Vuori (vuori.csc.fi) is a high-performance supercluster. It has 304 compute nodes, first 272 nodes having two six-core 2.6 GHz AMD Opteron processor each, following 24 nodes each having two six-core 2.6 GHz Intel Xeon processors, and final 8 GPGPU nodes with NVidia Tesla GPGPU accelerators and two six-core 2.6 GHz Intel Xeon processors. The amount of compute cores is thereby 3648. In addition there are two login nodes (vuori1.csc.fi and vuori2.csc.fi) and one control node for administrative purposes. The compute nodes are equipped with different memory configurations, e.g. nodes 1-272 have 16 GB or 1.33 GB per core (128 nodes, 1536 cores) and 32 GB or 2.67 GB per core (144 nodes, 1728 cores), big memory nodes 273-296 have 8 GB per core (24 nodes, 288 cores).

The nodes 241-272 (384 cores) are solely dedicated for the Corander Research group:

http://www.csc.fi/english/csc/news/news/smartbayes

Processors are connected via InfiniBand (4x QDR) network, GPGPU nodes and 24 big memory nodes are connected via QDR InfiniBand interconnect. The theoretical peak performance of Vuori when 240 nodes (2880 cores) are used is 30 Tflop/s (Rmax) and the real performance in Linpack is 23.4 Tflop/s (Rmax). With the Corander partition using 272 nodes (3264 cores): Rpeak = 33,9 Tflop/s and Rmax = 26,5 Tflop/s.

For program development on Vuori there are available Fortran-, C- and C++ -compilers from Portland Group (PGI) , PathScale, GNU Compiler Collection (GCC) and Intel. The supported MPI library is MVAPICH2 but there is also OpenMPI.

Vuori is intended for serial jobs and for middle-size parallel jobs. The scalability of the job must be demonstrated by a special scalability test, if a user wants to use more than 144 cores.