment was to determine the lowest core voltage at which we could successfully benchmark each separate speed grade of the Athlon64 processor. The BIOS settings of the DFI Lanparty motherboard were used to lower the Core voltage and CPU multiplier at each step and a combination of SuperPI, 3Dmark2001SE and 3DMark2005 provided the benchmarks. The Prime95 burn-in program was also used to test operating stability at each voltage level.

Starting at the Athlon 64 4000+'s default clock speed of 2.4GHz, default multiplier of 12x200MHz and default core voltage of 1.5V, the core voltage was lowered to 1.4V. This proved to be the lowest stable voltage the processor would accept at stock speeds. A temperature reading was taken at full load during the Prime95 test, then the set of three benchmarks was run.

The multiplier was then dropped to 11.5x200MHz, reducing core speed to 2.3GHz and voltage was set to 1.35V. The system proved to be unstable at this point, so voltage was set back to 1.4V and temperature readings and benchmarks taken.

The core speed was then dropped to 2.2GHz and core voltage to 1.35V and we continued to stress test and benchmark the system, lowering core multiplier in steps and determining the lowest stable voltage for each. The results are listed below.

Temperature under Load by Speed/Voltage*
Clock Speed/Voltage	Temperature under Load (C)	Ranking
2.4 GHz/1.4V	33.5
2.3 GHz/1.4V	33.1
2.2 GHz/1.35V	33
2.1 GHz/1.3V	32
2 GHz/1.25V	30.4
1.9 GHz/1.15V	29.7
1.8 GHz/1.1V	29.5
1.7 GHz/1.05V	29.3
1.6 GHz/1V	28.8
1.5 GHz/1V	28.7
1.4 GHz/1V	28.3
1.3 GHz/0.95V	27.9
1.2 GHz/0.95V	27.9
1.1 GHz/0.95V	27.8
1 GHz/0.85V	27.6
900 MHz/0.85V	27.1
800 MHz/0.8V	26.9

*All temperatures were measured using a Fluke digital thermometer and thermocouple probe placed on top of the copper base of the heatsink. This gives a good idea of the temperatures the heatsink was radiating into the case. The ambient temperature was approximately 22 degrees Celsius.

While there was a noticeable drop in temperature, it was not a huge one. From 2.4GHz to 800MHz, the temperature decreased by only 6.5 degrees Celsius. To put it another way, for a 66% drop in speed there was a 6.5 degree drop in temperature. This makes a bit more sense if you look at the numbers in terms of Voltage not speed; a 43% drop in voltage producing a 6.5 degree drop in temperature seems more reasonable. The largest temperature drop occurred between 1.3V and 1.25V, where the temperature fell almost 1.6C.

The Athlon 64 4000+ can run stably at a fraction of its original voltage and clock speed, which was interesting to see, but how much CPU power was sacrificed in doing this?

Let's take a look at the performance benchmarks on the next page.