At the Computex event in Taipei this week, Nvidia announced four new systems equipped with its Grace- and Hopper-generation hardware, including two in Taiwan. Those two are Taiwania 4, powered by Nvidia’s Grace CPU Superchip, and Taipei-1, based on Nvidia’s H100 GPUs. The others: Helios and Israel-1.
Taiwania 4
The first system, Taiwania 4, will be built by Asus and sited at the National Center for High-Performance Computing (NCHC) run by the National Applied Research Laboratories (NARLabs). It will consist of 44 nodes powered by the Grace CPU Superchip, which includes two Arm-based Grace CPUs in a single package. The nodes will be networked with Nvidia’s Quantum-2 InfiniBand. Taiwania 4, as one might expect, is set to succeed Taiwania 3 – an Intel Xeon “Cascade Lake”-based system that first appeared in 2020 – when it arrives in 2024.
At the NCHC, Taiwania 4 is expected to serve crucial needs mostly centered around atmospheric modeling, such as climate modeling, air-quality management and computational fluid dynamics. In keeping with these priorities, Nvidia and the NCHC are touting the energy efficiency that will come with the Grace Superchips – something Nvidia has been touting elsewhere, as well, as energy use becomes an ever-more pressing concern.
“The Arm-based NVIDIA Grace CPU will make Taiwania 4 among the most energy-efficient CPU-based supercomputers in Asia, in line with its mission to tackle the global climate crisis,” the NCHC said in a statement.
Taipei-1
The second system, Taipei-1, will be owned and operated by Nvidia. It will include 64 DGX H100 systems – each of which includes eight H100 GPUs (for a total of 512) – and 64 Omniverse-focused OVX systems, each of which includes four L40 GPUs and a BlueField-3 DPU. The system will use unspecified Nvidia networking, and it is expected to come online “later this year.”
Taipei-1 will have more general applications than Taiwania 4: the system is aimed at wide-ranging use cases like large language models (LLMs), healthcare, robotics, industrial digital twins and – like Taiwania 4 – climate science. For its first project, though, National Taiwan University (one of the system’s first users, among other leading Taiwan education and research institutes) will be using it to research speech learning by LLMs.
“National Taiwan University researchers are dedicated to advancing science across a broad range of disciplines, a commitment that increasingly requires accelerated computing,” said Shao-Hua Sun, assistant professor, Electrical Engineering Department at National Taiwan University. “The NVIDIA Taipei-1 supercomputer will help our researchers, faculty and students leverage AI and digital twins to address complex challenges across many industries.”
Helios & Israel-1
Also new: Nvidia’s Helios supercomputer, based on the DGX GH200 system and including 1,024 Grace Hopper Superchips (learn more about that system and the GH200 supercomputer infrastructure here). Nvidia is also building a generative AI-focused system called Israel-1 for its Israeli datacenter, which will use the Nvidia HGX H100 platform as well as BlueField-3 DPUs.
These are only the latest in an increasing barrage of system announcements from Nvidia, which also recently announced the Isambard 3 system – also based on the Grace CPU Superchip – for the UK. Also in the spotlight are MareNostrum 5, a EuroHPC system that will use Grace CPUs, as well as Venado (Los Alamos National Laboratory), Alps (CSCS) and Shaheen-III (KAUST), all of which will use the Grace Hopper Superchip.