To rapidly address the modeling requests of our collaborators, the simulation codes we develop and use can run on parallel architectures.
The framework is organized as follows.
Community code development
Remote distributed memory computer centers (> 1,000,000 core-hours per year)
- PRACE-EU projects
- IT4I (Ostrava, Czech Republic)
- METAcentrum (National Computational Ressources of Czech Republic)
These resources are available via the CESNET network, via the institution subscription at FZU (IT service).
Local shared memory architectures (small jobs submission)
- Max capacity: 560,000 core-hours per year.
CPU 2 x 16-core HyperThreaded Intel(R) Xeon(R) CPU E5-2683 v4 @ 2.10GHz (TDP 120 W) machine (64 virtual cores). Funded by BIATRI project (Czech Republic),
RAM: 125 GB. HDD: 11 TB.
Local workstations (code development)
- Max capacity: 140,000 core-hours per year.
CPU 2 x 8 cores, Intel(R) Xeon(R) CPU E5-2650 v2 @ 2.60GHz (TDP 95 W),
32 GB RAM.
Equipped with GPU NVIDIA GeForce GT 630(D3), GPU 96 cores @ 700 MHz, 2 GB RAM. CUDA Toolkit 2.1 < Curr: 6.5.16 < CUDA Toolkit 8.0
[Maximum capacity of the GPU: 840’960 core-hours / year].
Local HD 886 GB. Portable HD 466 GB.
- Max capacity: 105,000 core-hours per year.
CPU 2 x 6 cores, Intel(R) Xeon(R) CPU E5-2630 v2 @ 2.60GHz (TDP 80 W), 16 GB RAM.
Radeon Tahiti PRO HD 7950/8950 OEM / R9 280.
Local HD: 848 GB.
- Max capacity: 70,000 core-hours per year.
CPU 2 x 4 cores, Intel(R) Core(TM) i7-3770 CPU (TDP 77W) @ 3.40 GHz, 8 GB RAM.
Local HD: 385 GB + 1.5 TB.
- CPU 2 x 2 cores, Intel(R) Core(TM) i5-5200U CPU (TDP 15W) @ 2.20 GHz, 4 GB RAM.
Local SSD: 256 GB.
Equipped with NVIDIA GeForce GT940M, GPU 384 cores @ 1072 MHz, 8 GB RAM (CUDA Toolkit 5.0+).
Potential capacity: 3’363’840 core-hours per year.
Currently reached performances on Octopus
We attempt to optimize the computer time we use. In particular, we prepare numbers of TD-runs using Octopus software.
Here we keep a small note on the performances reached on various computers.
- MPCDF Draco > Octopus (version of 2019 09 11): 2.5 s per TD step (4 nodes of 32 cores).
- IT4I Salomon > Octopus (version of 2019 06 10): 5 s per TD step (4 nodes of 24 cores).
- PRACE Prometheus (Poland) > Octopus (version of 2019 10 16): 5 s per TD step (4 nodes of 24 cores).