Mangi

The new Mangi compute cluster expands MSI's hardware resources to run a wider variety of jobs.

Mangi is a heterogeneous HPE cluster featuring nodes with leading-edge AMD processors that are tightly integrated via a high-speed InfiniBand network. In addition, it contains ten nodes with large memory (2 TB per node), accelerator nodes (GPU), and all the nodes are equipped with solid-state storage devices (SSD) for high-performance local scratch.

Compute infrastructure:  

• A total of 164 AMD ROME compute nodes with a total of 20,992 compute cores. 12 nodes include a 2-way NVidia Tesla V100 GPU configuration, two nodes include 4 NVidia Tesla V100 GPUs, and one node includes an 8-way NVidia Tesla V100 GPU configuration. 
• Memory: 144 nodes have 256 GB of RAM, 10 nodes feature 512 GB, and 10 nodes have 2 TB of RAM each. The 2-way and 4-way GPU nodes have 384 GB of RAM each and the 8-way has 768 GB of RAM. 
• SSD input/output nodes: all nodes have 480 GB solid state drives (SSDs) for high-performance input/output.  The total system SSD capacity is 85 TB.

The cluster is connected via HDR100 (100 Gb/second) Infiniband for high-speed communication between nodes.

The name “Mangi” comes from an Ojibwe word meaning “make bigger.” The name reflects Minnesota’s cultural heritage, as well as the idea that Mangi serves as an addition to MSI’s existing HPC system, Mesabi.

What can I do with Mangi?

Mangi features a heterogeneous architecture designed to a diversity of job types. You can schedule one or many 128-core nodes as a parallel threaded or MPI job, or request a single processor on a single node for several days. Depending on the partition, your job may have full control of a node or share nodes with other jobs. Other partitions provide access to specialized hardware such as high-memory nodes (up to 2TB of RAM) and NVIDIA V100 GPUs.

See also the documentation for job partitions.

How do I access Mangi?

Mangi is most efficiently accessed through a terminal environment. MSI users can follow directions in the Connecting to HPC Resources quick start guide to access Mangi and all other MSI HPC systems. Once connected to Mangi you will be able to submit jobs to Mangi's partitions using scripts with specific Slurm commands also known as Slurm scripts. Mangi should be your first choice for computing at MSI, but Agate is also available.