On 5 Mar 2001, at 18:42, Guillermo Ballester Valor wrote:
> Last weekend the number of AMD-athlons registered in Primenet has
> reached 10000. In few weeks they will be the most popular among GIMPSers
> if the actual trend continues.
Not surprising, since (a) 1.2 GHz Athlons seem to be reasonably
readily available, whereas PIII's faster than 1GHz either don't exist
or don't work; (b) at a given clock speed and running Prime95 or
mprime (code written for Pentium Pro), Athlon outperforms PIII by at
least 25%; (c) Athlon CPUs are also cheaper than PIIIs, and _much_
cheaper than P4s, whilst the cost of the other components needed to
build a system is at least no worse. (P4s require at least two
expensive RIMM modules to function at all).
>
> George, it is possible an optimization for Athlons?. I mean if it is
> worthwhile, as far as I know (I'm possibily wrong) the documentation for
> AMD-processors is no so clear as for Intel's.
>
An optimization should be possible. The cache structure is different
to the Intel model, in particular the L1 cache is enormous compared
with the Intel design; the associativity and granularity is also not
the same. It _may_ be possible to achieve around 5% improvement by
tuning memory accesses to the cache structure.
I don't know whether you'd need a different scheme for the Duron -
which is the same as the current Athlon (Thunderbird) except that it
has only 64KB of level 2 cache, and is (as yet) available only in 200
MHz FSB (technically double-data-rate clocked at 100 MHz) whereas
faster Athlons are now also available with 266 MHz FSB.
The FPU timings and throughput are different - actually the increase
in FFT performance over the Pentium is precisely because of this,
though there may be some room for further improvement by re-ordering
instructions in some of the macros.
As for documentation, I gather that Intel's published documentation
on the P4 is not that great, either ...
Having said all that, the performance gained by optimizing for Athlon
would be an improvement rather than a whole new ball game - whereas
the opposite is true in the case of the P4!
BTW I'm now running two Athlons (1.2 GHz & 650 MHz), though neither
show up in the PrimeNet statistics as they're both doing QA runs on
large exponents.
Regards
Brian Beesley
_________________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
