i've just posted the file 'fscudabins-linux-centos4.tgz' here:
ftp://surfer.nmr.mgh.harvard.edu/pub/dist/freesurfer/misc/linux-centos4/
note that we have recently discovered a new problem with running mri_em_register_cuda on brains with skull: the results are off compared to the cpu version. so updates to that will be available next week sometime.
n.
On Fri, 2010-08-20 at 16:07 -0400, Gonzalo Rojas Costa wrote:
Hi:
Any solution for the 32 bit version ?...
Sincerely,
Gonzalo Rojas Costa
Message: 6 Date: Fri, 20 Aug 2010 15:55:39 -0400 From: Nick Schmansky nicks@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] CUDA Error - all CUDA-capable devices are busy or unavailable To: Daniel Guellmar Daniel.Guellmar@med.uni-jena.de Cc: freesurfer@nmr.mgh.harvard.edu Message-ID: 1282334139.31414.17.camel@terrier.nmr.mgh.harvard.edu Content-Type: text/plain; charset=utf-8
CUDA users,
The problem with the _cuda binaries (message: CUDA Error in file 'devicemanagement.cu'...) has been fixed, and new binaries are available for download from here:
ftp://surfer.nmr.mgh.harvard.edu/pub/dist/freesurfer/misc/linux-centos4_x86_64/
download the file 'fscudabins-linux-centos4_x86_64.tgz' and extract it (tar zxvf <filename>) into your $FREESURFER_HOME/bin directory. This is for the 64b linux platform only (centos 4 and 5).
fyi, the problem was that the utility UPX, which is run on the publicly distributed freesurfer binaries to reduce their size, seemed to cause this strange problem.
Let us know how these utilities work. In particular we'd like to get a sense of the type of NVIDIA GPU cards people are using.
Nick