NVIDIA Foundational Technology Montage I GTC Taipei and COMPUTEX 2026 Edition
NVIDIA built CUDA 20 years ago as a single architecture for accelerated computing to address the need for general-purpose hardware acceleration. That architecture is backed by a broad ecosystem of roughly a thousand CUDA‑X libraries that target domain-specific problems and are described as tools for agents. The talk lists concrete library examples—cuLitho, cuOpt, cuDSS, AIQ, Aerial, Warp, Parabricks—showing coverage across lithography, optimization, sparse solvers, document research, AI RAN, differentiable physics, and genomics. The stated foundation of this ecosystem is algorithms, which are presented as the core enabling element for breakthroughs across science and engineering.
Key technical points
- CUDA was developed 20 years ago as a single architecture for accelerated computing.
- CUDA is positioned as having reinvented computing by providing unified acceleration.
- There exists an ecosystem of approximately a thousand CUDA‑X libraries.
- CUDA‑X libraries are described as helping developers make breakthroughs across every field of science and engineering.
- The presentation frames CUDA‑X libraries explicitly as tools for agents.
- cuLitho is named as a CUDA‑X library for computational lithography.
- cuOpt is named as a CUDA‑X library for decision optimization.
- cuDSS is named as a CUDA‑X library for direct sparse solvers.
- AIQ is named as a CUDA‑X library for deep research across structured and unstructured documents.
- Aerial is named as a CUDA‑X library for AI RAN (radio access network) use cases.
- Warp is named as a CUDA‑X library for differentiable physics.
- Parabricks is named as a CUDA‑X library for genomics.
- The talk emphasizes that the foundation of these libraries is algorithms.
- The presenter characterizes those foundational algorithms as ’beautiful,’ emphasizing the centrality of algorithmic design.
Treat CUDA as a unified acceleration platform and the CUDA‑X collection as domain‑specific building blocks rather than general implementation recipes. The transcript provides explicit mappings from library name to problem domain (e.g., cuLitho → computational lithography, cuOpt → decision optimization, cuDSS → direct sparse solvers, AIQ → document research, Aerial → AI RAN, Warp → differentiable physics, Parabricks → genomics), which engineers can use to identify candidate libraries for workload acceleration. Note that the source specifies intent and coverage but does not include API details, performance metrics, evaluation methods, production guardrails, failure modes, or deployment patterns—those must be obtained from the corresponding library documentation and empirical validation. Finally, recognize the emphasis on algorithms: algorithmic design and selection are presented as the primary technical lever in this ecosystem.
For engineering teams in Japan, the cited CUDA‑X libraries provide explicit domain targets to evaluate for acceleration projects: lithography, optimization, sparse linear algebra, document research, RAN, differentiable physics, and genomics. The talk supplies names and domains but no implementation or operational details, so local adoption efforts should start by mapping project requirements to the listed libraries and then pursuing tests and integrations using each library’s official technical materials. Emphasize algorithmic work when planning improvements, since the presentation highlights algorithms as the foundational element of the CUDA‑X ecosystem.