search
CUDA
Trends
- 1Janus: Go binary runs GGUF models via Vulkan on any GPU●Show HN: Janus – Go binary that runs GGUF models via Vulkan on AMD/Intel/Nvidia
A developer has released Janus, an open-source Go binary that runs GGUF large language models through Vulkan, removing the need for CUDA and making it compatible with AMD, Intel and Nvidia GPUs. The project is shared on GitHub and is drawing attention on Hacker News, where users are discussing its potential to simplify local model inference across different hardware vendors.
- 2NVIDIA adopts Shibaura Institute computing method into CUDA●NVIDIA、芝浦工大発の計算手法をCUDAに採用。AI向けGPUの低精度演算器で「FP64相当」の計算を高速化 – ライブドアニュース https://www. yayafa.com/2901870/ # AgenticAi # AI #
NVIDIA has incorporated a computing method developed at Shibaura Institute of Technology into CUDA, its parallel computing platform. The technique uses low-precision arithmetic units on AI-oriented GPUs to accelerate calculations equivalent to FP64 double precision, promising faster scientific and AI workloads on consumer-grade hardware. The announcement, reported by Livedoor News, is drawing attention in Japanese tech and AI communities as a notable example of domestic Japanese research shaping GPU computing.
Repos
- magnitudedev/magnitude Open source inference engine for agents that optimizes itself for your exact hardware. Compiles and tunes its kernels on
- tile-ai/tilelang Domain-specific language designed to streamline the development of high-performance GPU/CPU/Accelerators kernels
- tensorflow/tensorflow An Open Source Machine Learning Framework for Everyone
- lostmsu/TurboGPT Train a tiny GPT in under a minute (CUDA only)