Skip to content

High-Performance Computing - Category Home

High-Performance Computing - Topics

Immersion vs air cooling for an HPC rack packed with GPU nodes Why do exascale supercomputers rarely reach peak FLOPS on real science codes? At what point does renting cloud GPUs beat owning your own HPC cluster? Can waste heat from a supercomputer warm the buildings around it? InfiniBand vs Ethernet for the interconnect fabric of a large GPU cluster Does liquid cooling change how many cards an AI rack can hold? How much airflow does a passively cooled data center accelerator need? Why can a job on a national supercomputer sit in the queue for a week? FP64 machines vs AI clusters built for low precision math Does a billion simulated miles actually prove a car is safe? Can photonics replace the copper interconnect inside a GPU cluster? Why does interconnect bandwidth throttle large model training more than raw FLOPS? Why do 96-core server chips feel slower than a desktop CPU on lightly threaded tasks? How much power do optical interconnects save over copper at rack scale? How do researchers actually apply for time on a national supercomputer? Silent data corruption at scale: how do supercomputer centres catch bad answers? Optical computing: which problem does light actually solve better than transistors? Why does GPU cluster utilization stay under half even when the queue is full? TOP500 ranks machines by LINPACK while real workloads look nothing like it What does it cost in electricity to run an exascale machine for a year? Checkpoint and restart when a node dies three days into an HPC run Scratch space vs parallel filesystems: where should HPC data actually live?