SnuCore

SnuCore is a 16-node experimental heterogeneous CPU/GPU cluster. After porting the double precision LINPACK to SnuCore (written in MPI + OpenCL) and optimizing it with our software techniques for multiple GPUs, we have achieved 991 GFLOPS per node (total 15.9 TFLOPS). The detailed specification of a SnuCore node is as follows:

• 2 × 12-core 2.1 GHz AMD Opteron 6172
• 3 × AMD Radeon HD 6990 graphics cards
  • 6 GPUs
  • Water cooled
• Main memory: 128 GB
• Motherboard: Tyan s8232
• 2 × 1.5 KW power supplies
• Single-port Mellanox InfiniBand QDR HCA