Sagar Dolas, Program Manager at SURF — the Dutch national organization for digital infrastructure in research and higher education — joins Kevin Jackson to challenge one of HPC's most enduring assumptions: that bigger machines produce more science.

Sagar argues that raw compute capacity and scientific productivity are not the same thing, and that the gap between them is growing. At SURF, he and his colleagues have tracked that classical simulation and modeling workflows can lose as much as 25 to 30 percent of their compute time to manual data movement between storage tiers — a problem rooted not in hardware limits but in orchestration, legacy software, and workflow design. He makes the case that if the HPC community had designed systems from the application side outward, a convergent infrastructure meeting the needs of astronomy, high-energ ...