Nacho <[EMAIL PROTECTED]> wrote:
> I will have a SparcStation at work and I want to put it to do some useful
> work with GIMPS.
{snip}
> A SparcStation is not a very fast machine and I will have it only for 6
> months, so perhaps it will not be very fast to do LL test in new exponents,
> and perhaps I'll do doublecheck better.
The "best" choice here depends on the type of Sparc, how many hours per day
you'll be getting on it, and your taste with respect to assignment type.
If you like to be able to turn something in at least once a month or so,
double-checking is the way to go. DCing is also a better use of the cycles
on hardware where there is a significant relative performance drop in going
from DC to first-time-test sized exponents. Interestingly, an old Sparc 1
is often better in this last regard than a top-of-the-line E450: the Sparc 1
shows lower relative performance at DC-size exponents, but suffers little or
no performance degradation at larger runlengths. So if you don't mind waiting
3-4 months for a result and prefer doing first-time tests, go ahead and set
your old Sparc to doing one.
> Are there binaries for Mfactor for Sparc/Solaris? I found Mlucas binaries but
> not Mfactor binaries. I don't have a f90 compiler.
No Sparc binaries at present, but Mfactor is written (not by me) in ANSI C,
so just follow the compile instructions in the source code header.
One caveat, though: for sieving up to depth 2^53, Mfactor performs OK on
the Sparc because it can use the FPU to help with the double-width integer
multiplies. However, the Sparc truly sucks at integer multiply (Jason
Papadopoulos likes to say that this is the *only* thing the Sparc sucks at),
so for sieving beyond 2^53 the Sparc will be pretty awful. You'd be better
doing the factoring on a PC (or not at all) and using the Sparc for LL
testing.
> I suppose that LL results are 100% identical to Prime95 results, so LL test
> will be accepted by primenet server in manual assignment, right?
Yes. You do need to paste *only* the exponents into the Mlucas worktodo.ini
file. When your latest job finishes, copy the last line of results.txt (or
of the *.stat file corresponding to the exponent in question) into the
PrimeNet manual tests form and submit. When reserving exponents, I suggest
you ask for the maximum amount of time (120 days); if the exponent isn't done
at the end of that period, you can extend the assignment by another 120 days.
Which CPU type you report to the server when requesting work is up to you.
I usually enter "Pentium" or "PII/PPro" so as to get a reasonable amount of
work (I'm not sure what weighting factor the server uses for machines entered
as "Other".) I suspect other Mlucas users do the same, which means that non-x86
machines are underrepresented (relative to their true numbers) on PrimeNet's
world test status page.
> If I use the manual assignment pages at PrimeNet, will I receive time credits
> at Primenet status pages?
Your manual test work won't show up on the PrimeNet server's pages, but will
appear on George's master list, www.mersenne.org/top.htm and ...top2.htm ).
I'm somewhere between #200 and #300 on the top2.htm page, and 90% of that
work is from Mlucas manual tests. (Admittedly, I have access to slightly
more hardware that a single Sparc, but every contribution is valuable.)
Happy hunting,
-Ernst
_________________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ -- http://www.exu.ilstu.edu/mersenne/faq-mers.txt
