We will be releasing a minor QuNex patch this week (
0.97.1) it fixes the following:
fc_preprocess, when motion was not included as a regressor,
.mov files were not read, but the data was still referenced later in the code. This interaction is now resolved.
- Fixed a bug where
fc_process could sometime create an invalid GLM table.
- Fixed a bug where FSL’s
imrm did not remove some images which caused
hcp_diffusion to crash.
- Major improvements to consistency of documentation.
If something does not work, let me know.
We will be releasing a patch that fixes
hcp_diffusion which currently does not work because of a bug in the HCP pipelines. Furthermore,
hcp_diffusion now supports processing on newer GPUs (CUDA 10 and 11).
Support for newer GPUs for other commands in the DWI pipeline (
dwi_legacy) will be most likely released at the end of March.
We will be releasing QuNex
0.97.3 today or tomorrow. It fixes a bug in
hcp_msmall and a bug where
hcp_diffusion would sometime crash.
If something does not work, let us know.
Thanks for the good news Jure. Does QuNex 0.97.3 include the hcp_diffusion updates to include newer GPUs, or will this be in a new release in the future?
hcp_diffusion now supports newer NVIDIA GPUs as well. Just use
--bash_pre="module load cuda/11.2" or similar to load the appropriate CUDA module and add the
--nv flag. You can leave
hcp_dwi_cudaversion at the default version.
Note that other other DWI commands, such as
bedpostx do not support new CUDAs yet. I am still working on that.
Hi Jure, can you please share what is the estimated time line for the new qunex container with the updated CUDA support for the other DWI processing steps (bedpostx, probtrackx)?
Alternatively, is it possible to run those DWI steps with the 0.97.3 version using older CUDA modules, for example, 9.1? Thank you.
I am confident that QuNex
0.98.0, which will have CUDA 10.2+ support for all DWI commands, will be released in May.
hcp_diffusion can be ran using new GPUs (CUDA 10.2+), other DWI commands need GPUs that support CUDA 9.1 (on our end, we mainly use CUDA 9.1.85). So, if you have such a GPU then you should have no issues running the commands, either with QuNex
0.96.2a or with
0.97.3. Feel free to open more specific topics if you need assistance.