As the largest AI company in the world and currently surpassing Amazon and Google as the fourth largest on the planet in all areas, NVIDIA has showcased its strength by advertising its new supercomputer. Known as NVIDIA Eos DGX SUPERPOD, what NVIDIA has created is a beast that, while not the fastest in the world in raw power, is indeed the fastest in Artificial Intelligence. With its launch, Huang’s team displays the potential of AI, both for their company and for those that require its online computing capacity.
The numbers are truly impressive, but not enough to compete with the global TOP 3, although that is not obviously its goal. Huang wants to compete in the sector he dominates, and there, this supercomputer has no rival, making use of an incredible number of DGX H100 systems.
NVIDIA Eos DGX SUPERPOD: Out-of-this-world numbers for AI
This supercomputer is not aimed at private use by NVIDIA, although the company will certainly use it as well. Its main goal is to offer its computing capacity to businesses, making it accessible from any point in the world and allowing us to move around the planet without having hardware at our destination.
NVIDIA itself encourages the idea, stating that “every day, Eos rises to face the challenges of thousands of NVIDIA’s internal developers working on AI research, and helps them solve what was previously unsolvable.”
Looking at the list of the 500 fastest Supercomputers in the world, NVIDIA Eos ranks ninth. The list includes a total of 485,888 cores, which we cannot fully understand, as the specifications do not match up.
NVIDIA talks about 576 DGX H100 systems, and each one integrates 8 H100 GPUs and two Intel Xeon Platinum 8480C, which equates to 112 Cores in total at 3.8 GHz each. Therefore, the figure to consider should be 69,120 cores, assuming that each GPU would be a unit of that count.
Outstanding performance
Regardless of the core count, the performance that NVIDIA Eos achieves is impressive. It scores 121.40 pflops/s in Linpack, with a theoretical maximum peak of 188.65 PFlops/s and an Nmax of 5,484,544. This power would be nothing without the corresponding software. In the video, NVIDIA gives many hints about what we can do with it:
EOS has an integrated software stack that includes AI development and deployment software, cluster orchestration and management, accelerated computing, storage and networking libraries, and an operating system optimized for AI workloads.
EOS, built on the knowledge gained from previous NVIDIA DGX supercomputers such as Saturn 5 and Seline, is the latest example of NVIDIA’s AI expertise in action.
So, it seems that we can create anything from a simple Chatbot to a specialized AI factory, as also shown in the video. Without a doubt, NVIDIA has raised the bar again, focusing on cloud and centralized computing. If we cannot access their GPUs or if we have an AI project that does not warrant the hardware investment, NVIDIA Eos is currently the best option for such cases, at a price that is still unknown.
The entry NVIDIA Eos, the supercomputer for AI with 4,608 H100 GPUs, has entered the top 10 global ranking, first appearing in El Chapuzas Informático.
Need Help?
Can't Find What Your Looking For? Just fill out our form and we'll ask our trusted distributors