Data-Driven Prediction of Optimal MPI Rank Placement and Node Count for Memory-Bound CFD on AMD EPYC Clusters
Users who run a memory-bound OpenFOAM case on an AMD EPYC cluster must decide how many nodes to use and how many MPI ranks to place on each core complex.
ProofPaper ↗
Key points
- The fastest choice is often not the cheapest one.
- This paper revisits our earlier analytical model of this performance and energy trade-off with 560 measured configurations of the motorBike case on five Zen 4 and Zen 5 EPYC processors, including socket power from processor telemetry.
- On up to four nodes, small meshes are both fastest and cheapest on a single node on most processors.
- For the largest mesh the largest node count is both the fastest and the cheapest option, because its speedup is superlinear.
Sources (1)
- [1]Data-Driven Prediction of Optimal MPI Rank Placement and Node Count for Memory-Bound CFD on AMD EPYC ClustersarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 16, 12:40 PM
Users who run a memory-bound OpenFOAM case on an AMD EPYC cluster must decide how many nodes to use and how many MPI ranks to place on each core complex.
The fastest choice is often not the cheapest one.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 16, 2026One Communicator, Three Vendors: Extending the MoA Predictive Theory to a Real NVIDIA+AMD+CPU MPI Ensemble
- Oct 15, 2026ASML Is Building a Hybrid Bonding Tool!
- Oct 15, 2026FRACTAL: A Self-Similar Multi-Agent Orchestration Architecture with Formal Communication, Convergence, and Depth Bounds
- Oct 15, 2026FALCON: FPGA-Accelerated Latency-Certifiable Occupancy Networks for Camera-Based Autonomous Driving
- Oct 15, 2026Local and Distributed Large LLM Inference with MLX and RDMA on Mac Studio
- Oct 14, 2026GPT-Image-style photo editing on your own GPU (12 GB+): I built a zero-setup app for Qwen-Image 2.1 on stable-diffusion.cpp
