MUSICA¶
Note
MUSICA is currently under the closed test phase. The open testing phase will be later in the year.
MUSICA (Multi-Site Computer Austria) is the latest addition to Austria's national supercomputing infrastructure. The system's compute resources are distributed across three locations in Austria: Vienna, Innsbruck, and Linz.
The system comprises a total of 440 compute nodes, all featuring direct liquid cooling based on the 3rd generation of Lenovo’s Neptune™ platform.
The hardware was supplied by Lenovo, the storage by MEGWARE, the installation was carried out by the Vienna-based company EDV-Design.
Cluster partitions¶
Vienna:
- 112 GPU nodes
- 72 CPU nodes
Linz & Innsbruck:
- 80 GPU nodes
- 48 CPU nodes
CPU nodes¶
Each CPU node is equipped with two AMD EPYC 9654 processors (96 cores per CPU), for a total of 192 cores. A node features 768 GB of DDR5 memory (24 × 32 GB DIMMs) and a 1.92 TB NVMe local storage device. CPU nodes are connected via Infiniband NDR200.
GPU nodes¶
GPU nodes come with the same CPU and memory configuration as the CPU nodes. In addition, each GPU node is equipped with 4× NVIDIA H100 SXM5 GPUs (94 GB memory per GPU), a local 7.68 TB NVMe drive, and 4× Infiniband NDR200 adapters.
Storage¶
Each MUSICA site is equipped with 4 PB of all-NVMe storage, based on the Weka parallel file system. The storage system consists of 28 servers, each equipped with 12 × 15.36 TB NVMe drives. It delivers up to 1.8 TB/s of read throughput and 750 GB/s of write throughput.
Cooling¶
The MUSICA system is optimized for year-round free cooling through direct liquid cooling of all compute nodes. Residual heat from components such as power supplies (PSUs), power distribution units (PDUs), and network switches is managed via rear-door heat exchangers integrated into the racks.
Performance¶
Using a standard HPL benchmark, the four GPUs of a single GPU node deliver an aggregate performance of 192.7 TFlop/s, while for a single CPU node a performance of 8.9 TFlop/s could be measured.
For testing purposes, the GPU cluster partitions of Vienna and Innsbruck have been linked, and an aggregate performance (Rmax) of 31.8 PFlop/s could be achieved.
In November 2024, MUSICA debuted at rank 50 on the TOP500 list of the world’s most powerful computing systems. The Linpack performance (Rmax) of 24.22 PFlop/s, used for the TOP500 ranking, reflected the initial deployment - of approximately two-thirds of the system installed in Vienna at that time. Once all three sites are operational, the accumulated performance is projected to reach 42 PFlop/s (Rmax).