Getting started with Mahti
This is a quick start guide for Mahti users. It is assumed that you have previously used CSC supercomputing resources like Puhti, Sisu or Taito. If not, you can start by looking here.
Go to my.csc.fi to apply for access to Mahti or view your projects and their project numbers if you already have access.
Connecting to Mahti
Connect using a normal ssh-client:
$ ssh <csc_username>@mahti.csc.fi
Default modules, which are loaded automatically, are
Compilers and MPI
Currently, Mahti has GNU compiler suites (versions 9.3.0 and 7.4.0), as
well as AMD and Intel compiler suites. All compiler suites can be used
mpicxx (C++), or
wrappers. We recommend to start with the GNU compiler suite, but for some
applications the other suites may provide better performance.
In Mahti, many applications benefit from hybrid MPI/OpenMP parallelization, so it is recommended to build a hybrid version if it is supported by your application.
More information about compilers here.
You need to have the MPI module loaded when submitting your jobs
High performance libraries
Mahti has several high performance libraries installed, more information here.
More information about specific applications can be found here. Note, the preinstalled selection is not as large as on Puhti.
In Mahti, many applications benefit from hybrid MPI/OpenMP parallelization, however, the optimum ratio of MPI tasks and OpenMP threads depends a lot on the particular application as well as on particular input data set. Mahti supports also simultaneous multithreading (SMT), i.e. two threads can be run on the same physical CPU core. Benefits of multithreading depend also on the application, in some cases it improves performance while in some cases performance becomes worse. Binding of threads to CPU cores can also have an impact on performance.
More information about controlling hybrid applications can be found here.
The project based shared storage can be found under
/scratch/<project>. Note that this folder is shared by all
users in a project. This folder is not meant for long term data
storage and files that have not been used for 90 days will be
automatically removed. The default quota for this folder is 1
TB. There is also a persistent project based storage with a
default quota of 50 GB. It is located under
/projappl/<project>. Each user can store up to 10 GB of data in
their home directory (
The disk areas for different supercomputers are separate, i.e. home, projappl and scratch in Puhti cannot be directly accessed from Mahti.
More detailed information about storage can be found here.
Moving data between Mahti and Puhti
Data can be moved between supercomputers via Allas by first uploading the data from one supercomputer and then downloading it to the other. This is the recommended approach if the data should also be preserved for a longer time.
Data can also be moved directly between the supercomputers with the rsync command. For example, in order to copy my_results (which can be either a file or a directory) from Puhti to the directory /scratch/project_2002291 in Mahti, one can issue in Puhti the command:
rsync -azP my_results <username>@mahti.csc.fi:/scratch/project_2002291
See Using rsync for more detailed instructions for rsync.
How Mahti and Puhti differ?
If you are new to supercomputes, or the details below is unfamiliar, you likely should start with Puhti and some introductory tutorials first. In a nutshell, Mahti is meant for large parallel jobs, and Puhti for a wide variety of small to medium sized jobs including special resources.
|Resources are granted||By full nodes||By finer detail (cores/memory/...)|
|Minimum job size||128 cores (1 node)||1 core (1/40 node)|
|Maxmimum job size (cores)||200 nodes (*) (25600)||100 (**) nodes (40000)|
|Memory per node (average per core)||245 GB (2 GB)||192 - 1500 GB (4 - 37 GB)|
|Fast local disk||no||yes (NMVe)|
(*) And even more via Grand Challenge calls. (**) To be scaled down to 25 nodes in autumn 2020.
Last edited Tue Aug 25 2020