Linear algebra reservations, progress and results

[fivemack]

I think I have the reservations and completed-work synchronised now

Taking C206_129_95 (ETA morning of July 6)

[fivemack]

Originally Posted by VBCurtis

If 148^83+83^148 can get another ~60-70MQ scheduled to get raw rels over 700M, I can get the LA started before I leave.

I have stuck another 100MQ on C195_148_83

[swellman]

Reserving C205_125_117 for postprocessing. Thank you.

[fivemack]

Taking C208_138_73

ETA July 3rd

Revised ETA July 6th, because I forgot to restart it to run over the weekend when my machine crashed late on Friday night

[swellman]

Originally Posted by VBCurtis

If 148^83+83^148 can get another ~60-70MQ scheduled to get raw rels over 700M, I can get the LA started before I leave. It's a 4-node Dell C6100, for which MadPoo sent me his drawer of server DDR2 4GB sticks; two nodes have 32gb and two have 24gb, so I can run a second LA job over summer.

I'm very curious to see if a) you can now get 148_83 into LA with your hardware; and b) how long it will take to solve it. Rooting for success!

What is the minimum number of relations for a 33-but job? I had assumed it was north of 900M, but you often seem to solve 31 and 32-bit jobs with less than the typical number of relations so I'm rethinking the issue. Good luck whatever the final number is.

[VBCurtis]

Originally Posted by swellman

I'm very curious to see if a) you can now get 148_83 into LA with your hardware; and b) how long it will take to solve it. Rooting for success!

What is the minimum number of relations for a 33-but job? I had assumed it was north of 900M, but you often seem to solve 31 and 32-bit jobs with less than the typical number of relations so I'm rethinking the issue. Good luck whatever the final number is.

I solved 13*2^864-1 with 15e/33LP and needed 620M raw relations to build a TD=124 matrix. But that was SNFS-264 or so, much smaller than this candidate. 13*2^837-1 was also 15e/33LP, took 557M raw relations for ~SNFS-256; I believe that one would have been quicker as 32LP, but not by much; that factorization provides a reasonable lower bound for # of relations needed, I think- that matrix built at TD=116.
13*2^827-1, a C251, built a TD=116 matrix with 326M raw relations with 15e/32.

Two data points with small inputs isn't enough to fit a regression for # of relations vs input size; I look forward to seeing if 750M is enough for this rather large task. I'll wait a day or two to try filtering, since I may need every last relation. If I accept that 8 digits of SNFS diffculty requies 60M more relations (per the two above data points), 750M relations should be sufficient for SNFS-279ish; well short of this GNFS-194. Maybe I'll need 825M to get TD above 120...

I plan to gather more 33LP data by running some snfs-250 through NFS@home as 14e/33; I think 550-575M rels will work, and wish to demonstrate this.

[fivemack]

I have done three 33-bit GNFS jobs above 190 digits.

192.3  15e  971.4Mr/680.3Mu  39.1M  1140hrs i7/5820Kx6  td=128, td=144 failed
193.1  15e  873.0Mr/659.5Mu  34.5M  706hrs  i7/4930Kx6  td=120
196.8  16e  885.4Mr/645.7Mu  33.4M  852hrs  K10/1900x24MPI  td=120

I would be quite surprised if you can make a usable matrix with less than 850M raw relations at this level; accordingly I have pushed Q_max up to 900M.

I have also done some SNFS jobs with 33-bit large primes

272.8  15e  1064.4Mr/790.2Mu  44.2M  1550hrs i7/5820Kx6 td=160
278.9  15e   985.4Mr/742.4Mu  42.8M  1290hrs  K10/1900x48MPI  td=136, td=144 failed

I think it would be useful to record the size as well as the density of the matrices you obtained in previous work

[swellman]

Originally Posted by swellman

Reserving C205_125_117 for postprocessing. Thank you.

ETA is 19 July.

[RichD]

I'll take C215_145_52 next.

[fivemack]

Taking C198_149_50

[swellman]

Originally Posted by VBCurtis

If 148^83+83^148 can get another ~60-70MQ scheduled to get raw rels over 700M, I can get the LA started before I leave. It's a 4-node Dell C6100, for which MadPoo sent me his drawer of server DDR2 4GB sticks; two nodes have 32gb and two have 24gb, so I can run a second LA job over summer.

Any progress on 148^83+83^148?