Warning: Nerd Post II
After my first post about overclocking my new computer, I spent some more time testing the system. This time I was interested in seeing the relationship between the increase in CPU frequency and the needed increase of core voltage on the CPU. The new Sandy Bridge Intel processors have made overclocking ridiculously easy. Overclocking is basically increasing the speed of the CPU in performing calculations, or making the computer faster. The Core i5 2500k is rated to run at a stock frequency of 3.3GHz. The FSB is set to 100 MHz, and the default multiplier is 33 (33 x 100MHz = 3.3GHz). Overlocking is as easy as changing the multiplier, eg. from 33 to 45 (45 x 100MHz = 4.5GHz). However, as the frequency of the CPU is increased, the CPU requires more voltage to remain stable. In turn, the increase of voltage also raises the temperature of the chip and so there is a limit to how much it can be overclocked before it overheats. Thus, overclocking is a procedure of trying to achieve a higher CPU frequency (more GHz) while raising the CPU core voltage as little as possible.
Through a lengthy process of testing, I was able to find stable settings (3+ hours of prime95) for 3.3, 3.6, 3.9, 4.2, and 4.5GHz. I compiled the information into the two figures below. Fig 1 uses the CPU core voltage settings from the BIOS. Fig 2 uses the actual CPU core voltages while under load taken from CPU-Z (a monitoring program). For example, even though I set the CPU core voltage (Vcore) to 1.025V in the BIOS, the actual Vcore drops to 1.008V when the CPU is working at 100%. This is because the system automatically lowers the Vcore when under load to lower the chip’s temperature and prevent overheating.
Fig 1: The Vcore setting in the BIOS vs CPU frequency
Fig 1: The actual core voltage while under load vs CPU frequency
The black dots in the Fig 1 and 2 are the empirical data that I took. To predict the kinds of core voltages I would require in order to overclock the CPU above 4.5 GHz, I used quadratic, cubic (3rd degree polynomial) and exponential approximations. The cubic fit was the closest with an error of 0.0030 and 0.0019 volts with the Vcore setting and actual core voltage respectively. The quadratic fit was not far behind with errors of 0.0033 and 0.0032 volts. The exponential fit was the worst with an error of 0.0137 and 0.0150 volts.
The conclusion is that I would have to set the core voltage to ~1.5V in the BIOS (which would be about 1.475V under load) in order to run a stable overclock at 5.0 Ghz. Since the safe 24/7 core voltage for the core i5 2500k is said to be around 1.350V, the max stable overclock that I would be able to achieve with my system is around 4.6GHz. My results do corroborate with what I have read online that it becomes inefficient to overclock above 4.5GHz as the voltage step needed to bump up the CPU frequency grows at a constant rate with respect to frequency (first derivative of a quadratic fit is linear).
Anyway, it’s been a week of putting my new computer through its paces and now it’s time for me to actually use it.
