✳flâneur — a map of the web's best reading

Nice Quarter Guys on X: "@aniruddh_mohan Fluctuating by ~25 MW is crazy right? Like 20,000 houses turning on" / X

x.com · 37 words · saved by 1 readers

To view keyboard shortcuts, press question mark View keyboard shortcuts More details Quote Aniruddh Mohan @aniruddh_mohan · Jul 10 Fascinating high resolution load data for an AI training data center (50 MW block) Periodic idling & surges - 6MW to 30MW spike in 290ms(!) - fluctuating demand followed by drop then spikes again Why does AI training load vary so much? - GPUs process data in batches, so power The graph from Aniruddh Mohan’s post shows AI training data center power consumption fluctuating between 6MW and 30MW in 290ms, driven by GPU batch processing, as confirmed by studies like those from MIT Sloan (2025) highlighting how batch sizes in AI models cause rapid energy spikes. These swings, equivalent to 20,000 households turning on/off, challenge traditional grid stability, necessitating synchronous generators and batteries, a solution supported by Xylenepower.com’s analysis of synchronous generation for rapid load changes. Skepticism from Josh Payne suggests the extreme va

Nice Quarter Guys @NiceQuarterGuys Replying to @aniruddh_mohan Fluctuating by ~25 MW is crazy right? Like 20,000 houses turning on 2:31 PM · Jul 10, 2025 1.7K Views 2 0 2 20 0 2 0 Read 2 replies

Explore this link on the map →

related reading