Mersenne Digest Tuesday, June 8 1999 Volume 01 : Number 570
----------------------------------------------------------------------
Date: Mon, 7 Jun 1999 11:01:43 +0200
From: "Steinar H. Gunderson" <[EMAIL PROTECTED]>
Subject: Mersenne: Re: Mersenne Digest V1 #568
On Sat, Jun 05, 1999 at 10:07:15PM -0700, Mersenne Digest wrote:
> Kevin Sexton
> 37% done with 5315483 after nearly finishing on v.17
If you would like credit for it, you could always contact Scott. From what
I've seen, he's nice when it comes to that part :-) (Well, nice in general...)
>If what you say is true, then whoever designed version 17 acted in a
>completely unconscionably rash manner by releasing it without thoroughly
>testing it for problems as serious as that. And has therefore shot the whole
>GIMPS effort in the foot by setting it back many weeks.
>
>Hopefully, realization of the impact on GIMPS is punishment enough to make
>sure this won't happen again...
While trying not to continue the flaming (or overreacting), I must
point out that the GIMPS QA team was formed after (and because of) the v17
bug. We have learned from our mistakes, and there is absolutely nothing more
we can do now. Being angry won't help now :-) (As for the source, it has
always been freely available, at least it was when I joined long before
PrimeNet, when my Cyrix 6x86 was counted as a 486/1353...)
Now, back to studying for my oral exam...
/* Steinar */
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 7 Jun 1999 02:12:11 -0700
From: Paul Leyland <[EMAIL PROTECTED]>
Subject: RE: Mersenne: RE: Mersenne Digest V1 #568 & prize question
> The $1500 I offered before the EFF contest began has since
> gone toward more PrimeNet hardware. Between some GIMPSers,
> a new prize pool of $1111.11 has been
> proposed (the radix is 10 :-) If you think it's worth a few
The radix is always 10. I guess you mean the radix is (1+1)*(1+1+1+1+1),
or, more concisely, (1+1+1)^(1+1) + 1.
Can anyone represent that number in fewer than (1+1+1)! ones?
;-)
Paul
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 7 Jun 99 13:46:07 CES
From: "Cornelius Caesar" <[EMAIL PROTECTED]>
Subject: Mersenne: RE: Mersenne Digest V1 #568 & prize question
Paul Leyland <[EMAIL PROTECTED]> wrote:
>> a new prize pool of $1111.11 has been
>> proposed (the radix is 10 :-) If you think it's worth a few
>
>The radix is always 10. I guess you mean the radix is (1+1)*(1+1+1+1+1),
That made me laugh! Good point...
>or, more concisely, (1+1+1)µ(1+1) + 1.
>
>Can anyone represent that number in fewer than (1+1+1)! ones?
Yes (but difficult to write in plain ASCII):
(1+1+1+1)
---------
\
\ i
/
/
---------
i = 1
or: Sum(i=(1) to (1+1+1+1)) over i.
Cornelius Caesar :-)
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 07 Jun 1999 07:51:49 -0400 (EDT)
From: "=?iso-8859-1?q?Rolf_Lochb=FChler?=" <[EMAIL PROTECTED]>
Subject: Re: Mersenne: SETI again
On Sun, 06 Jun 1999 11:31:42 -0700, Spike Jones wrote:
>I suspected that existence of SETI@Home would somehow benefit
>GIMPS in the long run, and now I see how it might happen: SETI@Home
>is a dog compared to GIMPS. It doesnt get outta your way when
>you want your computer's undivided attention, and now it appears
>they have been handing out the same assignment over and over
>[...]
... which however is their official policy as you can read at
http://setiathome.ssl.berkeley.edu/faq.html#q3.11:
[...]
3.11. Is it likely that so many people sign up that you won't
always have enough Arecibo data to feed all the clients? If so,
how will this be handled?
It's possible. Up to a point, we will handle it by sending the
same data to more than one user. Beyond that, if we can afford
it, we will set up another data recorder at Arecibo and record
a wider frequency range (our current system records only 2.5
MHz out of SERENDIP's 100 MHz bandwidth).
[...]
- --
Rolf Lochbühler
[EMAIL PROTECTED]
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 7 Jun 1999 08:05:17 -0700
From: Paul Leyland <[EMAIL PROTECTED]>
Subject: RE: Mersenne: RE: Mersenne Digest V1 #568 & prize question
> Paul Leyland <[EMAIL PROTECTED]> wrote:
>
> >> a new prize pool of $1111.11 has been
> >> proposed (the radix is 10 :-) If you think it's worth a few
> >
> >The radix is always 10. I guess you mean the radix is
> (1+1)*(1+1+1+1+1),
>
> That made me laugh! Good point...
>
> >or, more concisely, (1+1+1)µ(1+1) + 1.
> >
> >Can anyone represent that number in fewer than (1+1+1)! ones?
>
> Yes (but difficult to write in plain ASCII):
>
> (1+1+1+1)
> ---------
> \
> \ i
> /
> /
> ---------
> i = 1
>
> or: Sum(i=(1) to (1+1+1+1)) over i.
>
> Cornelius Caesar :-)
But that's cheating! You're allowed only 1 as a numeric quantity, possibly
modified with any of +-*/! and sqrt.
Paul
(Brian: forget that spurious solution I gave you. It doesn't work in
general, as I now realise.)
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 7 Jun 1999 16:51:28 +0200
From: "Steinar H. Gunderson" <[EMAIL PROTECTED]>
Subject: Mersenne: Re: Mersenne Digest V1 #569
On Mon, Jun 07, 1999 at 01:14:19AM -0700, Mersenne Digest wrote:
>I just noticed that the EFF is now offering $100,000 prize for the first
>10,000,000 digit prime. I assume that this means that they consider the
>1,000,000 digit prize essentially considered to have been claimed?
The $100,000 prize has been there all along, as somebody else pointed out.
However, somebody (can't remember who) complained that a 1,000,000,000
digit prime would not be found in a very, very long time. Heard that one
before? :-)
- ---snip---
>As someone else said, in the time it takes to find a giga-digit prime, why
>not throw a couple hundred bucks into a money market account or some other
>interest bearing fund...in the time it'd take to find a prime that big,
>you'd have already received that much in interest. :-)
Perhaps a gigadigit prime, but I think 10,000,000 should certainly be within
the limits. My poor PC is right now trying to find the first possible
10-million _exponent_... (Simple human-like algebra -- I'm sure all you
exagurus out there find a much better solution. For now, it _seems_ like
we get 3010 extra digits (possibly +1) for each 10000th iteration (of *2), but
that is of course pure guesswork. Extrapolating from that, we should get
an exponent of a little above 33 million, which is, believe it or not,
within reach. But perhaps I'm just too optimistic here, just as other people
tend to be pessimistic.)
- ---snip---
><div>+----------------------------------+</div>
><div>| Jud
>McCranie
>|</div>
><div>|
>|</div>
><div>| I have macular stars in my eyes. |</div>
><div>+----------------------------------+</div>
><br>
></html>
>
>- --=====================_233641312==_.ALT--
Again, general list stuff: Could somebody please make a HTML stripper? Anybody
with a more normal mail program can clearly see the problem here, I think...
>I wouldn't recommend the "> /dev/null" command. If you start it remotely,
>the stdout and stderr usually (but not always) go to e-mail. This way you
>can still see the output and not lose any critical messages...
mprime never closes stdout/stderr (unless terminated), so this e-mail will
never be sent :-) Better (IMHO) is using standard `nohup', because it sends
all output to a file called `nohup.out'.
More generally, perhaps we should make mprime ignore SIGHUP altogether, as
a standard? It only takes one line of code, namely:
signal(SIGHUP, SIG_IGN);
>I have heard some insider news that Intel *could* hit the 1 GigaHertz mark
>by years end if they had a reason too (if AMD jumped out with a unexpected
>surprise).
They probably could, but there is not enough market for such a chip yet, I
suppose. Those who really need that speed are buying non-Intel, and will
probably continue to do so for a while.
>Multiple processor systems are already becoming more mainstream.
You can say that again. Even the gamers (represented by id software) are
using SMP now! `Where go the gamers, so go the rest of the world.' (Picked
up on Slashdot.) This is a clear sign that it will be more normal very soon.
- ---snip---
>I also wonder how much Merced will affect coding practices. With it's EPIC
>architecture, programmers will have to rely on their wits more (the way they
>should be), hopefully getting tighter code. Of course, George already has
>his stuff about as efficient as you can get
Certainly not... Yes, true, it is extremely efficient, but it can still be
improved. (Some of this comes from Intel, who's got some real bad documentation
lying around.) When v19 comes out (it will be a while to, I believe), we will
not only get the promised 5% speedup (256K FFT, being lost in FFTs I don't
have a clue about what exponent that is), but George has also promised he
will do special optimization for P6 (PPro/PII/PIII), and wring even more
performance out of our systems...
BTW, coding for Merced will be something _entirely_ different than coding for
x86. And forcing programmers to `rely on their wits' really doesn't work, at
least it has never worked yet. A lot of programmers (including myself) spit
out extremely bad code at times.
- ---snip---
>[phma@littlecat bin]$ cat runtail
>#!/usr/bin/tclsh
If anybody are interested, I could convert this to a standard (ba)sh script.
- ---snip---
>Don't credit me with any great vision. I have repeatedly made boneheaded
>statements like "Our goal is to test all exponents below 1.3 million by
>the year 2000". Scott Kurowski and Luke Welsh have had much better vision.
Forgive me for my ignorance: the names George (Woltman) and Scott (Kurowski)
are well-known, but what does Luke (Welsh) do? Mailing-list operator? QA team
leader? I remember he's posted here, but I can't recall what his `official'
position is.
>(Heck, there's even a market still for 386 processors, they make darn good
>small office routers)
As you get P120s virtually for free, I don't think there's much money to
get for 386es now. But, of course, they are still usable (we picked up a
free 486 last year, and it's routing all it can), although not very much
as GIMPS machines.
- ---snip---
>However we can't deny having seen cash prizes, in addition to this satisfaction,
>make an observable difference in accelerating Mersenne number research results.
Can we? I'm sure you (Scott) is the one with most data on this...
/* Steinar */
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 7 Jun 99 18:55:43 CES
From: "Cornelius Caesar" <[EMAIL PROTECTED]>
Subject: Mersenne: Re: Mersenne Digest V1 #569
Steinar H. Gunderson <[EMAIL PROTECTED]> wrote:
>My poor PC is right now trying to find the first possible
>10-million _exponent_...
>For now, it _seems_ like we get 3010 extra digits (possibly +1) for
>each 10000th iteration (of *2)
Not surprising, since log 2 = 0.30102999...
You need to test exponent (10000000 / log 2) = 33219281 or above.
Cornelius Caesar
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 7 Jun 1999 11:08:23 -0600
From: "Blosser, Jeremy" <[EMAIL PROTECTED]>
Subject: RE: Mersenne: RE: Mersenne Digest V1 #568 & prize question
Is this considered cheating then?
((1+1)<<1+1)<<1
<snip>
But that's cheating! You're allowed only 1 as a numeric quantity, possibly
modified with any of +-*/! and sqrt.
Paul
(Brian: forget that spurious solution I gave you. It doesn't work in
general, as I now realise.)
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 7 Jun 1999 10:15:37 -0700
From: Paul Leyland <[EMAIL PROTECTED]>
Subject: RE: Mersenne: RE: Mersenne Digest V1 #568 & prize question
> Is this considered cheating then?
> ((1+1)<<1+1)<<1
Nice!
You're assuming a binary computer (or, equivalently, defining "<<1" as a
doubling operator) so the purists wouldn't like it.
Paul
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 07 Jun 1999 13:21:11 -0400
From: Jud McCranie <[EMAIL PROTECTED]>
Subject: Re: Mersenne: Re: Mersenne Digest V1 #569
- --=====================_320232889==_.ALT
Content-Type: text/plain; charset="us-ascii"
At 04:51 PM 6/7/99 +0200, Steinar H. Gunderson wrote:
>Perhaps a gigadigit prime, but I think 10,000,000 should certainly be within
>the limits.
I was thinking about it informally the other day, and I thought that I expected
to see 10,000,000-digit primes and 100,000,000-digit primes in my lifetime, but
probably not 1,000,000,000-digit primes. Then I looked at the historical info
about the largest known prime
http://www.utm.edu/research/primes/notes/by_year.html and a billion-digit prime
has a decent chance of being discovered in my lifetime. The prediction of
2009, however, seems much overly optimistic to me.
>exagurus out there find a much better solution. For now, it _seems_ like
>we get 3010 extra digits (possibly +1) for each 10000th iteration (of *2), but
>that is of course pure guesswork.
It is a fact that each additional bit adds log10(2) digits = 0.30103.
+----------------------------------------------+
| Jud "program first and think later" McCranie |
+----------------------------------------------+
- --=====================_320232889==_.ALT
Content-Type: text/html; charset="us-ascii"
<html>
<font size=3>At 04:51 PM 6/7/99 +0200, Steinar H. Gunderson wrote:<br>
<br>
>Perhaps a gigadigit prime, but I think 10,000,000 should certainly be
within<br>
>the limits.<br>
<br>
I was thinking about it informally the other day, and I thought that I
expected to see 10,000,000-digit primes and 100,000,000-digit primes in
my lifetime, but probably not 1,000,000,000-digit primes. Then I
looked at the historical info about the largest known prime<br>
<a href="http://www.utm.edu/research/primes/notes/by_year.html"
eudora="autourl">http://www.utm.edu/research/primes/notes/by_year.html</a>
and a billion-digit prime has a decent chance of being discovered in my
lifetime. The prediction of 2009, however, seems much overly optimistic to
me.<br>
<br>
<br>
>exagurus out there find a much better solution. For now, it _seems_ like<br>
>we get 3010 extra digits (possibly +1) for each 10000th iteration (of *2), but<br>
>that is of course pure guesswork. <br>
<br>
It is a fact that each additional bit adds log10(2) digits = 0.30103.<br>
<br>
<br>
</font><br>
<div>+----------------------------------------------+</div>
<div>| Jud "program first and think later" McCranie |</div>
<div>+----------------------------------------------+</div>
<br>
</html>
- --=====================_320232889==_.ALT--
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 7 Jun 1999 14:37:22 -0400
From: "Ernst W. Mayer" <[EMAIL PROTECTED]>
Subject: Mersenne: Re: algorithmic improvements
George Woltman wrote:
>As recently 6 years ago, Crandall and Fagin published
>the details for doubling the speed of the test. Perhaps there are
>more theoretical or algorithmic improvements to come. (Anyone on this
>list working on that right now??)
Does 50+ bits per IEEE 64-bit float count as an algorithmic improvement?
The old:
1000 iterations of M 54011 with FFT length 3072
Res64: E4B9382BD681646B. Program: E2.6x
Clocks = 00:00:05.135
5.02u 0.09s 0:05 98% 0+5k 0+7io 0pf+0w
The new:
1000 iterations of M 54011 with FFT length 1024
Res64: E4D9382BD681646B. Program: E3.0x
Clocks = 00:00:07.760
7.38u 0.18s 0:07 96% 0+4k 0+0io 0pf+0w
Of course, cutting that 7.38s CPU time by a factor of 3-4 is a challenge-
if it weren't, it wouldn't be fun!
Cheers,
Ernst
p.s.: And no, I'm not cheating and using the nonstandard real*16 data type.
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 7 Jun 1999 18:42:33 +0000 (GMT)
From: Henrik Olsen <[EMAIL PROTECTED]>
Subject: RE: Mersenne: EFF and 10,000,000 digits
On Sun, 6 Jun 1999, Jud McCranie wrote:
> At 11:30 AM 6/6/99 -0600, Aaron Blosser wrote:
>
> >I suppose it depends on whether Moore's Law can continue to hold true. I'm
> >not so sure that we can keep doubling speeds of processors every 18 months
> >as predicted...
>
>
> That's often stated, but it hasn't been holding true. We are a factor of about
> 65,000 short of where we should be - if it was true. Moore actually said "18
> to 24 months", but most everyone says "18". Anyhow, I did a test on it from
> 1971 to 98, and it actually averaged about 26 months for the doubling time (and
> it was slowing).
Was that doubling based on clock speeds, or actual processing power?
Clockspeeds are useless for comparing processors, even among processors of
the same type.
Case in point, I've seen a 350MHz P-II system that's more than 1.5 times
as fast as a 300MHz P-II system for NFS calculations, because the former
is a 100MHz*3.5, and the latter is 66MHz*4.5, making the memory access for
the first system a lot faster.
- --
Henrik Olsen, Dawn Solutions I/S URL=http://www.iaeste.dk/~henrik/
Do not meddle in the affairs of dragons,
for you are crunchy and taste good with ketchup.
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 07 Jun 1999 19:43:59 GMT
From: [EMAIL PROTECTED] (Steven Whitaker)
Subject: Re: Mersenne: SETI again
On Mon, 07 Jun 1999 07:51:49 -0400 (EDT), you wrote:
<SNIPPED>
>
>[...]
>3.11. Is it likely that so many people sign up that you won't
>always have enough Arecibo data to feed all the clients? If so,
>how will this be handled?
>
>It's possible. Up to a point, we will handle it by sending the
>same data to more than one user. Beyond that, if we can afford
>it, we will set up another data recorder at Arecibo and record
>a wider frequency range (our current system records only 2.5
>MHz out of SERENDIP's 100 MHz bandwidth).
>[...]
I thought the problem was not that they had run out of data but there
was a problem with the server.
According to their main page:
'Our "data pipeline" is not flowing at top speed yet, so we're sending
out the same work units (mostly recorded Jan 7 and Jan 8) repeatedly.
This will be fixed shortly'.
Not sure what they mean by that.
My desktop PC has gone back to finding factors.
By the way, this discussion doesn't really belong on the Mersenne
list. Is there a SETI mailing list or has anyone thought about setting
one up?
- --
Steve Whitaker ([EMAIL PROTECTED])
Egham, Surrey, UK
"Start the day with a smile and get it over with" - WC Fields
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 7 Jun 1999 22:03:29 +0200
From: "Steinar H. Gunderson" <[EMAIL PROTECTED]>
Subject: Re: Mersenne: Re: Mersenne Digest V1 #569
On Mon, Jun 07, 1999 at 02:11:01PM -0400, Pierre Abbat wrote:
>I think it would be better to act like a daemon and reread its configuration
>file when hung up on. That way we wouldn't have to stop it, start it
>interactively, then start it again in the background to tell it when we go on
>vacation.
Any way, just not die silently. (I run it in the foreground -- it gives me
satisfaction to see a `3%' increase in the morning...)
/* Steinar */
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 7 Jun 1999 16:37:32 -0500
From: JON STRAYER <[EMAIL PROTECTED]>
Subject: RE: Mersenne: Re: Mersenne Digest V1 #569
I'm I correct in thinking that since the exponent is five times the size of
current exponents that the LL test will take 25 times as long (give the same
processor)?
> You need to test exponent (10000000 / log 2) = 33219281 or above.
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 7 Jun 1999 16:39:31 -0700 (PDT)
From: "J. Williams" <[EMAIL PROTECTED]>
Subject: Mersenne: Prime 95 Error Messages/ Misc
Hi, I am running this and received the following message:
Error: Illegal Sumout
Is this anything I need to be concerned about and is there a listing of
such messages?
Also, a while ago the program reserved about 20 numbers for testing (all
the way into the year 2001) then it and released them a few minutes later.
Is this normal operation?
Any tips for optimization and usage?
Finally, when and why does it communicate with the server (besides getting
new numbers to test)?
Thanks, John
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 07 Jun 1999 19:25:51 -0500
From: Gary Diehl <[EMAIL PROTECTED]>
Subject: Re: Mersenne: Prime 95 Error Messages/ Misc
I also got the "illegal sumout" and it locked up my machine. It did it
four times in a row at the 6564090 iteration (the lock ups were hard
locks - had to recycle power to recover).
However, I had my pentium-II 266 overclocked to 333, and when I reset
the speed back to 266 the problem went away. I have had it overclocked
for about 5 months now with no problems. I have been running prime95 at
333mhz since 16 May 99. After resetting the speed to 266 I went back to
an older backup file (and lost about 4 hours work) just to make sure I
am still working with good data. Interesting to note -- the average
iteration speed at idle (0.275 sec) remained unchanged regardless of the
CPU speed change. Does overclocking even help?
Maybe this error is a bug, maybe a hardware error, although the latter
seems more suspect in my case.
Gary Diehl
"J. Williams" wrote:
>
> Hi, I am running this and received the following message:
>
> Error: Illegal Sumout
>
> Is this anything I need to be concerned about and is there a listing of
> such messages?
>
> Also, a while ago the program reserved about 20 numbers for testing (all
> the way into the year 2001) then it and released them a few minutes later.
> Is this normal operation?
>
> Any tips for optimization and usage?
>
> Finally, when and why does it communicate with the server (besides getting
> new numbers to test)?
>
> Thanks, John
>
> ________________________________________________________________
> Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Tue, 8 Jun 1999 01:17:14 +0100
From: Tony Forbes <[EMAIL PROTECTED]>
Subject: Re: Mersenne: RE: Mersenne Digest V1 #568 & prize question
Paul Leyland <[EMAIL PROTECTED]> writes
>
>> Paul Leyland <[EMAIL PROTECTED]> wrote:
>>
>> >> a new prize pool of $1111.11 has been
>> >> proposed (the radix is 10 :-) If you think it's worth a few
>> >
>> >The radix is always 10. I guess you mean the radix is
>> (1+1)*(1+1+1+1+1),
>>
>> That made me laugh! Good point...
>>
>> >or, more concisely, (1+1+1)(1+1) + 1.
>> >
>> >Can anyone represent that number in fewer than (1+1+1)! ones?
>>
>> Yes (but difficult to write in plain ASCII):
>>
>> (1+1+1+1)
>> ---------
>> \
>> \ i
>> /
>> /
>> ---------
>> i = 1
>>
>> or: Sum(i=(1) to (1+1+1+1)) over i.
>>
>> Cornelius Caesar :-)
>
>But that's cheating! You're allowed only 1 as a numeric quantity, possibly
>modified with any of +-*/! and sqrt.
>
How about
[sqrt([sqrt(sqrt((1+1+1)!!))]!)],
where the square brackets [x] denotes, as usual, the integer part of x.
- --
Tony
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 7 Jun 1999 20:19:20 -0600
From: "Aaron Blosser" <[EMAIL PROTECTED]>
Subject: RE: Mersenne: Prime 95 Error Messages/ Misc
> I also got the "illegal sumout" and it locked up my machine. It did it
> four times in a row at the 6564090 iteration (the lock ups were hard
> locks - had to recycle power to recover).
>
> However, I had my pentium-II 266 overclocked to 333, and when I reset
> the speed back to 266 the problem went away. I have had it overclocked
> for about 5 months now with no problems. I have been running prime95 at
> 333mhz since 16 May 99. After resetting the speed to 266 I went back to
> an older backup file (and lost about 4 hours work) just to make sure I
> am still working with good data. Interesting to note -- the average
> iteration speed at idle (0.275 sec) remained unchanged regardless of the
> CPU speed change. Does overclocking even help?
>
> Maybe this error is a bug, maybe a hardware error, although the latter
> seems more suspect in my case.
I've seen this happen a lot on computers that are either overclocked or just
have a faulty CPU/memory module.
For instance, I've got a nice IBM PPro 200 machine that's been running
NTPrime for nearly a year now, rock solid, nary a problem. On a whim, I
thought I'd see if the machine could handle 233MHz. Everything else ran
okay, NT, Office, etc. but NTPrime started kicking out errors like a
banshee.
Reverted to a backup I had made (I was smart...backed up the whole prime
directory before doing anything...if anything I'm over-cautious) to make
sure I didn't corrupt any results...set the speed back to 200MHz and no
problems since.
I did notice in the prime.log that when errors occur, that info is sent to
the Primenet server.
On that note, Scott...what becomes of info sent indicating errors in the
calculations? Are those exponents flagged in someway, indicating that they
are "suspect"?
Anyway, long and short is that overclocking might seem to work fine, but a
really CPU intensive program like NTPrime/Prime95 is likely to show
problems. Heck, it's a great way to find out if your overclocked system is
really working as well as you thought...just back up any temp files
beforehand so you can revert back to them when/if you start getting errors.
Personally, I'm glad we double-check numbers because some errors are bound
to slip by due to overclocked systems not working right. I've decided that
none of the systems I run NTPrime on should be overclocked, to maintain the
integrity of the results as much as possible.
Aaron
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Tue, 8 Jun 1999 03:36:20 +0000 (GMT)
From: Henrik Olsen <[EMAIL PROTECTED]>
Subject: Re: Mersenne: Prime 95 Error Messages/ Misc
On Mon, 7 Jun 1999, Gary Diehl wrote:
> I also got the "illegal sumout" and it locked up my machine. It did it
> four times in a row at the 6564090 iteration (the lock ups were hard
> locks - had to recycle power to recover).
>
> However, I had my pentium-II 266 overclocked to 333, and when I reset
> the speed back to 266 the problem went away. I have had it overclocked
> for about 5 months now with no problems. I have been running prime95 at
> 333mhz since 16 May 99. After resetting the speed to 266 I went back to
> an older backup file (and lost about 4 hours work) just to make sure I
> am still working with good data. Interesting to note -- the average
> iteration speed at idle (0.275 sec) remained unchanged regardless of the
> CPU speed change. Does overclocking even help?
It does help but you have to remember that the reported time per iteration
is gotten by counting clockcycles, then adjusting for the speed you
yourself said your processor was running at.
My guess is that you didn't change the cpu speed you'd told prime95 when
you lowered the frequency, and since the problem takes the same number of
clockcycles, it's reported as taking the same time.
- --
Henrik Olsen, Dawn Solutions I/S URL=http://www.iaeste.dk/~henrik/
Do not meddle in the affairs of dragons,
for you are crunchy and taste good with ketchup.
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 07 Jun 1999 21:19:46 -0700
From: Luke Welsh <[EMAIL PROTECTED]>
Subject: Re: Mersenne: Re: Mersenne Digest V1 #569
At 04:51 PM 6/7/99 +0200, Steinar H. Gunderson wrote:
>>Scott Kurowski and Luke Welsh have had much better vision.
[...]
>Forgive me for my ignorance: the names George (Woltman) and Scott
>(Kurowski) are well-known, but what does Luke (Welsh) do?
M29 and 2nd member of GIMPS.
>but I can't recall what his `official' position is.
I named GIMPS. Does that make me the godfather?
- --Luke, who has such great vision that he lost a bet and
owes George a dinner (prime rib, of course)
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Mon, 7 Jun 1999 23:42:19 -0700
From: "Ethan Hansen" <[EMAIL PROTECTED]>
Subject: RE: Mersenne: Prime 95 Error Messages/ Misc
General comments on overclocking and Prime95:
If you're overclocking a CPU, it is a good idea to run the full Prime95 self
test (Options menu) to ensure that the CPU is stable and functional for all
FFT lengths. This is especially important if you are using a Xeon
processor, as there are interesting cache functionality problems that only
appear when a certain percentage of the die is used. Note: This behavior
appears even more often on the latest Intel Coppermine processors -- only
Xeon-like configurations checked so far.
It is also possible to have an overclocked CPU pass the full self test
suite, but later exhibit problems. The likely culprit is simple wearout --
the CPU initially was barely functional at the overclocked speed, but slowed
enough that it no longer runs. If this happens, you usually can still run
the CPU at the rated speed.
Regards,
Ethan
> -----Original Message-----
> From: [EMAIL PROTECTED]
> On Behalf Of Aaron Blosser
>
> > I also got the "illegal sumout" and it locked up my machine. It did it
> > four times in a row at the 6564090 iteration (the lock ups were hard
> > locks - had to recycle power to recover).
> >
> > However, I had my pentium-II 266 overclocked to 333, and when I reset
> > the speed back to 266 the problem went away. I have had it overclocked
> > for about 5 months now with no problems. I have been running prime95 at
> > 333mhz since 16 May 99. After resetting the speed to 266 I went back to
> > an older backup file (and lost about 4 hours work) just to make sure I
> > am still working with good data. Interesting to note -- the average
> > iteration speed at idle (0.275 sec) remained unchanged regardless of the
> > CPU speed change. Does overclocking even help?
> >
> > Maybe this error is a bug, maybe a hardware error, although the latter
> > seems more suspect in my case.
>
> I've seen this happen a lot on computers that are either
> overclocked or just
> have a faulty CPU/memory module.
>
> For instance, I've got a nice IBM PPro 200 machine that's been running
> NTPrime for nearly a year now, rock solid, nary a problem. On a whim, I
> thought I'd see if the machine could handle 233MHz. Everything else ran
> okay, NT, Office, etc. but NTPrime started kicking out errors like a
> banshee.
[SNIP]
> Anyway, long and short is that overclocking might seem to work fine, but a
> really CPU intensive program like NTPrime/Prime95 is likely to show
> problems. Heck, it's a great way to find out if your overclocked
> system is
> really working as well as you thought...just back up any temp files
> beforehand so you can revert back to them when/if you start
> getting errors.
>
[SNIP]
>
> Aaron
>
> ________________________________________________________________
> Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
>
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Tue, 8 Jun 1999 08:59:02 GMT
From: "Brian J Beesley" <[EMAIL PROTECTED]>
Subject: RE: Mersenne: Prime 95 Error Messages/ Misc
"Ethan Hansen" <[EMAIL PROTECTED]> writes:
> This is especially important if you are using a Xeon
> processor, as there are interesting cache functionality problems that only
> appear when a certain percentage of the die is used.
Eh?
256K FFT = 2Mbyte work vector. Two copies needed, plus odds &
sods of other memory (including executable code). There are going
to be cache misses, even on a 2MB L2 cache Xeon, for any FFT
size >= 128K. Which means all current LL test assignments.
IMHO if you have a Xeon (or a PII/PIII > 400 MHz, for that matter)
you would probably be better running LL test assignments, leave
the double checking for people with older processors.
> It is also possible to have an overclocked CPU pass the full self test
> suite, but later exhibit problems. The likely culprit is simple wearout --
> the CPU initially was barely functional at the overclocked speed, but slowed
> enough that it no longer runs. If this happens, you usually can still run
> the CPU at the rated speed.
Or, the errors are there all the time but at a low rate e.g. on
average 1 every day. This is very likely to pass the 1 hour "self
test" but will show up in actual use, or on the continuous "torture
test" (see the "Options" menu in Prime95).
If you've overclocked your system at all (_not_ reccomended, but I
know it can be successful in some cases) then I suggest you let
the torture test run for a couple of days before committing yourself
to doing "real" work. It can happen that an overclocked system
appears to run fine for "office" applications but causes problems
with Prime95 because very few other applications use the floating-
point unit even half as intensively as Prime95 does.
Note that simple overheating can also cause problems, even if your
system _isn't_ overclocked. Might be an idea to check that case
and processor cooling fans are operating. They have been known to
fail!
Finally (I think this is right - I'm sure George will chip in if not) there
is a small but finite chance that you could get a very occasional
"sum out error" even if your system is 100% perfect. This is due to
abnormal combinations of data in the FFT triggering the "sanity
check" in the code; the result may well be OK. The program should
check that the "error" is "deterministic" (repeatable) rather than
random and continue automatically if it is - though there will be an
error log entry - PrimeNet uses this information to flag the result as
"suspect", the exponent should then be re-assigned for an early
double-check instead of waiting its turn as usual.
Regards
Brian Beesley
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Tue, 8 Jun 1999 09:10:08 +0100
From: Nick Craig-Wood <[EMAIL PROTECTED]>
Subject: Re: Mersenne: Re: Mersenne Digest V1 #569
On Mon, Jun 07, 1999 at 04:37:32PM -0500, JON STRAYER wrote:
> I'm I correct in thinking that since the exponent is five times the
> size of current exponents that the LL test will take 25 times as
> long (give the same processor)?
>
> > You need to test exponent (10000000 / log 2) = 33219281 or above.
Time for an n bit multiply using DWT is O(n log n) plus some fiddly
bits to do with decreasing precision available when using longer
convolutions - I'd probably estimate this as O(log log n). As it
takes n iterations for each test then to test an n digit number using
the Lucas Lehmer test with DWT takes O(n^2 . log n . log log n).
If we say that the current exponent is about 6.5 million then we need
to test one which is approx 5 times bigger. Using the above formula I
make this 29.97 times longer for a 5 times increase in exponent size
starting at 6.5 million.
Probably ;-)
- --
Nick Craig-Wood
[EMAIL PROTECTED]
http://www.axis.demon.co.uk/
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
Date: Tue, 8 Jun 1999 12:34:36 +0200
From: "Steinar H. Gunderson" <[EMAIL PROTECTED]>
Subject: Mersenne: Re: [First 10 million digit exponent]
On Tue, Jun 08, 1999 at 11:53:30AM +0200, Alex Kruppa wrote:
>The math is really quite simple: a number n has log_10(n) (logarithm with base 10)
>decimal digits or ld(n) (ld = log_2 = logarithm with base 2) bits.
>You can do a base conversion between logarithms from base a to base b by
>dividing log_a(n) by log_a(b). The log_10(2) =~ 0,301 =~ 1 / 3.3219.
>So if you want to know the number of bits in a 10,000,000 digit number,
>you do 10,000,000 / log_10(2) =~ 10,000,000 * 3,3219 = 33,219,000.
Looks like I shouldn't have asked this question, I'm getting flooded
with replies here :-) The exact formula (as Brian pointed out) is:
ceil((x-1)/log10(2))
Set x=10000000, and you get an exponent of 33,219,278 (at least that's
what gcc and my assembly program tells me, Brian found it to be 33,219,277).
>The first prime exponent Mersenne with 10,000,000 digits is M33219281.
And below M36000000, there are 159,975 exponents (again repeating Brian)
with at least 10 million digits.
/* Steinar */
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
------------------------------
End of Mersenne Digest V1 #570
******************************
