Nvidia PUE-PUEs datacenter efficiency ratings, calls for application-specific metrics

Power use effectiveness – PUE for short – has long been the spec by which datacenter efficiency has been measured. But after nearly two decades Nvidia believes it's time for a new metric. PUE itself is a rather simple measure of efficiency. It describes how much power consumed by datacenters goes toward compute, storage, or …

[Pascal Monett]

It's Nvidia

We know Nvidia quite well. We know that Nvidia will skew any benchmark in its favor in any way it can.

Now, Nvidia is siezing on an opportunity to push forth benchmarks that would appear to be in its favor.

I'm just waiting for the inevitable day when Nvidia is going to complain that some competitor using the same benchmark is unfairly skewing it in the competitor's favor.

'Cause that is going to happen.

[Korev]

Re: It's Nvidia

Years ago I remember reading about a 10X speed increase in Blastp using a GPU; when I tried it on our own data it came back at ~10% of the speed!

[51mes]

Nvidia See a benmchmark - Own a Benchmark - Any Benchmark.

I really don't see this,

PUE is all about the efficiency of the facility i.e. the power the facility needs to provide 1Kw of capacity to whatever sits in it.

What that capacity is or does is nothing to do with the facility and therefore nothing to do with PUE. by comparing PUE you get an ideal of how efficient (sustainability) a facility is as you can remove from the facility costs any generation (solar/wind) by that facility.

There is a second level of benchmark - what is the efficiency in which my IT system can consume that power to do work is I agree of of vital importance in solution design and we do need a benchmark for it

- but surely a decision to buy a particular technology shouldn't be affected by the efficiency of the building I put it in..

[ChrisElvidge]

"The argument effectively boils down to: datacenter metrics should take into account how effectively a facility is able to convert watts into work "

But shouldn't that be "convert watts into _useful_ work"?

Is AI training really useful? Is slurping all that user data useful?