Cloud and Energy

In Data Centers and Energy I wrote about Hyperscaler Data Centers and Open Compute, and how they bring down the PUE of data centers, making them more efficient, and in Streaming and Energy I followed up on this, explaining how Netflix energy usage fits into this. Now the Uptime Institute has released a study that claims “Data center energy efficiency gains have flattened out”.
It is summarized as:
The average power usage effectiveness (PUE) ratio for a data center in 2020 is 1.58, only marginally better than 7 years ago, according to the latest annual Uptime Institute survey (findings to be published shortly).
But with a lot of caveats:
As ever, the data does not tell a complete story. This data is based on the average PUE per site, regardless of size or age. Newer data centers, usually built by hyperscale or colocation companies, tend to be much more efficient, and larger. A growing amount of work is therefore done in larger, more efficient data centers (Uptime Institute data in 2019 shows data centers above 20 MW to have lower PUEs). Data released by Google shows almost exactly the same curve shape — but at much lower values.
Which basically says the opposite: actually newer, larger hyperscaler cloud datacenters are way more energy efficient, and continued to do so after 2013. But a lot of IT is stagnating in old build, and is not being upgraded nor moved anywhere, so the numbers seem to stagnate when they are not.
It suggests the best way to improve the PUE of your data center is to move to get cloud and get rid of it, but Uptime being Uptime they cannot phrase it that way.
(via @speicherstief )