CUDA 13.0—New Features and Beyond | NVIDIA GTC D.C.
NVIDIA CUDA Team
NVIDIA GTC 2025 · Session
This talk, delivered by Rob from the NVIDIA CUDA team at GTC D.C., provides an in-depth look at the advancements and strategic directions introduced with **CUDA 13.0**, NVIDIA's latest major release of its parallel computing platform and programming model. Released in August, CUDA 13.0 marks a significant milestone, being the first full-bore release with comprehensive support for the **Blackwell architecture**. As a major version bump, it lays critical groundwork for future development, addressing fundamental changes and setting the stage for the next 18 months of CUDA evolution.
AI review
A competent and detailed rundown of CUDA 13.0's feature set from someone who clearly knows the platform. The CUDA Tile model is the genuinely interesting bit — it's a real architectural response to a real problem (tensor core portability breaking across generations). But this is a product announcement talk dressed up as an engineering talk, and the article summary doesn't help: it's thin on implementation specifics, benchmarks are hand-wavy, and there's nothing here that would let you actually use any of these features today. Worth watching if you're building kernels for Blackwell; probably…