As AI products gain popularity among consumers in the market, products for AI servers due to the increasing AI-driven data transmission and computing power are appearing one after another.

NVIDIA announced on the 3rd that it has begun commercializing the NVIDIA DGX H100 system. It is reported that the DGX H100 system, the latest and highest-spec data center product that was officially launched earlier this year, has begun commercialization with various customers amid the competitive AI market.

The DGX H100 system features eight NVIDIA H100 GPUs, each equipped with a Transformer Engine designed to accelerate generative AI models. Delivering an average of nearly six times more performance than the previous-generation DGX A100, the DGX H100 system features up to 640 GB of total GPU memory, 18 NVIDIA NVLinks per GPU and 900 GB/s of GPU-to-GPU bidirectional bandwidth.

The new system features two 400Gbps network interfaces, allowing hundreds of DGX H100 nodes to be connected into an AI supercomputer using NVIDIA Quantum InfiniBand, which is twice as fast as the previous generation and has ultra-low latency. It also doubles the kilowatts per petaflop of the previous generation.

The NVIDIA DGX H100 system, which can be utilized in smart manufacturing and smart factories around the world, is a product expected to be applied to various applications based on deep learning technology, such as smart valet parking, generative AI and LLM, and healthcare solutions.

The DGX H100 system includes NVIDIA AI Enterprise software to accelerate data science pipelines and simplify the development and deployment of generative AI, computer vision, and more.

Manuvir Das, vice president of enterprise computing at NVIDIA, announced the commercial availability of the DGX H100 system during a talk at MIT Technology Review's Future Compute event.

It is reported that customers have already adopted or are considering adopting the DGX H100 system. Boston Dynamics AI Institute, a research arm of Boston Dynamics, a leader in robotics, announced that it is using DGX H100.

Initially, the DGX H100 will handle tasks related to reinforcement learning, a core technology in robotics, and will later be connected directly to prototype bots in the lab to run AI inference tasks. “It is possible to easily develop and deploy AI models by integrating high-performance computers in a relatively compact space,” added AI Lab CTO Al Ridge.

In Germany, DeepL will use several DGX H100 systems to scale services such as translation between dozens of languages for clients including Nikkei, Japan’s largest publisher. DeepL recently launched DeepL Write, an AI writing assistant.

Meanwhile, as the AI market is becoming more active through LLM products such as ChatGPT and generative AI service products, the demand for AI servers in the hardware sector is rapidly increasing. As demand for high-performance server GPUs for AI operations increases, demand for high-capacity, high-performance memory such as HBM3 and DDR5 is also increasing.