Status: April 2017
======================================================
DEIC
======================================================

https://www.deic.dk/

is our contact point to HPC resources:

https://www.deic.dk/supercomputing

Of the three supercomputers, the relevant for us is ...

======================================================
Abacus
======================================================

https://abacus.deic.dk/

3 teams from ITU have had succesful pilot projects (free of charge) so far:

Sebastian Risi, Peter Sestoft, now Ole Kristensen / Intermedia (for studies!)

We have a 100% success rate and positive experiences throughout.

Get access to Abacus 2.0 in the next four-month window:
July 1, 2017 to October 31, 2017.
Call opens April 1, 2017 and closes: May 15, 2017 at 12:00.

Compute nodes: 584
CPU Cores: 14,016
Performance:

Theoretical max: 766.6 TFlop/s (entire system)
Theoretical max: 560.6 TFlop/s (CPUs only)
Linpack performance: 462.4 TFlop/s (CPUs only)

Memory: 64.5 TB RAM
72 GPU nodes

As the base configuration except that each node contains 2 nVidia K40 GPU cards each with 2880 CUDA cores,
and 12 GB RAM. The theoretical performance for each K40 is 1.43 TFlop/s.


One level up, there is

=====================================================
Prace
=====================================================
Partnership for Advanced Computing in Europe

http://www.prace-ri.eu

Currently 7 machines:
about an order of magnitude or two bigger than Abacus, depending on your metrics.


Cores
RAM[TB]
Flops (PF)
  1. TOP500List 2016
Curie (F)
53k
378
2
74
Hazel Hen (DE)
185k
964
7.4
14
MareNostrum (ES)
50k
100
1
465
SuperMuc (DE)
245k
537
6.9
36
Marconi (IT)
299k
527
13
57
JuQueen (DE)
458k
459
5.9
19
PizDaint (CH, P100 GPU)
206k
194
16
8

Learn more:

http://www.prace-ri.eu/prace-resources/



plus:

+ free, once application accepted
+ good chances for success
+ seriously big machines

minus:

- considerable effort
- learning curve
- requires long term perspective
- you will have to develop new code, rather than use what you have



3 levels of access:


1. Tier-0
typical 15 million corehours, 12 pages applic, ambitious
2. Tier-1
3. Preparatory grants (!)

1. & 2. are for science, and experienced teams.

3. helps prepare/learn usage of PRACE machines:

Prep grants come with 6 personmonths of support!
run for 2-12 months.

If interested in any of this: sebastian@itu.dk