strutture:cnaf:cnaf_rd:infrastruttura:cluster_calcolo
Table of Contents
Cluster Calcolo
Il cluster R&D del CNAF si compone attualmente delle seguenti macchine:
- rd-ui: VM per il login degli utenti e per la sottomissione dei job
- rd-lsfmaster: VM con il master daemon e scheduler del batch system LSF
- rd-slurm: VM con il master daemon e scheduler del batch system SLURM
- rd-coka-01: server con 1 MIC e 2 GPU K20
- rd-coka-02: server con 2 MIC
- rd-gpu-02: server con 2 GPU C2050
- carma-devkit: host con processore ARM e GPU nVidia
- rd-arm-compiler: host per la cross compilazione
Descrizione Hardware
| Name | Chassis | MotherBoard | Cores | Memory | GPU/Co-processor | OS |
|---|---|---|---|---|---|---|
| rd-coka-01 | E4 | Supermicro X9DRG-HF | SandyBridge 2 x 12 cores @ 2.00 GHz | 32GB @ 1333MHz | 2 x nVidia K20m + Intel Xeon Phi 3110P | CentOS 6.4 |
| rd-coka-02 | E4 | Supermicro X9DRG-HF | SandyBridge 2 x 12 cores @ 2.00 GHz | 32GB @ 1333MHz | 2 x Intel Xeon Phi 5110P | SL 6.4 |
| rd-gpu-02 | E4 | Supermicro X8DTG-QF | Westmere 2 x 8 cores @ 2.40 GHz | 24GB @ 1066MHz | 3 x nVidia C2050 | CentOS 6.4 |
| carma-devkit | Seco | cardhu | ARM A9 1 x 4 cores @ 1.4GHz | 2GB DDR3 | nVidia Quadro 1000M | Ubuntu 12.04 |
Dettagli Hardware
| GPU K20 | |
|---|---|
| Modello | NVIDIA Tesla K20m |
| Cores | 13 (Multiprocessors) x 192 (CUDA cores) = 2496 CUDA cores |
| GPU clock rate | 706 MHz |
| Memory | 5 GB GDDR5 |
| Memory clock rate | 2600 MHz |
| Memory bandwidth | 208 GB/sec |
| Max power cons. | 225 Watt |
| Peak SP perf. | 3.52 Tflops |
| Peak DP perf. | 1.17 Tflops |
| GPU C2050 | |
|---|---|
| Modello | NVIDIA Tesla C2050 |
| Cores | 14 (Multiprocessors) x 32 (CUDA cores) = 448 CUDA cores |
| GPU clock rate | 1147 MHz |
| Memory | 3 GB GDDR5 |
| Memory clock rate | 1500 MHz |
| Memory bandwidth | 144 GB/sec |
| Max power cons. | 238 Watt |
| Peak SP perf. | 1.03 Tflops |
| Peak DP perf. | 515 Gflops |
| GPU Quadro 1000M | |
|---|---|
| Modello | NVIDIA Quadro 1000M |
| Cores | 2 (Multiprocessors) x 48 (CUDA cores) = 96 CUDA cores |
| GPU clock rate | 1400 MHz |
| Memory | 2048 MBytes |
| Memory clock rate | 900 MHz |
| Memory bandwidth | 28.8 GB/sec |
| Max power cons. | 45 Watt |
| Peak SP perf. | 268.8 Gflops |
| Peak DP perf. | … Tflops |
| MIC 3110P | |
|---|---|
| Modello | Intel Xeon Phi 3110P |
| Cores | 57 (Multiprocessors) x 4 (hardware threads) = 228 cores |
| CPU clock rate | 1100 MHz |
| Memory | 3 GBytes |
| Memory clock rate | 2500 MHz |
| Memory bandwidth | 240 GB/s |
| Max power cons. | 225 Watt |
| Peak SP perf. | … Tflops |
| Peak DP perf. | 1.0 Tflops |
| MIC 5110P | |
|---|---|
| Modello | Intel Xeon Phi 5110P |
| Cores | 60 (Multiprocessors) x 4 (hardware threads) = 240 cores |
| CPU clock rate | 1053 MHz |
| Memory | 8 GBytes |
| Memory clock rate | 2500 MHz |
| Memory bandwidth | 320 GB/s |
| Max power cons. | 225 Watt |
| Peak SP perf. | … Tflops |
| Peak DP perf. | 1.01 Tflops |
Descrizione Software
GPU
Per utilizzare le GPU sono disponibili:
- CUDA 5.0
- CUDA 5.5
- Intel OpenCL 1.2-3.0.67279
Intel Xeon Phi
Per l'uso dei co-processori Phi sono disponibili:
- Intel Parallel Studio XE 2013 (icc 13.1.1)
- Intel Parallel Studio XE 2013 SP1 (icc 14.0.0)
MPI
- Intel MPI
- MVAPICH 2
- OpenMPI
strutture/cnaf/cnaf_rd/infrastruttura/cluster_calcolo.txt · Last modified: 2013/10/15 13:20 by caberletti@infn.it