t cht lk
BUILDING AND MANAGING AN AI DATA CENTRE
AI hardware evolves at such a pace, it makes infrastructure investment and ROI a moving target. Power and cooling become existential questions, particularly in regions where energy costs are high. Designing and running an AI-optimised facility requires a blend of infrastructure, AI, and software expertise that few organisations have in-house. Top executives from ManageEngine, Qlik, VAST Data, Vertiv share their insights.
Data centres are evolving to address the next generation of AI workload demands, which are distinctly different from traditional computing tasks. They require massive parallel processing, high memory bandwidth, and low-latency interconnects to handle large-scale data sets and complex computations efficiently.
Advanced liquid cooling methods like immersion or direct-to-chip systems are becoming crucial, especially for graphics processing units, GPUs such as NVIDIA’ s H100, which requires 700W thermal design power. Custom AI chips such as tensor processing unit and ASICs improve efficiency, while technologies like HBM and NVLink boost data transfer and processing speeds throughout.
The conversation is no longer about raw compute power; it is about architecting infrastructure that can handle unpredictability of AI workloads.
“ Sustainability remains a key goal, with hyperscalers aiming to maintain power usage effectiveness, PUE under 1.2 for optimal and efficient usage through AI-based energy optimisations,” says Rajkumar Vijayarangakannan, Lead Network Design and DevOps, ManageEngine.
www. intelligentcio. com INTELLIGENTCIO MIDDLE EAST 79