historic speed of Matlab

Hi guys,

I am writing a paper about how a matlab toolbox can be used in teaching at universities and schools. It uses real-time sound feedback and the audio-toolbox. In order to find out which models can run, I need some guess about rough computer speed, as all of the calculations are time critical to allow for a real-time audio (and visual) experience that is required.

I can measure matlab speed on the usual workstations at with the Bench command (the one I am sitting around 70), and that is extremely useful for me to write which computer roughly other universities (or schools) would need to buy to achieve this performance. However, most places have presumably somewhat older computers and less funds, so it would be great to be able to get OLD matlab benchmark measures. Is there any way to access (maybe someone has them?) or measure OLD machines systematically as function of time. I would be interested in ‘the average PC available to students in 2000/2005/2015) had these benchmark results. It doesn’t have to be precise, this is not a precise art, but any date would be useful to assess if the growth is linear or flattening.

I did an internet search, but apparently no-one is collecting such information.

Thanks,

Stefan Bleeck

[Jan]

The benchmark function has been modified with the different Matlab versions, such that you cannot compare the results between a Matlab 6.5 machine with R2019a. You need to decide for a specific benchmark version at first. I do not know, if delays in the audio interface are considered at all in the benchmark tools.

What is "the average PC available to students in 2000"? If it was a single core, the activity of the virus scanner is essential for real-time processing. The "average" PC in 2000 might use 1.6 cores - but this is not meaningful.

The graphics output has slowed down dramatically in the last years, especially with introducing HG2. While creating or updating a simple GUI with some buttons takes seconds today, it ran in 0.4 seconds in R2009a and 0.2 seconds in R6.5. (Note: I took these timings on a 4-core i7 for Matlab 2018b and 2009a, but on a PentiumIII for R6.5.) But if you draw 3D patch objects with hundreds or thousands of elements, modern Matlab versions will outperform the older ones.

This means that asking for benchmarks will not be only "imprecise", but not useful in any way to estimate the performance of a very specific job. It is like asking for the average speed of cars in the last 100 years: It depends on the roads and the speed limits, on the relation between high-ways and streets inside cities, and on the decision if trucks are considered as cars or not.