NVIDIA is currently accelerating innovation in science and engineering with its CUDA-X library at GTC 2025. CUDA-X leverages NVIDIA's latest GB200 and GH200 superchip architectures to enhance integration between CPUs and GPUs, and significantly improve computational performance. This enables up to 11x faster performance and 5x larger computations than traditional accelerated computing architectures12.
CPU-GPU Integration: CUDA-X enables memory sharing between CPUs and GPUs through NVIDIA's Grace CPU architecture and NVLink-C2C interconnect. This allows developers to solve bigger problems without writing any special code 12.
Domain-specific libraries: NVIDIA has developed more than 900 domain-specific CUDA-X libraries and AI models, facilitating the adoption of accelerated computing. It is utilized in a variety of fields, including astronomy, particle physics, quantum physics, automotive, aerospace, semiconductor design, etc.12.
Performance Improvements: NVIDIA's cuDSS library is used to solve large-scale engineering simulation problems, which are utilized for design optimization, electromagnetic simulation, and more. cuDSS efficiently handles large matrices through Grace GPU memory and NVLink-C2C interconnects12.
CUDA-X can be used in a variety of industries:
Engineering Simulations: Ansys and Altair OptiStruct significantly improve their performance in electromagnetic simulations and finite element analysis with cuDSS. Ansys' HFSS software achieved up to 11x speedup12.
Quantum Computing: NVIDIA's cuQuantum library accelerates quantum algorithm simulation to support quantum computing research. The GH200 architecture delivers up to three times faster performance than the H100 system 12.
CUDA-X will bring about innovative changes in science and engineering, helping developers and researchers solve complex problems in a faster and more efficient way. With CUDA-X, NVIDIA will accelerate the spread of accelerated computing and contribute to innovative achievements in various industries.
Source: https://blogs.nvidia.com/blog/cuda-x-grace-hopper-blackwell/
Share this post: