per GW.
Harnessing NVIDIA’s AI Power
NVIDIA’s current generation H100 AI GPU is deployed far and wide, providing an expansive reach with an estimated 3.5 million units expected to be shipped in 2024. Stock Talk’s market analyse reported an outrageous figure, where the power consumption of the H100 AI GPUs plus future NVIDIA products will qualify as a Countries ‘energy bill’, consuming 13,000 GWh of energy annually where the likes of Guatamela and Lithuania carry 8,953 and 8,567 electricity demand. This has been a topic of Weekly News before, citing back in 2020 just a single Bitcoin mining farm in Kazakhstan had the power Constrint of 180,000 people – an analogue echoed to those experience registrations of power crunch within Argentina and Holland with intense Crypto activities.
NVIDIA brings solutions however with impressive 90% market share of the industry, the A100 and H100 AI GPUs have been seen driving the newestinstallations like the Jupiter Supercomputer, integrated with some impressive features such as 93 ExaFLOPS AI performance rendered by GH200 Grace Hopper Superchips (complemented with 24,000 instances of independent operation), harnessing Keyboard-ConnectX Network rail with 1.2PB Per Second of speed., while carrying the 1.0 EF turnover of a typical HPC facility with only 18.2W accuneted pwer watts spent.
This account is laid out to be supplemented by the B100 AI GPU equipped with Blackwell Istrucutre and emerging processing solutions, compromising improved consumption of valuable Electricity ; while A tremendous spike up to 1.5 – 2million gpu in 2024 as projected. Jupiters Supercomputer will now become a viable factor to Scientific discoveries in morphology, Dualities studies and mode research to just owning stock. Phew.
