Dear all:

Bill Rea just sent me a binary of Mlucas 2.7a which will run under
Solaris 2.5 - apparently a lot of the older Sparcs still run this
version of Solaris, and the binary compiled for Solaris 2.6 and 2.7
won't run under 2.5. This binary (like all the others) is available via

ftp://209.133.33.182/pub/mayer/README.html

Bill also sent me a detailed set of per-iteration timings for the
above binary on a 167MHz Sparc 1 with 512KB L2 cache. These timings
are quite interesting. For FFT lengths below 256K, the Sparc 1
behaves very similarly to an Alpha ev4. But above 256K the ev4
performance suffers somewhat of a deterioration, whereas the Sparc 1
keeps chugging onward: at 320K the Sparc timings match those of my
200MHz Alpha ev4, and above 320K they are significantly better,
despite the Sparc's slightly lower clock speed. Perhaps this is
the Sparc 1's non-blocking cache at work.

The Sparc 1 also has a near-linear iteration time trend up to very
long runlengths, and beyond 1024K has virtually identical relative
performance (normalized by clock rate) as a state-of-the-art Sparc
E450 with a much bigger 4MB L2 cache. Interesting...

The upshot is that older Sparcs can still contribute quite valuable
work for GIMPS: on a 167MHz Sparc 1 as above, double-checking an
exponent ~5M should take about 3 weeks. First-time testing an LL
exponent ~10M should take around 3 months, which is long, but still
quite a bit faster than my P166 laptop needs, and well within the
120-day limit for manual test assignments. (And such assignments
can also be extended by up to another 120 days, if need be.)

Complete timings are at ftp://209.133.33.182/pub/mayer/gimps_timings.html

Cheers,
-Ernst

p.s.: I would also appreciate some timings for SGI workstations
with pre-R10000 MIPS processors. If you have an SGI with a MIPS
3000 to 8000 version CPU and want to do some timings, please do
all the 128K-4608K FFT length timing tests listed on the README
page, and send them to me, along with the output of the Irix 'hinv'
command.

_________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ      -- http://www.tasam.com/~lrwiman/FAQ-mers

Reply via email to