Leo – Hungarian HPC Cluster
The HPC is the newest machines named after Leo Szilard, a Physicist and inventor of the nuclear reactor. The Top500 qualified heavily accelerated machine has 252 Nvidia GPUs to enable running highly accelerated codes to help the 168 Sandy Bridge 8-core CPUs. The system is located at Debrecen, the second largest city in Hungary, and shown in Figure. The machine is integrated to PRACE European HPC network. Other technical information is provided in the table below.
Administrative data
Name | Leo |
Short Description | Hungarian HPC cluster |
Owner | NIIF Institute |
Country | Hungary |
Dedication to VI-SEEM
CPU (percent) | 10% |
Storage (percent) | 0% |
Accelerators (percent) | 10% |
CPU (core-hours per year) | 588,672 |
Storage in TB | 0 |
Accelerators (hours per year) | 125,269,402 |
Integration
System operational since | 2015 |
Available to the project from | PM04 |
Expected date to be phased | - |
Interfaces | SSH |
Computational Power
Number of servers | 84 | Interconnect type | FDR InfiniBand | Peak performance (Tflops) | 248 | |
Server specification | HP SL250S | Interconnect latency | 2.5 μs | Real performance (Tflops) | 208 | |
CPU per server | 2 | Interconnect bandwidth | 40 Gbps | Operating system | Red Hat Enterpise Linux | |
RAM per server | 125 GB | Local filesystem type | Lustre | Version | 6.5 | |
Total number of CPU-cores | 1,344 | Total storage (TB) | 585 TB | Batch system/scheduler | SLURM | |
Max number of parallel processes | 2,688 | Peak performance CPU (Tflops) | - | Accelerators type | Nvidia K20x and K40x misc. | |
Number of cores | 2688 | Servers equipped with accelerators | 84 | Peak performance accelerators (Tflops) | - | |
Accelerators per server | 3 | |||||
Development tools | Intel SDK | |||||
Libraries | Intel MKL, etc. | |||||
Applications | Maple, Matlab, etc. |