BOGOMIPS are really BOGUS

Lies, Damn Lies and Benchmarks

I enrolled in a course with the Linux Foundation that asked me to insure my systems were up to the task of handling the courseware. Their performance metric and benchmark was using Bogomips that measures the amount of time it takes to execute a Busy Loop while booting the kernel.

I am not a newcomer to benchmarking CPU performance since I was the founding chair of the SPEC Steering Committee a lifetime ago. Needless to say, I was appalled at the measurement of my humble Raspberry PI4 with 8GB of RAM vs. their courseware expectations. Granted that a classic laptop with an AMD or Intel CPU would pass this with flying colors, but according to Bogomips, my system was 1/60th of the performance of a machine with an Intel CPU. That did not feel correct to me since I had hooked up Raspberry PI4 machines to the TVs in my bedroom and living room (yes retired single life is that boring!).

Most of the "off the shelf benchmarks" (e.g. existing SPEC and Phoronix benchmarks) were too focused on the classic database access or computationally challenged tasks for simulations and modeling for my taste. If you saw my previous blog, you noted that I have played around with cryptographic hashes.

Bottom Line

I spent about 6 weeks writing my own benchmarks with cryptographic hash programs and using "dd" to compare the amount of CPU and wall time it took to simply copy data, versus taking the hash of the data. In my testing I found that the ratio was not 60:1 but only 2.2:1 or 2.8:1, which matched my own intuitive feel.

Here are the Performance ratios of my hash benchmarks run across a variety of machines. All of the gory details, including bash scripts, and a longer and boring version of this note can be found at:

I include one graph and two tables for your bottom line.

Your Mileage May Vary

This phrase was popular when the EPA first started testing gas mileage efficiency for cars. Since your results may vary, you can rerun all of my simplistic tests. I am certain that overclocked gaming CPUs with lots of RAM will have much higher performance ratios. I just wanted to know how to compare against the laptop and deskside computers that I used for writing, editing and testing code would compare. I have found these inexpensive computers to be remarkably capable compared to what past employers have purchased for my use... but then I come from the dark ages before there were true laptops.