Skip to content
GIRIVON
Infrastructure+
Systems+
Components+
Computing Devices+
Solutions+
Compute Tools+
Knowledge+

HPC Infrastructure

High-performance computing clusters — compute nodes, low-latency interconnects, parallel file systems and schedulers — for simulation and scientific workloads.

Peak versus delivered performance

Theoretical peak (Rpeak) is cores × clock × floating-point operations per cycle, plus accelerator peak. Delivered performance depends on memory bandwidth, interconnect latency and how much of the code parallelises — Amdahl's law caps the speedup of any job with a serial fraction. The HPC cluster planner reports Rpeak and an estimated sustained figure from an efficiency you set.

Memory bandwidth per core

Many CFD, weather and finite-element codes are memory-bandwidth bound. Populating every memory channel and choosing core counts that keep bandwidth per core healthy often matters more than peak FLOPS.

Interconnect and storage

Tightly coupled MPI jobs need low-latency fabrics; embarrassingly parallel jobs tolerate commodity Ethernet. Parallel file systems are sized for aggregate checkpoint and I/O bandwidth, not just capacity.

10 products

Compare selected

Specifications to compare

  • CPU cores (total)
  • Memory
  • GPU count
  • Interconnect
  • Network speed

Calculate before you buy

Typical workloads