Cublas
PulseAugur coverage of Cublas — every cluster mentioning Cublas across labs, papers, and developer communities, ranked by signal.
4 day(s) with sentiment data
-
LLM framework AUTO automates GPU code optimization, beats baselines
Researchers have developed AUTO, a novel framework that leverages large language models (LLMs) for automated design optimization. This system employs a Strategist agent for high-level planning and multiple Implementor a…
-
StitchCUDA framework automates end-to-end GPU programming with multi-agent RL
Researchers have developed StitchCUDA, a novel multi-agent framework designed for end-to-end GPU program generation. This system employs specialized agents for planning, coding, and verification to optimize machine lear…
-
Nova compiler unlocks deep learning performance with aggressive optimization
Researchers have developed Nova, an end-to-end Just-In-Time (JIT) compiler designed to optimize deep learning model performance by gaining granular control over hardware and memory. Nova achieves this by fusing operatio…
-
New benchmarks and optimization tools emerge for LLM-generated hardware kernels
Two new research papers introduce benchmarks and optimization frameworks for large language models (LLMs) generating code for hardware accelerators. The first paper, KernelGenBench, offers a unified benchmark to evaluat…
-
InstantSfM offers GPU-native structure-from-motion for deep learning era
Researchers have developed InstantSfM, a novel GPU-native structure-from-motion system designed to integrate seamlessly with deep learning pipelines. This system addresses the limitations of traditional CPU-centric SfM …
-
AI system CUDA-L2 surpasses NVIDIA's cuBLAS for matrix multiplication
Researchers have developed CUDA-L2, a system that leverages large language models and reinforcement learning to automatically optimize matrix multiplication CUDA kernels. This system significantly outperforms existing b…
-
New CTA-pipelining method slashes multi-GPU latency for LLMs
Researchers have introduced CTA-pipelining, a novel execution paradigm for multi-GPU systems that optimizes for latency in serving large language models. This method exploits dependencies at the Cooperative Thread Array…
-
New kernels boost LLM inference speed by fusing SwiGLU activations
Researchers have developed new techniques to accelerate the inference of large language models (LLMs) by fusing SwiGLU activation functions directly into GEMM operations at the tile level. These methods, implemented usi…
-
AutoMegaKernel compiles Llama models into single CUDA kernels
Researchers have developed AutoMegaKernel (AMK), a system that compiles HuggingFace Llama-family models into a single, persistent CUDA kernel for efficient forward passes. AMK's static validator ensures schedule safety,…
-
GPU Matrix Multiplications Faster With Predictable Data
Researchers have discovered that matrix multiplications on GPUs can perform faster when the input data is "predictable." Initially, a project called CUTLASS showed a 10% performance improvement over NVIDIA's CuBLAS. How…
-
DeepSeek V4 achieves faster performance with custom kernels, replacing cuBLAS
DeepSeek has developed a custom kernel stack, DeepGEMM and TileLang, which not only matches but surpasses the performance of NVIDIA's cuBLAS. This custom implementation achieves bitwise determinism and batch invariance,…