Nvidia introduced the “world’s strongest synthetic intelligence supercomputer”

Nvidia, a number one producer out there of video playing cards (GPU), pc chips and computing {hardware}, offered new developments on the Supercomputing 2023 occasion. Amongst them stands out JUPITER, described as “essentially the most highly effective synthetic intelligence supercomputer on the planet ».

JUPITER is a brand new class of supercomputers powered by the Nvidia Grace Hopper accelerated computing structure.whose first installations are projected for 2024. Particularly, the supercomputer will use 24,000 NVIDIA GH200 chips, one other of the merchandise that the corporate unveiled on the occasion held in Denver, Colorado.

This would be the “strongest synthetic intelligence system on the planet,” in line with the corporate, with a efficiency of greater than 90 exaflops for AI coaching and 1 exaflop for high-performance (HPC) purposes, consuming solely 18.2 megawatts of energy . An exaflop is a unit of measurement of computing efficiency, particularly within the subject of supercomputing. It represents the power to carry out one billion (10^18) floating level operations per second.

Nvidia emphasizes the potential use of this supercomputer to deal with scientific challenges in areas comparable to local weather prediction, drug discovery, quantum computing applied sciences and industrial design.

The corporate’s announcement, ratified and deepened in an official assertion later, confirms Nvidia’s orientation in direction of applied sciences comparable to synthetic intelligence and quantum computing. Lately, the corporate primarily based in California, United States, was a central participant within the subject of GPUs used for mining cryptocurrencies, primarily Ethereum.

New Nvidia chips for high-performance computing

Along with the JUPITER venture that can formally be launched subsequent yr, Nvidia offered different developments centered on synthetic intelligence and quantum computing. Quantum computing is a paradigm in pc science that makes use of ideas of quantum mechanics to course of info in a different way than classical computer systems. In contrast to classical bits, that are binary and will be in a state of 0 or 1, quantum qubits can exist in a number of states concurrently because of quantum superposition.

Concerning Nvidia displays, we speak in regards to the H200 chips, for instance. That is an improve to Nvidia’s GPU Hopper chips, with improved efficiency and higher price and energy effectivity. The Nvidia H200 chips introduce HBM3e reminiscence, which is quicker and better capability.

With 14 GB of reminiscence and the power to course of 4.8 terabytes per second, the H200 chips goal to speed up industrial design, simulations and particularly generative synthetic intelligence purposes and huge language fashions.

However, the Grace Hopper Superchips had been unveiled, the results of the mix of Nvidia’s Grace and Hopper applied sciences. The GH200, a part of the Grace Hopper sequence, is utilized in analysis and computing-intensive duties, with a mixed efficiency of 200 exaflops of AI efficiency per second. All these chips are utilized in supercomputers made by massive producers comparable to Dell and HPE.

Lastly, Nvidia introduced the event of quantum processing models (QPUs). On this sense, the open programming mannequin known as CUDA Quantum stands out, whose objective is to facilitate the mixing of quantum processors in future supercomputers.

Bitcoin mining pivots in direction of AI

The truth that Nvidia, as soon as a pacesetter in manufacturing {hardware} for cryptocurrency mining, focuses on synthetic intelligence is one other instance of a pattern that was lately reported in CriptoNoticias. The factor is that synthetic intelligence is an increasing enterprise, as had been cryptocurrency networks on the time. There, lTraders see enterprise alternatives and supply them with processing energy for his or her operations.

An exemplary case is that of Hive Digital, one of many principal Bitcoin mining firms on the planet. In response to what its CEO, Frank Holmes, instructed this newspaper, the funding in processing duties with synthetic intelligence allowed the corporate to earn greater than USD 250,000 in a single quarter with only a few GPUs devoted to duties comparable to coaching neural networks, calculations intensive or picture and video processing.