Raspberry Pi Overclocking Benchmarks

Over-clocking a Raspberry Pi?

The Raspberry Pi is an amazing device for $35. I have been tinkering with mine for a few weeks. Recently, the Raspberry Pi group offered directions for over-clocking a Raspberry Pi from its stock CPU frequency of 700 MHz to a “turbo” mode of 1000 MHz. Even better, they have guaranteed that using the “turbo” mode will not void your warranty or damage the Raspberry Pi.

The question I had was – what would the impact be of this overclocking? Therefore, I decided to test the CPU under a number of frequencies from the stock setting of 700 MHz through the “turbo” mode of 1000 MHz. I used the nbench Linux tool to benchmark my Raspberry Pi’s CPU. I changed the CPU frequency via the raspi-config command-line utility.

The results

The following table shows the CPU frequency of my Raspberry Pi and the “integer index” and “floating-point index” calculated by nbench. Higher values are better.

CPU (MHz) Integer index Floating-point index
700 10.752 3.474
800 13.024 4.244
900 14.475 4.653
950 14.575 4.501
1000 16.237 5.475

As the table shows, the difference from the stock speed of 700 MHz to 1000 MHz is substantial. This chart shows the steps visually.

Raspberry Pi Over-clocking Results

What is of special interest to me are the actual performance gains over stock, over the previous level, and the performance per dollar of acquisition costs.

Performance improvements over stock

CPU (MHz) Int improvement (%) FP improvement (%)
700 100 100
800 121 122
900 135 134
950 136 130
1000 151 158

Performance improvements over previous speed

CPU (MHz) Int improvement (%) FP improvement (%)
700 100 100
800 121 122
900 111 110
950 101 97
1000 111 122

Performance index per dollar

This table breaks down the performance per dollar of the Raspberry Pi’s cost.

CPU (MHz) Cost per Int ($) Cost per FP ($)
700 3.26 10.07
800 2.69 8.25
900 2.42 7.52
950 2.40 7.78
1000 2.16 6.39

Conclusions

These benchmarks allow me to draw a few key conclusions.

First, my benchmark results are close (but not identical to) the results from the Raspberry Pi foundation. Therefore, some variability likely explains different results. I used a stock install so I am not sure what they did (or did not do) to achieve higher benchmarks.

Second, the “turbo” mode makes a huge difference! My Raspberry Pi is getting 151% integer performance and 158% floating-point performance over stock. Those are some astounding numbers for a software tweak that is vendor-supported (and endorsed!)

Third, If you do not want to take a risk (some SD card corruption has been reported with the 1000 MHz setting) than using 900 MHz offers the nearly the same performance as the 950 MHz setting. This slower speed may offer more “safety” with a little impact.

Testing notes

To run these tests I disabled wpa_supplicant and transmission-daemon which consistently negatively impacted the nbench results about 3.4%.

Leave a Reply