CP2K Benchmark - H2O-512 Scaling on the Cloud

Last Updated on: 2025-09-17

The original input files can be found here.

🔬 Simulation Details

Why H2O-512?

The H2O-64 benchmark is designed for local runs or small clusters.
However, when scaling to high-performance cloud environments, its small size underutilizes resources and fails to stress-test parallel performance.
Thus, we use H2O-512 to:

• Test scaling efficiency on multiple cores
• Measure cost-performance trade-offs on the cloud
• Better simulate real-world production workloads

Software Versions

The software versions used in this benchmark were the following:

Computational Resources Tested

The benchmark was conducted on the following machine series:

Intel-based

• C2
  Intel Xeon Scalable (Cascade Lake)
• C3
  Intel Xeon Scalable (Sapphire Rapids)
• C4
  Intel Xeon Scalable (Granite Rapids) — latest generation

AMD-based

• C2D
  AMD EPYC (Milan)
• C3D
  AMD EPYC (Genoa)
• C4D
  AMD EPYC (Turin) — latest generation

Notes on Machine Selection

• Some configurations were excluded due to insufficient memory:
  • c3d-highcpu-16
  • c2d-highcpu-16
  • local test machines
• For the C2 series, only the standard configuration was tested, as highcpu variants are not available.

See the results of the benchmark on the following pages:

• Execution Times Comparison
• Cost vs Time