When talking to interested people about liquid cooling as defined by “liquid invasion of the IT”, it seems PUE calculations do not show necessary benefits. For example, we retrofit a liquid cooling technology to the IT and for argument sake we drop the total energy draw of the digital components by 10%. We then raise the temperature onto the heat rejection system making it operate more efficiently, the average coefficient of performance increases from 3 to say 4 and UPS efficiency drops from say 91% to 89% as the critical load decreases on average. Then if we started with a PUE of 1.4 before retrofitting the liquid cooling solution would give a PUE of 1.5 (thus not helping the case). The arithmetic indicates simply that the changes in the denominator have dropped more than the numerator.
However if we use TUE , which is defined as total energy consumed by the entire data centre divided by the digital “only” components of the data centre, we get a different story. So before retrofitting of the liquid cooling solution the PUE of 1.4 would be a TUE of 1.63 and after retrofitting the TUE of the liquid cooling solution would be 1.58, showing improvement.
The interesting notion of this TUE is that we have a numerator which is affected by the combined digital workload and ambient conditions, whereas the denominator is only affected by the digital workload! In fact introducing ITUE (see ), which treats the IT (which has its own power and cooling) in the same vein as PUE for data centres, then
TUE = PUE x ITUE
TUE would have less manipulative capability for some obvious reasons. Should IT manufactures quote their ITUE at different workloads? In fact I am aware that ITUE was being looked at by the International Standards.
 Energy Efficient HPC Working Group https://eehpcwg.llnl.gov/pages/infra_itue.htm