I will diagnose and fix your hpc cluster performance problems

Most HPC clusters run at 3040% of their actual capability.

Not because the hardware is wrong. Because the configuration was never tuned for the actual workload.

I've diagnosed this exact problem at research institutions, AI labs, and engineering teams. The fixes are almost always in software and configuration not hardware.

What the audit covers:

Slurm configuration gaps (DefMemPerCPU, cgroup, fairshare)

InfiniBand fabric health and link speed validation  

Storage throughput (Lustre/BeeGFS/NFS stripe config)

MPI process binding and NUMA topology

HPL efficiency vs theoretical peak

Node health and silent hardware fault detection

What you receive:

Written diagnosis with severity rating per finding

Exact fix for each issue commands included Before/after benchmark numbers

Priority order: what to fix first for maximum impact

What I need from you: SSH access to login node,  your cluster specs, and 2 hours of low-activity time for benchmarking.

Turnaround: 24-48 hours after access is granted.