FPT imports NVIDIA's DGX H100 system into Vietnam

On May 9, 2024, FPT imported Vietnam's first DGX H100 server system, marking the initiation of a strategic collaboration with NVIDIA to advance artificial intelligence (AI) development in the country.

FPT has begun executing the memorandum of cooperation signed with NVIDIA at the end of April. It involves the implementation of the AI Factory project, which aims to develop a sovereign AI and Cloud ecosystem in Vietnam.

FPT has successfully imported Vietnam's first DGX H100 servers

This is the most advanced server system in NVIDIA's lineup. Renowned as the most powerful AI machine available today, the DGX H100 is in high demand among leading technology companies such as Tesla, Meta, and Microsoft. Since late 2023, the demand for these servers has surged, leading to a significant backlog. According to an analysis report from Omdia, businesses now face a waiting period of 36 to 52 weeks when placing orders with NVIDIA.

A representative from FPT Smart Cloud, a subsidiary of FPT responsible for project implementation, emphasized that the import of this system marks the initial step toward advancing technology integration and establishing supercomputers dedicated to supporting AI development. It also represents the first phase in implementing the plan to construct an AI Factory. Additional DGX-H100 systems will be imported to Vietnam shortly to enhance technological capabilities, expedite product research and development, and lay the groundwork for next-gen AI and Cloud solutions and services.

In collaboration with NVIDIA for the AI Factory construction project in Vietnam, FPT aims to offer GPU cloud services. This initiative seeks to enable FPT's global business clientele to access essential resources for enhancing research capabilities and accelerating AI application development, particularly in Generative AI. The goal is to achieve significant breakthroughs in productivity enhancement and customer experience improvement.

Each DGX H100 server comprises 8 H100 GPUs interconnected via NVIDIA NVLink, along with dual Intel Xeon Platinum 8480C processors, totaling 112 cores and 2TB of system memory, ideal for handling the most demanding AI tasks. Additionally, the servers feature two NVIDIA BlueField-3 DPUs designed to offload, accelerate, and isolate advanced network, storage, and security services.

The DGX H100 delivers six times higher performance and twice as fast networking compared to previous-generation servers. This capability addresses the immense computing demands associated with large language models (LLMs), which often involve processing trillions of parameters, recommendation systems, medical research, and climate science.

The DGX H100 plays a crucial role in constructing a refined, systematic, and scalable AI platform, enabling businesses to process language, analyze data, and more. It is the foundational system for creating Eos, NVIDIA's supercomputer, which ranks 9th among the world's fastest supercomputers. NVIDIA asserts that the DGX H100 has the capability to transcend the boundaries of AI technology and infrastructure.

The DGX H100 server has found applications across various corporations and technology firms globally. Tesla, for instance, utilizes the H100 GPU platform to develop self-driving vehicles. In Japan, startup Xeureka employs the DGX-H100 in drug research and development processes, enhancing medical examination and treatment activities. In Ecuador and Sweden, a plethora of projects in the financial, healthcare, legal, IT, and telecommunications sectors leverage AI on the DGX-H100 platform.