NVIDIA and RIKEN power Japan’s AI & quantum supercomputing
The Age of AI Will Drive Supercomputing with Networking and Accelerated Computing.
NVIDIA introduced many advancements in accelerated computing, next-generation networking, and quantum systems at SC25 to show the rapid growth of AI supercomputing worldwide.
Ian Buck, NVIDIA's general manager and accelerated computing VP, presented a special keynote. NVIDIA announced new AI physics models, quantum jumps through NVQLink, Quantum-X Photonics InfiniBand CPO networking switches, and BlueField-4 DPU.
NVIDIA-RIKEN Partnership Advances Japanese Science
NVIDIA is working with Japan's leading scientific institute, RIKEN, to build two GPU-accelerated supercomputers. These systems will strengthen Japan's quantum computing and AI for science leadership.
Through the GB200 NVL4 platform and NVIDIA Quantum-X800 InfiniBand networking, 2,140 NVIDIA Blackwell GPUs will be connected to boost Japan's secure domestic infrastructure and sovereign AI goal.
The two new systems include:
AI for Science System: 1,600 Blackwell GPUs enable climate and weather forecasting, materials science, life sciences, manufacturing, and laboratory automation research.
Quantum computing system: 540 Blackwell GPUs accelerate quantum-classical, hybrid, and quantum algorithms.
A spring 2026 launch of these supercomputers is anticipated.
This alliance expands RIKEN, Fujitsu, and NVIDIA's codesign of FugakuNEXT, the Fugaku supercomputer's successor. Production-level quantum computers will be incorporated by 2030, and FugakuNEXT will improve application performance 100x. New GPU-accelerated supercomputers will codesign and develop FugakuNEXT's hardware, software, and applications.
The RIKEN Centre for Computational Science's director, Satoshi Matsuoka, said that integrating the NVIDIA GB200 NVL4 accelerated computing platform with their next-generation supercomputers will help Japan's scientific infrastructure become one of the world's top unified platforms for AI, quantum, and high-performance computing.
Linking Quantum and Classical Computing with NVQLink Over a dozen of the world's top scientific computing centres employ NVQLink to connect quantum processors for accelerated computation. Ian Buck claims that NVIDIA is developing the next generation of quantum GPU, CPU GPU supercomputers with global supercomputing centres.
Open architecture NVQLink connects quantum processors and NVIDIA GPUs to power large-scale processes utilising CUDA-Q. Supercomputing centres can incorporate several quantum processors due to this important connectivity. Its AI performance is 40 petaflops with FP4 accuracy.
Notable NVQLink technical achievements include:
By combining its new Helios QPU with NVIDIA GPUs through NVQLink, quantum computing company Quantinuum decoded scalable qLDPC quantum error-correction codes in real time.
Without NVQLink correction, system integrity was 95%; with it, 99%. The system could react in 60 microseconds, 16 times faster than Helios' 1-millisecond limit. A decoder implementation has a reaction time of 67 microseconds, 32 times faster than Helios' two-millisecond threshold, according to another source.
A universal bridge between quantum and conventional hardware, NVQLink lets researchers and developers design real-time quantum-GPU operations, hybrid systems, and scalable error correction.
Adopting facilities worldwide:
The G-QuAT (AIST) and RIKEN Centre for Computational Science in Japan, the Pawsey Supercomputing Research Centre in Australia, the NCHC in Taiwan, KISTI in Korea, and the National Quantum Computing Hub in Singapore are Asia-Pacific.
Germany's Jülich Supercomputing Centre (JSC), Saudi Arabia's KAUST, the Czech Republic's IT4I, Italy's CINECA, Denmark's DCAI, France's GENCI, and Poland's PCSS.
Top US national laboratories include MIT Lincoln, NERSC, Oak Ridge, Pacific Northwest, Sandia, Lawrence Berkeley, Fermi, Brookhaven, and Los Alamos.
AI Factories Use Photonics Networking and BlueField-4 BlueField-4 DPU, presented by NVIDIA, powers AI factories' operating system. Networking, storage, and security are offloaded, accelerated, and isolated by BlueField-4 DPUs to free CPUs and GPUs for compute-intensive workloads. With NVIDIA ConnectX-9 networking and a 64-core NVIDIA Grace CPU, BlueField-4 delivers excellent performance, efficiency, and zero-trust security at scale.
Innovative storage companies like DDN, VAST Data, and WEKA are using BlueField-4 to improve AI and scientific workload performance. The DPU runs storage services on BlueField-4, where WEKA is implementing its NeuralMesh architecture.
TACC, Lambda, and CoreWeave plan to add NVIDIA Quantum-X Photonics InfiniBand CPO networking switches to next-generation systems next year. These switches reduce energy and operating expenses in supercomputing facilities and AI factories. Through direct optics integration and the removal of pluggable transceivers, which often cause task runtime issues, NVIDIA Photonics switch systems:
Increase power efficiency 3.5 times.
Perform with ten times more resilience.
Make apps run five times longer uninterrupted.
In order to grow huge AI workloads, NVIDIA Quantum X Photonics is improving power economy and reliability, according to CoreWeave co-founder and chief technology officer Peter Salanki.
New software and systems
NVIDIA also introduced hardware and software to accelerate physics and AI simulations:
The world's smallest AI supercomputer, the DGX Spark, was released by NVIDIA. This desktop compact has 128GB of unified memory and a petaflop of AI performance. Developers can directly fine-tune and infer models with up to 200 billion parameters using the Grace Blackwell architecture.
At SC25, NVIDIA Apollo introduced open AI Physics models. Models like these help Applied Materials, Cadence, Siemens, and Synopsys simulate and speed up weather, semiconductor, and computational fluid dynamics design. Apollo uses domain-specific knowledge, diffusion, transformers, and neural operators.
NVIDIA Warp: An open-source Python framework for AI and computational physics with up to 245x GPU acceleration. Warp simplifies high-performance simulation workflow development with Python's accessibility and CUDA-level performance. Luminary Cloud, Siemens, and Neural Concept use NVIDIA Warp.
Arm Adopts NVLink Fusion
Arm's Neoverse architecture is adding NVIDIA NVLink Fusion, a high-bandwidth, coherent connection debuted with Grace Blackwell. NVLink Fusion integrates CPUs, GPUs, and accelerators into a rack-scale architecture to eliminate memory and bandwidth constraints that limit AI performance. This integration uses Arm's AMBA CHI C2C protocol to ensure smooth data transfer between CPUs and accelerators.
This agreement sets a new AI infrastructure standard, enabling ecosystem participants to design energy-efficient solutions. Ian Buck said that 'Folks developing their own ARM CPU, or using an Arm IP can really have access to NVLink Fusion, connect that ARM CPU to an Nvidia GPU or the rest of the NVLink ecosystem'.
Comparative Analysis
Integrating quantum processors with GPU supercomputers with NVQLink is like building a universal translator over a rapid fiber-optic connection. Unlike the GPU, the quantum processor speaks quantum physics, the language of nature. NVQLink's ultra-low latency connection and translation layer (CUDA-Q) APIs let these two very different systems to collaborate in real time on difficult tasks like mistake correction, speeding up and improving scientific results.
