Hi,
Turns out if you rename libcudart.so.6.0 to the 5.0 in your respective lib64,
you can pass the cudadetect for Freesurfer. However after it will crash with a
segfault error. This leads me to believe that I installed correctly.
I think Freesurfer only supports up to CUDA 5.0 so far (although I know
Freesurfer group stopped supporting the option and left the flags around). Two
options exist: Either in a future patch the program can just be enabled to call
the newer version of CUDA (or if I can find where it tries to find CUDA from,
edit it to accept any version), or to install any version of CUDA less than or
equal to 5.0
To others: The OpenMP and -use-gpu flag are quite promising. I took a recon
down from 12 hours to 3.5 hours using the -openmp 8 flag for a 4 core processor
(i7 4770k). I am currently trying to see if I can speed up faster using the
CUDA enabled option. There was a paper titled "Improving the Execution
Performance of Freesurfer" which implements a resource scheduler on top of the
CUDA and openMP option. No comment on the accuracy of the recon-all using all
of these in combination.
- Bryan
Date: Mon, 30 Jun 2014 16:34:45 +0000
From: "Chiu, Bryan (PHTH)" <bryan.c...@ubc.ca>
Subject: Re: [Freesurfer] recon-all CUDA error (Chiu, Bryan (PHTH))
To: "freesurfer@nmr.mgh.harvard.edu" <freesurfer@nmr.mgh.harvard.edu>
Message-ID:
<680d9b8d5feaaf40a1a5878265424ca44a594...@s-itsv-mbx06p.ead.ubc.ca>
Content-Type: text/plain; charset="iso-8859-1"
Hi all,
I am still working on this problem. Does anyone know if Freesurfer is asking
for CUDA 5.0, except that I have CUDA 6.0 installed?
My next steps will be to find out if Freesurfer is calling an older version of
CUDA that isn't installed, or trying the CUDA samples. I really want to get
this to work.
-Bryan
Date: Fri, 27 Jun 2014 17:59:37 +0000
From: "Chiu, Bryan (PHTH)" <bryan.c...@ubc.ca>
Subject: [Freesurfer] recon-all CUDA error
To: "freesurfer@nmr.mgh.harvard.edu" <freesurfer@nmr.mgh.harvard.edu>
Message-ID:
<680d9b8d5feaaf40a1a5878265424ca44a594...@s-itsv-mbx06p.ead.ubc.ca>
Content-Type: text/plain; charset="iso-8859-1"
Hi,
I'm fairly novice at Linux and I've been trying to setup CUDA for recon-all. I
get an error as follows:
[TLA@COGMOB-linuxPC ~]$ recon-all -s bert -use-gpu -openmp 8
Testing for CUDA device:
/usr/freesurfer/bin/mri_em_register_cuda: error while loading shared libraries:
libcudart.so.5.0: cannot open shared object file: No such file or directory
Linux COGMOB-linuxPC 2.6.32-431.el6.x86_64 #1 SMP Fri Nov 22 03:15:09 UTC 2013
x86_64 x86_64 x86_64 GNU/Linux
recon-all -s bert exited with ERRORS at Fri Jun 27 10:53:29 PDT 2014
For more details, see the log file
To report a problem, see http://surfer.nmr.mgh.harvard.edu/fswiki/BugReporting
Any idea what is going on? I setup CUDA and my Nvidia drivers according to this
tutorial
(http://aspratyush.wordpress.com/2012/05/06/install-nvidia-cuda-on-centos-6/).
With, of course some slight changes to install the latest toolkit.
Thanks in advance,
- Bryan
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.
