The COSMA8 prototype system entered service in October 2020. It was then extended in early 2021, to become the initial COSMA8 installation. This is undergoing science commissioning until October 2021, when it will enter full service. It is the first COSMA system to reach Petascale, with 1.37PF achieved on the HPL benchmark (RMax), and a 1.9PF peak performance. Each node has 4.2 TFLOP HPL performance.
It is currently comprised of:
- 360 compute nodes with 1 TB RAM and dual 64-core AMD EPYC 7H12 water-cooled processors at 2.6GHz
- 2 login nodes with 2 TB RAM and dual 32-core AMD EPYC 7542 processors at 2.9GHz
- 2 fat nodes with 4 TB RAM and dual 64-core AMD EPYC 7702 processors at 2.2GHz
- 1 AMD GPU nodes with 6 MI50 GPUs (32GB), 1TB RAM, dual 16-core AMD EPYC 7282 processors at 2.8GHz
- 1 AMD Milan node with a MI100 GPU, 1TB RAM, dual 64-core AMD EPYC Milan 7713 processors at 2GHz
- 1 NVIDIA GPU node with 10 V100 GPUs (32GB), 768GB RAM, dual Intel Xeon Gold 5218 processors at 2.3GHz
- 2 console nodes with a single 16-core AMD EPYC 7302 processor at 3GHz and 256GB RAM
The interconnect is Mellanox HDR, 200GBit/s, with a non-blocking fat tree topology.
The full initial COSMA8 system is due to enter service in October 2021.
Information about using COSMA8 can be found here.