The UGent compute infrastructure consists of several specialised clusters, jointly called Stevin. These clusters share a lot of their file space so that users can easily move between clusters depending on the specific job they have to run.

Login nodes

The HPC-UGent Tier-2 login nodes can be access through the generic name login.hpc.ugent.be.

Connecting to a specific login node

There are multiple login nodes and you will be connected with one of them when using the generic alias login.hpc.ugent.be. (You can check which one you are connected to using the hostname command).

Note that since end of November 2018, direct access to a specific login node is no longer possible, due to a stricter security policy.

Compute clusters

#nodes CPU Mem/node Diskspace/node Network

delcatty

123 2 x 8-core Intel E5-2670
(Sandy Bridge @ 2.6 GHz)
64 GB 400 GB FDR InfiniBand

phanpy

16 2 x 12-core Intel E5-2680v3
(Haswell-EP @ 2.5 GHz)
512 GB 3x 400 GB (SSD, striped) FDR InfiniBand

golett

200 2 x 12-core Intel E5-2680v3
(Haswell-EP @ 2.5 GHz)
64 GB 500 GB FDR-10 InfiniBand

swalot

128 2 x 10-core Intel E5-2660v3
(Haswell-EP @ 2.6 GHz)
128 GB 1 TB FDR InfiniBand

skitty

72 2 x 18-core Intel Xeon Gold
6140 (Skylake @ 2.3 GHz)
192 GB 1 TB
240 GB SSD
EDR InfiniBand

victini*

96 2 x 18-core Intel Xeon Gold
6140 (Skylake @ 2.3 GHz)
96 GB 1 TB
240 GB SSD
10 GbE

Only clusters with an InfiniBand interconnect network are suited for multi-node jobs.
Other clusters are for single-node usage only.
(*) victini is the default cluster (since end of November 2018)

Shared storage

General Parallel File System (GPFS) partitions:

  • $VSC_HOME: 35 TB
  • $VSC_DATA: 702 TB
  • $VSC_SCRATCH: 1 PB (equivalent to $VSC_SCRATCH_KYUKON)
  • $VSC_SCRATCH_PHANPY: 35TB (very fast, powered by SSDs)

Look for...