From: Norman Geist (norman.geist_at_uni-greifswald.de)
Date: Thu Jan 17 2013 - 00:01:27 CST
this was easy because it didn't crash without one specific GPU (it was a new
one, so it was likely because the error occurred from there on), I'm lucky
enough to have more than one machine. Additionally, it was only this
specific simulation system, others were running without problems. Sometimes,
there can be weird cases that will run into such problems, while others just
Von: Ajasja Ljubetiè [mailto:ajasja.ljubetic_at_gmail.com]
Gesendet: Mittwoch, 16. Januar 2013 17:04
An: Norman Geist
Cc: Eric Hill; Namd Mailing List
Betreff: Re: namd-l: Re: AW: AW: CUDA problem?
IMHO memcheck doesn't say much about hardware issues on a GPU as it only
checks the vram not the GPU itself and not the communication across the
pcie. When I had a broken GPU, everything looked fine. No errors no display
problems. Also the simulations where running until they crashed with "atoms
moving too fast" error". Memcheck told nothing aswell.
How did you then diagnose the problem was the GPU? I'm asking, because I
have some simulations that crash only on the GPU and only with ABF.
This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:20:51 CST