lanczos error: submatrix is not invertible

[pinhodecarlos]

Just don't understand this error. LA was restarted after 3 days with a "lanczos error: submatrix is not invertible" whilst away from the machine. What's this about? Should I cancel this job and sieve more?

Sat Dec  3 23:38:13 2016  commencing Lanczos iteration (8 threads)
Sat Dec  3 23:38:13 2016  memory use: 5457.8 MB
Sat Dec  3 23:39:30 2016  linear algebra at 0.0%, ETA 190h16m
Sat Dec  3 23:39:54 2016  checkpointing every 80000 dimensions
Wed Dec  7 16:54:36 2016  lanczos error: submatrix is not invertible
Wed Dec  7 16:54:36 2016  lanczos halted after 109172 iterations (dim = 6903580)
Wed Dec  7 16:54:36 2016  linear algebra failed; retrying...
Wed Dec  7 16:54:36 2016  commencing Lanczos iteration (8 threads)
Wed Dec  7 16:54:36 2016  memory use: 5457.8 MB
Wed Dec  7 16:55:52 2016  linear algebra at 0.0%, ETA 187h32m
Wed Dec  7 16:56:15 2016  checkpointing every 80000 dimensions

Edit: Didn't think about this error algebra related. I understand its concept. What should I do next, sieve more?

[Batalov]

This error is memory related. Every time we saw it, it was due to memory ever so slightly faulty.

One thing to try is to rename *.bak.chk (if you were able to save it) to .chk and try to restart.

[pinhodecarlos]

Originally Posted by Batalov

This error is memory related. Every time we saw it, it was due to memory ever so slightly faulty.

One thing to try is to rename *.bak.chk (if you were able to save it) to .chk and try to restart.

Thank you. I've restarted it with the 3 pm GMT backup file. Let´s see what happens.

[fivemack]

Originally Posted by Batalov

This error is memory related. Every time we saw it, it was due to memory ever so slightly faulty.

Bother, I got a great swathe of this error on my three-year-old i7-4930K machine over the last couple of days.

I have noticed that I see it more when running cudalucas, giving me a slight question as to whether it's a driver problem rather than a physical defect.

(that is, I was getting it once every five minutes, and after a reboot I've got it zero times in twelve hours)