Anyone have an ETA for M1061?

[pinhodecarlos]

Originally Posted by Dubslow

How much memory?

39725.3 MB

[fivemack]

Originally Posted by Dubslow

How much memory?

Nearly $500 worth (though probably you should pay $800 and get the full 64G); memory is amazingly cheap nowadays and you don't even need an absurd server board to fit that much.

[fivemack]

Originally Posted by frmky

LA for M1061 is underway:

Fri Jun 1 10:06:56 2012 matrix is 90270707 x 90270879 (39725.3 MB) with weight 11308345602 (125.27/col)
Fri Jun 1 10:06:56 2012 sparse part has weight 9420766130 (104.36/col)
Fri Jun 1 18:27:17 2012 linear algebra at 0.1%, ETA 663h48m

This will take a bit longer than that thanks to waiting in cluster queues, but I'm hoping early July.

Awesome. 16x16 grid? (90M in 663h vs my 33M in 850h suggests you have ten times my 24 processors, or perhaps ten times my 16 DDR3 channels ...)

[pinhodecarlos]

Originally Posted by fivemack

Awesome. 16x16 grid? (90M in 663h vs my 33M in 850h suggests you have ten times my 24 processors, or perhaps ten times my 16 DDR3 channels ...)

Greg said this on NFS@home forum

The postprocessing for 2,1061- is a bit complicated. I anticipate that a portion of it will be run on Trestles at the San Diego Supercomputing Center, and that a portion will be run on Lonestar at the Texas Advanced Computing Center. It will likely use 576 cores, which is 18 32-core nodes at Trestles and 48 12-core nodes at Lonestar.

http://escatter11.fullerton.edu/nfs/...owrap=true#886

[frmky]

Yes, the scaling isn't linear. It's currently running at SDSC Trestles on 18 32-core AMD Magny-Cours computers in a 24x24 grid.

[chalsall]

Originally Posted by frmky

Yes, the scaling isn't linear. It's currently running at SDSC Trestles on 18 32-core AMD Magny-Cours computers in a 24x24 grid.

Cool.

Just wondering... Although it would be "cool" having M1061 factored, is there any mathematical or scientific benefit from doing so?

A sincere question.

[jasonp]

Greg wanting to run large NFS postprocessing jobs on big iron led to IMO the first highly parallel NFS linear algebra code that was publicly available. That code has also been incorporating research from Ilya Popovyan that has made it much faster.

Plus, factoring M1061 will finally shut up all the people here who keep asking about it :)

[chalsall]

Originally Posted by jasonp

Greg wanting to run large NFS postprocessing jobs on big iron led to IMO the first highly parallel NFS linear algebra code that was publicly available. That code has also been incorporating research from Ilya Popovyan that has made it much faster.

Cool.

So then the answer is, there is no actual direct benefit from factoring M1061, other than to have "driving problems" which require the development of new systems and software which are able to do so.

I say again: Cool.

[Dubslow]

Originally Posted by jasonp

Greg wanting to run large NFS postprocessing jobs on big iron led to IMO the first highly parallel NFS linear algebra code that was publicly available. That code has also been incorporating research from Ilya Popovyan that has made it much faster.

Is it safe to say that Msieve is the only software out there that could handle this post processing?

[Batalov]

Originally Posted by Dubslow

Is it safe to say that Msieve is the only software out there that could handle this post processing?

No.

[KingKurly]

Originally Posted by chalsall

Cool.

So then the answer is, there is no actual direct benefit from factoring M1061, other than to have "driving problems" which require the development of new systems and software which are able to do so.

I say again: Cool.

Yup. The main interest is that M1061 is currently the smallest known-composite Mersenne number for which there is no known factor.