Dear Eline,
I took a look at the eddy_s2v.out you sent me. It is a little confusing in that it seems to contain the output from four separate runs of eddy. Also they all seem to fail at a much earlier stage than the s2v part of eddy. It seems that they fail the first time eddy attempts to transfer an image volume to the GPU, and always with the error message
thrust::system_error thrown in CudaVolume::common_assignment_from_newimage_vol after resize() with message: parallel_for failed: out of memory
parallel_for failed: out of memory
Hence, it looks like your GPU is out of memory. It looks like eddy is only attempting to transfer two volumes to the GPU, so I can’t believe it is running out of memory because of that. My guess is that either there is already other jobs running on the GPU, and they have exhausted the available memory. Or that some earlier jobs have crashed in a way so that the GPU hasn’t freed up the memory. I have had that happen on my local GPU and then I have had to reboot it after which things have worked again.
Jesper
> On 24 May 2019, at 14:07, Eline Roelofs <[log in to unmask]> wrote:
>
> Dear Jesper,
>
> Thank you for your advice! Unfortunately, I still get an error (see file attached) after the re-run with --very_verbose.
>
> Regarding .slspec.txt: the .json file didn't contain any information about slice timing or order. However, I've checked the scanner acquisition file (scanner: Philips Achieva 3.0T) which noted Slice Scan Order = default. I've decided on scan order using this website http://dbic.dartmouth.edu/wiki/index.php/Slice_Acquisition_Order .
> I had an uneven number of slices (n=75) but had to remove the top one because of topup and have been running the rest of my analyses with the adjusted scans. Could it be that I should use the orginal scans with uneven number of slices?
>
> Kind regards,
> Eline
> ________________________________________
> Van: FSL - FMRIB's Software Library [[log in to unmask]] namens Jesper Andersson [[log in to unmask]]
> Verzonden: vrijdag 24 mei 2019 14:36
> Aan: [log in to unmask]
> Onderwerp: Re: [FSL] eddy s2v error
>
> Dear Eline,
>
> your error is a little surprising as it does not occur in a part of the code that is really affected by s2v. Can you please re-run it with --very_verbose, pipe the output from that into a file and send me that?
>
>>
>> when running eddy_cuda9.1 on single band data with s2v correction I get an error (see below). I cannot find out the number of excitations, so had to guess mporder.
>
> If your data is single-band the number of excitations will be the same as the number of slices. But I think that an mporder of 6 is mostly a decent compromise. How did you decide on what the slspec.txt should look?
>
> Jesper
>
>
>> Without the extra commands eddy seems to be okay.
>>
>> Command:
>> eddy_cuda9.1
>> --imain=${subj}/anat_DTI_merged_topoff --mask=${subj}/b0_hifi_brain_final --index=index.txt --acqp=acqparams.txt --bvecs=${subj}/bvec_merged.bvec --bvals=${subj}/bval_merged.bval --topup=${subj}/topup_b0_merged --out=${subj}/anat_DTI_eddycorr \
>> --fwhm=10,8,4,2,0,0,0,0 \
>> --flm=linear \
>> --niter=8 \
>> --mporder=6 --s2v_niter=6 --s2v_lambda=1 --s2v_interp=trilinear --slspec=slspec.txt \
>> --repol --ol_type=sw
>> --cnr_maps=true
>>
>> Error:
>> ...................Allocated GPU # 0...................
>> Segmentation violation, Address not mapped, Offending address = 0x700
>> eddy_cuda9.1 EDDY::ReplacementManager::Update(EDDY::DiffStatsVector const&) [0x6488e7]
>> eddy_cuda9.1 EDDY::EddyInternalGpuUtils::detect_outliers(EDDY::EddyCommandLineOptions const&, EDDY::ScanType, std::shared_ptr<EDDY::DWIPredictionMaker>, NEWIMAGE::volume<float> const&, EDDY::ECScanManager const&, EDDY::ReplacementManager&, EDDY::DiffStatsVector&) [0x68d617]
>> eddy_cuda9.1 EDDY::EddyGpuUtils::DetectOutliers(EDDY::EddyCommandLineOptions const&, EDDY::ScanType, std::shared_ptr<EDDY::DWIPredictionMaker>, NEWIMAGE::volume<float> const&, EDDY::ECScanManager const&, EDDY::ReplacementManager&) [0x67b544]
>> eddy_cuda9.1() [0x54e6ea]
>> eddy_cuda9.1 EDDY::DoVolumeToVolumeRegistration(EDDY::EddyCommandLineOptions const&, EDDY::ECScanManager&) [0x54f607]
>> eddy_cuda9.1 main [0x522e14]
>> /lib/x86_64-linux-gnu/libc.so.6 __libc_start_main [0x7fe8529ebb97]
>> eddy_cuda9.1() [0x547664]
>>
>> Thanks in advance!
>>
>> Kind regards,
>>
>> Eline
>>
>> ########################################################################
>>
>> To unsubscribe from the FSL list, click the following link:
>> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
>
> ########################################################################
>
> To unsubscribe from the FSL list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
>
> ########################################################################
>
> To unsubscribe from the FSL list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
> <eddy_s2v.out>
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
|