Clarke's remarks also touched upon the industry's move towards more efficient cooling solutions, indicating a shift towards enabling large-scale water cooling to manage the heat generated by high-power GPUs. This development marks a notable increase in power consumption from Nvidia's current top AI accelerator card, the H100, which has a maximum power draw of 700W. The B200 "Blackwell" accelerator, which was not previously outlined in Nvidia's announced roadmap, represents a significant step forward in GPU processor technology. Details about the B100, presumably a predecessor to the B200, remain sparse, with further information anticipated to be released at Nvidia's upcoming developer conference.
This announcement underscores the rapid advancements in GPU technology and the increasing demands on power and cooling infrastructure in data centers, highlighting the ongoing efforts within the industry to address these engineering challenges.
Source: ithome