Hardware

We provide multiple systems for working with sensitive data:

Verdi: General purpose system with large long term storage coupled to compute resources (both CPU and GPU nodes).
DGX-2: Expert system for training AI-models based on NVIDIA’s DGX-2 system

Verdi

Verdi is the current incarnation of our Data Science Platform, supporting use of both sensitive and non-sensitive data aimed at users of a wide range of expertise. It’s intended to enable long term data retention - allowing data to be accessed and worked with during periods of lower activity and making it easy to do compute at high end e-infrastructure provided by NAISS. It consists of three main parts: storage, CPU compute and GPU compute.

The nodes of the cluster is connected by dual 25 Gigabit Ethernet for all nodes.

Storage:

6 Storage nodes, each with:
- 24 Hard drives, 22 TB capacity each
  - 528 TB Total HD Storage per node
- 4 Solid state drives, 6.4 TB capacity each
  - 25.6 TB Total SSD Storage per node
Total
- 3.2 PB HD Storage
- 153 TB SSD Storage

CPU compute:

40 nodes, each with:
- 32 CPU cores (2 Intel® Xeon® Gold 6526Y 2.8G)
- 1 TB RAM
- 6.4 TB SSD
Total
- 1280 CPU Cores
- 40 TB RAM
- 256 TB SSD storage

GPU Compute

6 L4 + 2 L40s nodes, each with:
- 4 GPUs per node, either:
  - NVIDIA L4 (24 GB per GPU), or
  - NVIDIA L40s (48 GB per GPU)
- 32 CPU cores (2 Intel Xeon Gold 5416S 2G)
- 512 GB RAM
- 8 TB SSD
Total
- 256 CPU Cores
- 4 TB RAM
- 64 TB SSD storage

DGX-2

The DGX-2 system is based on NVIDIA’s system of the same name. The system consists of a DGX-2 server with 15 V100 (32 GB) GPUs connected to a storage server with 10 Gigabit Ethernet.

15 NVIDIA V100 (32 GB) GPUs, allocated in blocks of 3 or 4 to users
More details

Contact

Our system design team:

	Pontus Freyhult IT Architect System design support, Policy support, Management, System development
	Joel Hedlund Head of AIDA Data Hub Management, Policy support, System design support, Data sharing, Economy

Please contact us for support or more information!