Compute
Compute Facilities available
The compute component of Edinburgh Compute and Data Facility (ECDF), known as Eddie, offers a number of services aimed to satisy as best as possible all researchers' computational requirements.
Why Use Eddie
Eddie can cut the time taken to compute problems by running the software in parallel, or by breaking the problem into many pieces, each of which can be run on a separate cpu in parallel. Examples of the speed improvements that researchers have realised already are in Psychiatry, where processing of brain scans from a Schizophrenia study took 28 hours instead of 469 days, and a protein structure prediction study which involved 810,000 simulations and used 1.5 CPU years of computation in less than 2 days.
Users who wish to discuss their use of Eddie with the ECDF team should submit a query through the IS Helpline IS.helpine@ed.ac.uk
Eddie Linux Compute Cluster
Currently, Eddie is primarily made up of two parts:
- Mark 2 Phase 1: 130 IBM dx360M2 iDataPlex servers with two Intel Westmere E5620 quad core processors and 24GB of RAM, all connected with gigabit ethernet
- Mark 2 Phase 2:156 IBM dx360M3 iDataPlex servers with two Intel Westmere E5645 six core processors and 24GB of RAM, 68 of which are connected with a QDR Infiniband network for large scale parallel jobs, and all connected with gigabit ethernet.
GPGPU Service
GPGPU stands for General-Purpose computation on Graphics Processing Units. CUDA is a toolkit for GPGPU written by NVIDIA. OpenCL is a new industry standard programming system for developing parallel programs that typically execute on heterogeneous computing systems. More information about using Cuda and OpenCL is available from the links below. Currently the GPU provision is:
- One Tesla s1070 box hosting four C1060 Tesla GPU processors. Two workers nodes each have two GPUs attached
- Four worker nodes each hosting two C2050 Tesla GPU processors, also known as Fermis.
Symmetric Multiprocessing (SMP) and Large Memory Systems
Large memory jobs and shared memory programs using methods such as OpenMP (Open Multi-Processing) can make use of the following:
- Two mark 2 phase 1 nodes with 48GB of RAM
- 24-core 4-socket Intel X7542 system with a total of 512GB
This article was published on Aug 25, 2011