Nova

The Nova cluster began in 2018 with compute nodes using Intel Skylake Xeon processors, 1.5TB or 11TB of fast NVME local storage and 192GB / 384GB / 3TB of memory.  Five of those nodes also have one or two Nvidia Tesla V100-32GB GPU cards. In 2021 the cluster was expanded with nodes using AMD processors, each having two 32-Core  AMD EPYC 7502 Processors, 1.5TB of fast NVME local storage and 512GB of memory. The new GPU nodes in addition have four NVidia A100 80GB GPU cards. The 2022 expansion consists of 54 regular compute nodes (with two 32-Core Intel 8358 processors, 1.6TB of local storage and 512GB of memory each) and 5 GPU nodes with two 24-Core AMD EPYC 7413 processors, eight A100 GPU cards, 960GB of local storage and 512GB of memory each.

The three service nodes include login node, data transfer node and the management node.

Large shared storage consists of six file servers and twelve JBODS configured to provide either 338TB of backed up storage or 457TB non-backed up storage per server. Later 7 more file servers have been added. The total storage capacity on Nova is about 5PB.

All nodes and storage are connected via Mellanox EDR (100Gbps) switch.

Some of the above equipment is used for education in a special instruction partition - 8 regular nodes and 3 GPU nodes are dedicated for classes, with additional compute nodes shared between the instruction and scavenger partitions.

 

Detailed Hardware Specification

Number of NodesProcessors per NodeCores per NodeMemory per NodeInterconnectLocal $TMPDIR DiskAccelerator CardJob Constraint Flags
72Two 18-Core
Intel Skylake 6140
36192 GB100G IB1.5 TBN/Anova18, intel, skylake, avx512
40Two 18-Core
Intel Skylake 6140
36384 GB100G IB1.5 TBN/Anova18, intel, skylake, avx512
28

Two 24-Core Intel Skylake 8260

48384 GB100G IB1.5 TBN/Anova18, intel, skylake, avx512
2Two 18-Core
Intel Skylake 6140
36192 GB100G IB1.5 TB2x NVIDIA Tesla V100-32GBnova18, intel, skylake, avx512
1Two 18-Core
Intel Skylake 6140
36192 GB100G IB1.5 TBone NVIDIA Tesla V100-32GBnova18, intel, skylake, avx512
2Two 18-Core
Intel Skylake 6140
36384 GB100G IB1.5 TB2x NVIDIA Tesla V100-32GBnova18, intel, skylake, avx512 
1Four 16-Core
Intel 6130
643 TB100G IB11 TBN/A

nova18, intel, skylake, avx512

2Four 24-Core
Intel 8260
963 TB100G IB1.5 TBN/A

nova18, intel, skylake, avx512

40Two 32-Core  AMD EPYC 750264512 GB100G IB1.5 TBN/Anova21, amd, epyc-7502
15Two 32-Core  AMD EPYC 750264512 GB100G IB1.5 TBfour NVidia A100 80GB nova21, amd, epyc-7502
56Two 32-Core Intel Icelake 8358 64512GB100G IB1.6TBN/Anova22, intel, icelake, avx512
5Two 24-Core  AMD EPYC 741348512GB100G IB960GBeight NVidia A100 80GB nova22, amd
16Two 32-Core Intel Icelake835864512GB100G IB1.5TBN/Anova23, intel, icelake, avx512
3Two 96-Core AMD EPYC 9654192768GB100G IB1.7TBN/Anova24, amd, epyc-9654, avx512

HPC group schedules regular maintenances between academic terms to update system software and to perform other tasks that require a downtime.

The date of the next maintenance is listed in the message of the day displayed at login (when ssh-ing to the cluster).

Note: Queued jobs will not start if they cannot complete before the maintenance begins. In the output of the squeue command the reason for those jobs will state (ReqNodeNotAvail, Reserved for maintenance) . The jobs will start after the scheduled outage completes.