Hey Iain,
I can completely relate to that problem - it took quite some while for me to wrap my head around that. The way I understand it:
CUDA is basically a way to talk to your GPU and make it perform parallel computations, which massively speeds up some processes as compared to running them on CPUs. There are different versions of CUDA. The compute capability, on the other hand, is the underlying programming architecture of the GPU and depends on the generation ("microarchitecture") of the GPU, with different generations having different compute capabilities. Certain versions of compute capabilities support certain versions of CUDA, but there is an overlap between (i.e., not every new generation automatically relates to a new version). The German Wikipedia page on CUDA has an overview on the compute capability-to-CUDA version relationship: https://de.wikipedia.org/wiki/CUDA#Unterst%C3%BCtzte_GPUs.
Currently, I am using eddy_cuda9.1 on a "Pascal" microarchitecture GPU with a relatively broad range of supported CUDA versions.
However, newer generations of GPUs (i.e., the ones that you can buy right now) usually only support CUDA versions 10 and higher, and the FSL GPU site (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GPU) hasn't been updated for some time, so no info on whether or not new versions of FSL (specifically, eddy_cuda) are running on these new CUDA versions.
I hope this helps!
Still, I don't know if newer versions of eddy_cuda able to run on CUDA 10.x or 11.x GPUs.
Cheers,
Johannes
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1
This message was issued to members of www.jiscmail.ac.uk/FSL, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
|