Get the latest tech news
Gpu.cpp: A lightweight library for portable low-level GPU computation
A lightweight library for portable low-level GPU computation using WebGPU. - GitHub - AnswerDotAI/gpu.cpp: A lightweight library for portable low-level GPU computation using WebGPU.
CUDA has been dominant at large scale training and inference but at the other end of the the spectrum in the world of GPU compute on personal devices, there exists far more heterogeneity in the hardware and software stack. gpu.cpp lets us implement and drop-in any algorithm with fine-grained control of data movement and GPU code, and explore outside boundaries of what is supported by existing production-oriented inference runtimes. At the same time we can write code that is portable and immediately usable on a wide variety of and GPU vendors and compute form factors - workstations, laptops, mobile, or even emerging hardware platforms such as AR/VR and robotics.
Or read this on Hacker News