f/fgeorge wrote:
[...]

machine is benchmarked the benchmark is so small that it can't be
accurate over the life of a workunit. So it is a guesstimate, these
are often wrong and Berkeley is working on a way to make them closer
to real life. ...
workunits as benchmarks of how long it takes your machine to crunch a
unit, it is trying to learn as it goes. This is not up to snuff yet
either.

The idea of using the WUs themselves as the "benchmark" is still just a proposal at the moment.

Paul and others have been arguing the pros and cons for some time now.

Next is for someone to program up the idea ("Calibration WUs") and try it out for real. Or at least run a simulation of the effects...

Happy crunchin',
Martin