Hardware

We take a "building block" approach to EIDF, to give it the flexibility it needs. EIDF's main hardware partner is HPE and we work with them to procure hardware on a rolling basis.

EIDF's hardware is the least interesting part of it! The hardware forms a low-level infrastructure layer and is largely hidden from service users. On top of the hardware is what we call the virtualised "platform layer", and on top of that we create the services that users interact with.

EIDF is located at EPCC's Advanced Computing Facility on the edge of Edinburgh. 

Current hardware

General EIDF

  • Storage
    • E1000 Clusterstor "hot" storage (20 PB Lustre parallel file system)
    • Ceph "warm" storage (13 PB block storage, CephFS and S3)
    • 2x matching SpectraLogic tape archives driven by HPE DMF
  • Compute
    • HPE ProLiant host servers for c. 500 large VMs (Intel Xeon Platinum and AMD Epyc CPUs)
    • HPE Apollo new generation GPU servers with 532 GPUs (80x H200, 160x H100 and 292x A100)
    • HPE SuperDome Flex large-memory systems (18TB RAM) with 576 cores
    • Cerebras CS-3 Wafer-Scale Cluster AI accelerator with 4 units. 

Safe Haven Services

  • Storage
    • Nimble "very hot" storage (1 PB)
    • BeeGFS "warm" storage (5PB)
  • Compute
    • HPE ProLiant host servers for c. 300 large VMs (Intel Xeon Platinum and AMD Epyc CPUs)
    • HPE Apollo new generation GPU servers — 144 (80GB and 40GB) GPUs
    • HPE SuperDome Flex large-memory systems (18 TB RAM) with 576 cores

 

This article was published on