[Beowulf] Quasi-Non-Von-Neumann hardware in a Beowulf

[Andrew Piskorski]

On Thu, Mar 10, 2005 at 08:08:17PM -0500, Robert G. Brown wrote:

> they can help it.  However, these folks are totally happy with
> kilorands/sec, let alone megarands/sec.  I need hundreds of
> megarands/sec.

The VIA x86 cpus have "PadLock ACE" RNG hardware, but they only claim
1.5 mbit/s or so.  Lots slower than your dual Opterons pushing PRNG
code, but then, those little Vias are probably happy to get other work
done too while spitting out those random bits.  :)

Some web links say that Diehard is happy with the Via's output.  But
then I've also heard that Diehard wasn't intended (still true?) for
finding the type of non-randomness hardware generators are prone to:

  http://www.robertnz.net/hwrng.htm
  http://www.robertnz.net/true_rng.html

I wonder if the Via's underlying randomness sampling mechanism is
inherently rate limited, or if they could scale it up to a single
small RNG-only chip with 100 or 1000 times the bit rate.  There does
seem to be some market for such a thing, so since they have not done
so, probably not?

-- 
Andrew Piskorski <atp at piskorski.com>
http://www.piskorski.com/