[3dem] Installing Relion on a GPU cluster

[Sayan Bhakta]

Dear Dr. Clara,
Thanks for your suggestions. I was thinking exactly in the same way. I have
done the first suggested point of yours. I just needed some confirmation.
Thank you.

On Thu 25 Oct, 2018, 10:30 Dr. Clara Cai, <marketing at singleparticle.com>
wrote:

> Hi, Sayan
>
> Our suggestions:
> 1. You can actually install CUDA on the headnode (you will not be able to
> run the program without the driver but compiling should be fine). Make sure
> that all nodes have the same CUDA installed and for compute nodes, NVIDIA
> driver is installed properly.
>
> 2. Then compile and make RELION into a module that can be loaded on all
> nodes.
>
> 3. Set up a job submission template for RELION. We typically use SLURM
> scheduler, but it really depends on your cluster setup.
>
> Good luck!
>
> Best regards,
>
> -Clara
> SingleParticle.com
>
>
> On Wed, Oct 24, 2018 at 12:41 PM Sayan Bhakta <19sayanbhakta89 at gmail.com>
> wrote:
>
>> Hi,
>> I am using a cluster with 4 blades. One is the master node without any
>> graphics card. Other three have CUDA compatible high end graphics card. In
>> the master node CUDA was installed without any graphics driver and in the
>> three slave nodes CUDA was installed with appropriate graphics driver. All
>> the four nodes has similar architecture and CUDA installation paths are
>> same on each blades.
>>
>> I was going to compile relion with cmake in the master node and I got
>> stuck how to source the paths of CUDA and Nvidia driver environments from
>> different nodes while compiling with cmake. As on the master node there are
>> no graphics driver, cmake will not be able to compile the GPU version of
>> relion.
>>
>> My plan was as below.
>>
>> First, making a module with CUDA environment variables and loading it
>> before relion compilation. But I don't know how to make the module with
>> CUDA from different nodes. If any guidance is there would be very helpful.
>>
>> Secondly, after successful installation, making a module for relion with
>> all CUDA and other necessary (like openmpi,  ctffind, gctf,
>> mitioncorrection, etc) environment variables and loading it while
>> submitting the job. But in this case also I don't know how to specify GPU
>> from different nodes in the module file.
>>
>> Please help and ask for further information if needed.
>>
>> _______________________________________________
>> 3dem mailing list
>> 3dem at ncmir.ucsd.edu
>> https://mail.ncmir.ucsd.edu/mailman/listinfo/3dem
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20181025/dc2e5455/attachment-0001.html>