2- table

[bdodson]

Originally Posted by bdodson

Another "small" prime factor, completely different meaning. And a
very short wait. From M1069 = 2,1069- C306, p50 = ...

leaving a c256. I'm adding .5*t50 to c300-c319 (having added t50 to
c320-c384, for 2*t50), ...

Wow!, an even shorter wait, from M1177 = 2,1177- c306,

p51 = 355693251295001123842782330386351058608227381732783

leaving a c255. The numbers for which the 1st pass of .5*t50 with
B1 = 110M is done using gmp-ecm's base-2 arithmetic are getting a 2nd
pass of .5*t50, and may perhaps get some more (along with M1061
testing), with B1=260M, which fits in memory for the quadcores, but
doesn't for the Opterons. Actually, I double checked, looks like M1193
c355 may not quite fit in 2Gb ... that's 16Gb per 8 core node ... and
some early test cases towards a 3rd t50 crashed; not sure whether
it was due to memory conflicts.

There's another report today over on the 7- thread. -Bruce

[fivemack]

I agreed that, if akruppa sieved 2^841-1 on the French supercomputer that he had access to, I would do the linear algebra.

Sadly, the machine which I built to do this on is unusably unreliable; I've tried several times, and the job fails after 24 to 48 hours.

Does anyone else (Joppe is the person I'm immediately thinking of) have a single machine with 8G of memory (6G would also suffice, but it's not going to fit in 4G) and a reasonable number of reasonably fast processors that they would be prepared to devote to finishing 2^841-1?

I estimated the job at 700 hours (almost exactly one month) on a 2.4GHz quad-core core2; it is checkpointable, but it's not going to finish in reasonable time on a machine usable only at weekends. The machine is usable while the msieve job is running, though obviously you don't want to run vastly CPU-intensive tasks.

The data's large enough that the postal system is the only convenient network with enough bandwidth; I can send either the matrix (one DVD) or the relations (four DVDs).

[Wacky]

Tom,

Hardware Overview:

Model Name: Mac Pro
Model Identifier: MacPro1,1
Processor Name: Dual-Core Intel Xeon
Processor Speed: 2.66 GHz
Number Of Processors: 2
Total Number Of Cores: 4
L2 Cache (per processor): 4 MB
Memory: 8 GB
Bus Speed: 1.33 GHz
Boot ROM Version: MP11.005D.B00
SMC Version: 1.7f10
Serial Number: G**********

[jasonp]

Wacky's machine would be significantly faster, but I have a dual-CPU (not dual core) 2GHz opteron with 6GB of memory that I could volunteer.

[Andi47]

Originally Posted by fivemack

Sadly, the machine which I built to do this on is unusably unreliable; I've tried several times, and the job fails after 24 to 48 hours.

Did you try to check the points given in this posting (quoted below)?

Originally Posted by garo

Usual crap. Eliminate the variables one at a time:

1. Memory
2. Heat
3. Overclock if any stressing either of the above or the FSB.
4. Motherboard
5. CPU

Unfortunately, there is no short cut.

[xilman]

Originally Posted by fivemack

I agreed that, if akruppa sieved 2^841-1 on the French supercomputer that he had access to, I would do the linear algebra.

Sadly, the machine which I built to do this on is unusably unreliable; I've tried several times, and the job fails after 24 to 48 hours.

Does anyone else (Joppe is the person I'm immediately thinking of) have a single machine with 8G of memory (6G would also suffice, but it's not going to fit in 4G) and a reasonable number of reasonably fast processors that they would be prepared to devote to finishing 2^841-1?

I estimated the job at 700 hours (almost exactly one month) on a 2.4GHz quad-core core2; it is checkpointable, but it's not going to finish in reasonable time on a machine usable only at weekends. The machine is usable while the msieve job is running, though obviously you don't want to run vastly CPU-intensive tasks.

The data's large enough that the postal system is the only convenient network with enough bandwidth; I can send either the matrix (one DVD) or the relations (four DVDs).

I could do this one too. I've a quad-core Xeon and a dual quad-core Xeon, each with 8G memory.

Paul

[Joppe_Bos]

Originally Posted by fivemack

I agreed that, if akruppa sieved 2^841-1 on the French supercomputer that he had access to, I would do the linear algebra.

Sadly, the machine which I built to do this on is unusably unreliable; I've tried several times, and the job fails after 24 to 48 hours.

Does anyone else (Joppe is the person I'm immediately thinking of) have a single machine with 8G of memory (6G would also suffice, but it's not going to fit in 4G) and a reasonable number of reasonably fast processors that they would be prepared to devote to finishing 2^841-1?

I estimated the job at 700 hours (almost exactly one month) on a 2.4GHz quad-core core2; it is checkpointable, but it's not going to finish in reasonable time on a machine usable only at weekends. The machine is usable while the msieve job is running, though obviously you don't want to run vastly CPU-intensive tasks.

The data's large enough that the postal system is the only convenient network with enough bandwidth; I can send either the matrix (one DVD) or the relations (four DVDs).

Already some people volunteered but I would also like to help out. If you haven't sent a DVD already you can use my address I gave you earlier (although the 8 GB machine here is a Core 2 Duo).

[fivemack]

Thanks very much for all your offers; I've agreed to send the data to Wacky, but I now know that there is lots of support out there if I can't get the blasted computer to work by the next time I want to do some serious processing.

[xilman]

Originally Posted by fivemack

Does anyone else (Joppe is the person I'm immediately thinking of) have a single machine with 8G of memory (6G would also suffice, but it's not going to fit in 4G) and a reasonable number of reasonably fast processors that they would be prepared to devote to finishing 2^841-1?

I estimated the job at 700 hours (almost exactly one month) on a 2.4GHz quad-core core2;

As a matter of interest, what's the size and weight of the matrix? I've calibrated a number of my machines quite well and could probably make an estimate good to 10% of the running time on my systems.

Paul

[jasonp]

Originally Posted by xilman

As a matter of interest, what's the size and weight of the matrix? I've calibrated a number of my machines quite well and could probably make an estimate good to 10% of the running time on my systems.

It's a monster, nearly the size of the matrix for 2,811-:

matrix is 13558193 x 13558441 (3675.6 MB) with weight 906834632 (66.88/col)
sparse part has weight 827949855 (61.07/col)

[xilman]

Originally Posted by jasonp

It's a monster, nearly the size of the matrix for 2,811-:

matrix is 13558193 x 13558441 (3675.6 MB) with weight 906834632 (66.88/col)
sparse part has weight 827949855 (61.07/col)

Ok, so relatively low density. I'm more accustomed to weights per column/row of 70-80.

My estimate is that my machine would take about 250-280 hours to solve it. This is based on a (very good) estimate of 44.4 hours for a 5.71M matrix with weight 69.9 per column. The difference between 280 hours and the previously quoted 700 hours is sufficiently great that I'm tempted to ask for a copy to see whether my estimate is correct.

Is the matrix available on a generally accessible ftp site? Unfortunately, I can only deal with CWI format matrices or relations at the moment. Otherwise I'll have to write conversion software which could well take more than 400 hours elapsed, as it would have to be done in otherwise over-allocated spare time.

Paul