Warning: Nerd Post

This week, I am on winter break and came back to the Bay Area and upgraded my desktop computer. It had been pooping out on me and there were some incredible deals that were just too tempting. Currently, the build is a MSI Z68A-GD65 (G3) motherboard, Intel Core i5 2500K processor, Nvidia GeForce 9800 GTX+, and 8GB of G.Skill Ripjaws DDR3 RAM. I don’t normally overclock my computer because I don’t really need to nor is it worth it. However, like a kid with a new toy and plenty of time to play, I took it for a spin. I took some baseline tests at the stock 3.3GHz and then overclocked to 4.5GHz to compare and there was about a 25% improvement in performance. This made me curious about the relationship of overclocking and the increase in performance. I decided to do some tests.

Test Methodology

HyperPi - Calculates Pi to 1 million digits, the less time it takes the better.
PassMark Performance Test 7.1025 - A series of tests for the CPU, higher score is better

Starting at the stock frequency of 3.3GHz of the Core i5 2500K, I raised the clock by 0.1 GHz and ran HyperPi and PassMark three times. I then took an average of the three results. This took awhile because I had to restart my computer each time I changed the CPU frequency.

Results


Table 1: Data from all the trials

To get a better idea of what all the data meant, I plotted the values in MATLAB and noticed they all lined up fairly nicely. So I did a least squares approximation using a linear fit and plotted them together to see how close the linear fit was to the experimental values.


Fig 1: Plot of HyperPi data and the least squares linear approximation


Fig 1: Plot of PassMark data and the least squares linear approximation

I thought it was interesting that the performance increase of the CPU was linear, I initially guessed it would be exponential or maybe quadratic because of the jump in performance from 3.3Ghz to 4.5 GHz. From the linear approximation, I calculated that for every 0.1GHz increase of the CPU frequency, there was about an 1.77% and 2.71% improvement in HyperPi and PassMark respectively. After overclocking by 1.2GHZ, that’s a 21.2% and 32.5% improvement in HyperPi and PassMark respectively, an average of 26.9%. For about a 36% increase of the CPU frequency, you get about 27% increase in performance. That’s actually not that bad.

The bottom line is that it’s pretty pointless to overclock when I spend 75% of my time surfing the web and using Microsoft Office, but it’s cool just because I can. These numbers that I generated are not definitive and there’s probably a margin of error to it. However, I found it interesting and other geeks out there will probably too. When I did the tests, I had my core voltages set at 1.35V because I knew that it was stable for sure up to 4.5GHz. After slight tweaking, I have found that I can run 4.5GHz stable set at 1.33V with low Vdroop (after 3 hours of prime95).