Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: Please listen to Ulrich here...



On 2013-08-12 16:24:57 +0200, Ulrich Kulisch wrote:
> Dear David,
> 
> I thank you again for your very interesting mail. It shows indeed that the
> long accumulator method is most energy efficient. Compared with computing a
> dot product by the conventional method in floating-point arithmetic it
> avoids a large number of data movements in the computer. [...]

That's probably true for this comparision. However "energy efficient"
is misleading. You must consider:

1. Does the long accumulator need energy
    * when it isn't used at all?
    * when it isn't used by an operation, but needs to retain
      its memory?

2. What would be the loss/gain of energy if the long accumulator
were replaced by some other feature (e.g. to reduce the needed
energy for accurate polynomial evaluation)?

3. How much % of energy would you gain in average applications?
I suppose that in most applications, the long accumulator would
be useless, so that the % would be too low to have an interest
for processor vendors.

-- 
Vincent Lefèvre <vincent@xxxxxxxxxx> - Web: <http://www.vinc17.net/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.net/blog/>
Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)