NVIDIA CUDA 13.3 Wants to Make It Easier to Squeeze GPUs: Tiles, Stable Python and up to 15% More Performance
This story is only covered by news sources that have yet to be evaluated by the independent media monitoring agencies we use to assess the quality and reliability of news outlets on our platform. Learn more here.3 Articles
3 Articles
With the release of CUDA 13.3, NVIDIA strengthens its GPU ecosystem on two important fronts. The Python version officially passes in 1.0 (so considered stable and usable in production), and CUDA Tile natively arrives for C++ developers. Small reminder for non-initiated: CUDA, this is the tool that everyone uses to rotate calculation on NVIDIA graphics cards, mainly for AI and scientific computation. Historically, it's C/C++ at 99%. NVIDIA has be…
NVIDIA already started with CUDA 13.1 to move its platform towards a way of programming the GPUs less tied to the classic model of working thread by thread. Specifically, and as a great novelty at the end of last year, we had CUDA Tile as the first serious warning to organize the work in more manageable blocks so that writing fast kernels does not always force you to go down to the hardest detail of the hardware. With CUDA 13.3, that idea grows …
NVIDIA has released CUDA 13.3, continuing the overhaul of its developer platform. The most notable new feature is CUDA Tile for C++, which, with the new toolkit, is designed to be directly integrated into existing C++ CUDA codebases. Other additions include CUDA Python 1.0, CompileIQ autotuning for critical kernels, C++23 support in NVCC and NVRTC, and updates to libraries and profiling tools. According to NVIDIA, […]
Coverage Details
Bias Distribution
- There is no tracked Bias information for the sources covering this story.
Factuality
To view factuality data please Upgrade to Premium
