Olivia

The name Olivia is inspired by ´olivine´, a mineral once extracted from the former Lefdal mine, now transformed into a state-of-the-art data centre, Lefdal Mine Data Centers (LMD), where the supercomputer is hosted.

Information for pilot users here.

The next generation powerful supercomputer for HPC and AI applications

Olivia is provided by Hewlett Packard Enterprise and will have a computational capacity 17 times greater than the current most powerful national supercomputer, Betzy.

Technical Details

Details

Olivia

System

HPE Cray Supercomputing EX

Number of CPU nodes

252

CPU type

AMD Epyc Turin

CPU cores in total

64512

CPU cores per node

256

Memory in total

189 TiB

Memory per node

768 GiB

GPU type

NVIDIA GraceHopper Superchip (NVIDIA GH200 120GiB)

Number of GPU nodes

76

GPUs per node

4

GPUs in total

304

Interconnect

HPE Slingshot Interconnect

Regular Compute nodes (HPC):

  • 252 compute 2 way nodes with a total of 64512 AMD cores

  • 768GiB memory per node, giving a total of 189 TiB

  • Slingshot injection port 200 Gbits/s, with combined capacity of 49 Tbits/s

  • 3.84 TB NVMe per node, yielding 967 TB for local scratch

Accelerate nodes (AI):

  • 76 nodes with GPUs,each node has four NVIDIA GraceHopper Superchip

    • 76 accelerated 4 way nodes with total of 21888(76472) ARM cores

    • 76 accelerated nodes with a total of 304 (76*4) Hopper accelerators

  • 384 GiB HMB3 accelerator memory per node total aggregated 28 TiB

  • 480 GiB LPDDR5 processor memory per node total aggregated 35TiB

  • Injection ports yielding 800 Gbits/s, giving combined bandwidth of 60 Tbits/s

Cluster Storage:

  • 3.159 PB spinning disks

  • 1.107 PB solid state storage