Enhancing Japan’s AI sovereignty and strengthening its analysis and improvement capabilities, Japan’s Nationwide Institute of Superior Industrial Science and Know-how (AIST) will combine 1000’s of NVIDIA H200 Tensor Core GPUs into its AI Bridging Cloud Infrastructure 3.0 supercomputer (ABCI 3.0). The HPE Cray XD system will characteristic NVIDIA Quantum-2 InfiniBand networking for superior efficiency and scalability.
ABCI 3.0 is the newest iteration of Japan’s large-scale Open AI Computing Infrastructure designed to advance AI R&D. This collaboration underlines Japan’s dedication to advancing its AI capabilities and fortifying its technological independence.
“In August 2018, we launched ABCI, the world’s first large-scale open AI computing infrastructure,” stated AIST Government Officer Yoshio Tanaka. “Constructing on our expertise over the previous a number of years managing ABCI, we’re now upgrading to ABCI 3.0. In collaboration with NVIDIA we goal to develop ABCI 3.0 right into a computing infrastructure that may advance additional analysis and improvement capabilities for generative AI in Japan.”
“As generative AI prepares to catalyze international change, it’s essential to quickly domesticate analysis and improvement capabilities inside Japan,” stated AIST Options Co. Producer and Head of ABCI Operations Hirotaka Ogawa. “I’m assured that this main improve of ABCI in our collaboration with NVIDIA and HPE will improve ABCI’s management in home business and academia, propelling Japan in direction of international competitiveness in AI improvement and serving because the bedrock for future innovation.”
ABCI 3.0: A New Period for Japanese AI Analysis and Improvement
ABCI 3.0 is constructed and operated by AIST, its enterprise subsidiary, AIST Options, and its system integrator, Hewlett Packard Enterprise (HPE).
The ABCI 3.0 mission follows assist from Japan’s Ministry of Economic system, Commerce and Business, referred to as METI, for strengthening its computing assets by the Financial Safety Fund and is a part of a broader $1 billion initiative by METI that features each ABCI efforts and investments in cloud AI computing.
NVIDIA is intently collaborating with METI on analysis and training following a go to final 12 months by firm founder and CEO, Jensen Huang, who met with political and enterprise leaders, together with Japanese Prime Minister Fumio Kishida, to debate the way forward for AI.
NVIDIA’s Dedication to Japan’s Future
Huang pledged to collaborate on analysis, notably in generative AI, robotics and quantum computing, to put money into AI startups and supply product assist, coaching and training on AI.
Throughout his go to, Huang emphasised that “AI factories” — next-generation knowledge facilities designed to deal with essentially the most computationally intensive AI duties — are essential for turning huge quantities of knowledge into intelligence.
“The AI manufacturing unit will turn into the bedrock of contemporary economies internationally,” Huang stated throughout a gathering with the Japanese press in December.
With its ultra-high-density knowledge middle and energy-efficient design, ABCI gives a sturdy infrastructure for creating AI and large knowledge purposes.
The system is anticipated to come back on-line by the tip of this 12 months and provide state-of-the-art AI analysis and improvement assets. It will likely be housed in Kashiwa, close to Tokyo.
Unmatched Computing Efficiency and Effectivity
The power will provide:
- 6 AI exaflops of computing capability, a measure of AI-specific efficiency with out sparsity
- 410 double-precision petaflops, a measure of basic computing capability
- Every node is related through the Quantum-2 InfiniBand platform at 200GB/s of bisectional bandwidth.
NVIDIA expertise kinds the spine of this initiative, with a whole bunch of nodes every outfitted with 8 NVLlink-connected H200 GPUs offering unprecedented computational efficiency and effectivity.
NVIDIA H200 is the primary GPU to supply over 140 gigabytes (GB) of HBM3e reminiscence at 4.8 terabytes per second (TB/s). The H200’s bigger and sooner reminiscence accelerates generative AI and LLMs, whereas advancing scientific computing for HPC workloads with higher power effectivity and decrease whole price of possession.
NVIDIA H200 GPUs are 15X extra energy-efficient than ABCI’s previous-generation structure for AI workloads equivalent to LLM token era.
The combination of superior NVIDIA Quantum-2 InfiniBand with In-Community computing — the place networking gadgets carry out computations on knowledge, offloading the work from the CPU — ensures environment friendly, high-speed, low-latency communication, essential for dealing with intensive AI workloads and huge datasets.
ABCI boasts world-class computing and knowledge processing energy, serving as a platform to speed up joint AI R&D with industries, academia and governments.
METI’s substantial funding is a testomony to Japan’s strategic imaginative and prescient to boost AI improvement capabilities and speed up using generative AI.
By subsidizing AI supercomputer improvement, Japan goals to cut back the time and prices of creating next-generation AI applied sciences, positioning itself as a frontrunner within the international AI panorama.