The CLAIX hardware is organized in the batch system as queues or partitions of nodes. Each partition consists of a collection of node types with similar specifications.

The following table shows how the various node types are organized in partitions that can be selected within Slurm.

Abbreviations:

• p.N. - per Node
• all CPU Codename are meant to be 'Intel', if not anything to the contrary
• Value '#Cores per Node' mean the physical cores (not Hyperthreading 'CPUs' reported in the OS system). By default the HyperThreading is OFF on our nodes; for those few with HT=ON remarks are posted here.

Slurm information					Node information										Cluster information
Node Type	Partition	Features	max recomm. memory per node [MB]	default memory per task [MB]	Hardware Node Type	CPU Codename	CPU Model	Clock Speed [GHz]	#Nodes	#Sockets per Node	#Cores per Socket	#Cores per Node	Memory per Node [GB]	SSD/ HDD size [GB]	Sum Sockets	Sum Cores	Sum Memory [GB]	Beginning of operation
ncm	c18m	skylake, skx8160, hpcwork	187.200	3.900	Intel HNS2600BPB	Skylake	Platinum 8160	2.1	1.032	2	24	48	192	480	2.064	49.536	198.144	December 2018
nrm	c18m	skylake, skx8160, hpcwork	187.200	3.900	Intel HNS2600BPB	Skylake	Platinum 8160	2.1	211	2	24	48	192	480	422	10.128	40.512	February 2019
ncg	c18g	skylake, skx8160, hpcwork	187.200	3.900	Supermicro 1029GQ-TVRT-01 2 x Tesla V100	Skylake	Platinum 8160	2.1	48	2	24	48	192	480	96	2.304	9.216	March 2019
nrg	c18g	skylake, skx8160, hpcwork	187.200	3.900	Supermicro 1029GQ-TVRT-01 2 x Tesla V100	Skylake	Platinum 8160	2.1	6	2	24	48	192	480	12	288	1.152	March 2019

SubNUMAClustering enabled for Skylake CPUs, which means, there exist 4 NUMA nodes with 12 cores each. Slurm interpretes these NUMA nodes as sockets, therefore these nodes have 4 sockets and 12 cores per socket.

Further systems are integrated into the RWTH High Performance Computing through the Integrative hosting service.