Peter Boyle (paboyle.delete@this.ph.ed.ac.uk) on July 27, 2012 7:57 am wrote:
> The reason
> BG/Q has bigger area/flop is because it has a huge
> cache.

Sure, the difference is not just because of different clock, voltage and number of units, but also architectural on a higher level.

> BG/Q would have
> higher power efficiency AND higher area efficiency
> if it dropped the huge cache

I doubt that, because it would probably not get to the data quickly enough.

It is a bit of nonsense, but I did write: "Under the circumstances common to those designs."
What I ment is, that this factor is also throwing architecture into the kettle. It is a very general line that describes the best that is currently possible regardless of whether it is more of a CPU or GPU. If you make all the right choices you would land on that line. You may choose where, depending on your cost preference.
That line is around the best that is possible independent of the cache.
Of cause flops are not flops, it depends on the problem and the system.