Fellow phylo-dorks,
I was wondering if anyone had hard numbers (or a paper reference) for the energy consumption required by large phylogenetic analyses (or comparable computational problems from other fields). Perhaps @rdmpage, @ematsen, @alexei_drummond, @Alexis_RAxML, @mtholder, @phylorich, or @beerli might be able to help me out?
Thanks! JWB.
Hi Joseph,
In this paper: http://www.nature.com/nature/journal/v470/n7333/full/nature09676.html
They say: “220,000 central processing unit (CPU)–hours were used producing at least 7 tonnes of CO2” in the acknowledgement section…
cheers, – François
Thanks @francois! CPU-hours are sometimes reported, but I do not know the conversion to CO2-tonnes.
It seems like the CO2-tonnes should be a function of the wattage of the chips, as well as how long the computations were run. http://en.wikipedia.org/wiki/Performance_per_watt#FLOPS_per_watt
I guess it also depends on the source of energy that is used to power the computer…
Thanks @bredelings; that looks useful.
I have no data on electric consumption for migrate: the easiest way to get a comparison would be to run a kill-a-watt on a single computer with many cores, consumption on university clusters is very convoluted because these machines are usually run in shared usage and usually also do not do a detailed energy accounting because they are on the University grid. Calculations like this will not really help because FST calculations etc will be much cheaper but do we really want to only use those? I guess that running a computer cluster will be cheap in comparison to any field sampling trip (that involves plane/ship). But all that said, it would be an interesting paper to see a comparison.
cheers, Peter
Hi Joseph,
I don’t think I ever calculated it explicitly, but there seem to be tools for this, see here:
Based on this you can then extrapolate.
Alexis