#CUDA kernel optimization
113 related articles
GitHub Daily · July 19: The Dual Advan…
·3 min
GitHub Daily · July 19: The Dual Advance of Heterogeneous Inference and Agent Toolchains
GitHub Trending July 19: ktransformers tops the list with heterogeneous inference optimization, while jcode, cua, and AstrBot signal a maturing Agent ecosystem.
Read more →
Block Low-Rank Compression: A Guide to…
·2 min
Block Low-Rank Compression: A Guide to Accelerating Large Model Inference on Memory-Constrained GPUs
Learn how Block Low-Rank (BLR) decomposition compresses large model memory usage and accelerates GPU inference, including CUDA kernel optimization and combination with quantization and pruning.
Read more →
·2 min
19-Year-Old AI Learner: How to Balance Passion and Employability?
A 19-year-old AI learner torn between passion for LLMs and job market pressure. This article breaks down AI Engineering vs. research paths and offers actionable strategies.
Read more →
·3 min
Bonsai-27B Extreme Quantization: Running 1-bit LLMs on Consumer Hardware
Bonsai-27B supports binary/ternary extreme quantization for 27B LLMs running on 8GB VRAM. Covers llama.cpp upstream progress, RTX 4060 benchmarks (30 t/s), and real-world limitations.
Read more →
·3 min
GPUStack in Action: Complete Guide to Day 0 Deployment of DeepSeek-V4-Flash
Deploy DeepSeek-V4-Flash DSpark on 8× H20-141G using GPUStack's SGLang backend on Day 0. Full walkthrough of Web UI config, parameter tuning, and 200 tokens/sec benchmark results.
Read more →
·2 min
ExLlamaV3 v1.0.0 Official Release: Faster Inference, Lossless KV Quantization, and Streamlined Dependencies
ExLlamaV3 v1.0.0 releases with lossless KV cache quantization via kernel fusion, removal of flash-attention-2/xformers, major GEMM/GEMV gains, and broader tensor parallelism support.
Read more →
·2 min
GPU Sharing in Kubernetes Production: A Deep Dive into MIG, MPS, and Time-Slicing
A deep dive into MIG, MPS, and Time-Slicing GPU sharing solutions for Kubernetes production. Compare isolation, performance, and use cases to make the right choice.
Read more →
·2 min
llama.cpp Milestone: The Open-Source Evolution of Local LLM Inference
llama.cpp hits a new milestone, growing from a solo hobby project into core local AI inference infrastructure. Explore its iteration speed, GGUF quantization, and how AI coding agents are reshaping open-source development.
Read more →
·3 min
Is Your Training Job Quietly Slowing Down? TraceML Exposes PyTorch's Hidden Performance Bottlenecks
GPU at 51% utilization — and no one noticed? See how TraceML exposes hidden PyTorch DataLoader bottlenecks, cuts training time 43% with 3 parameter changes.
Read more →
4 Alternatives for Running CUDA on Non…
·3 min
4 Alternatives for Running CUDA on Non-NVIDIA Hardware: A Comparative Guide
A deep dive into running CUDA on non-NVIDIA hardware (AMD, Intel): comparing ROCm/HIP, ZLUDA, SYCL/oneAPI, and OpenCL across principles, use cases, and limitations.
Read more →
CUDA Kernel Fusion: A Practical Guide …
·2 min
CUDA Kernel Fusion: A Practical Guide to Reducing Memory Bandwidth and Launch Overhead
Learn how CUDA Kernel Fusion merges multiple GPU kernels to reduce global memory traffic and launch overhead, with real-world examples from AI inference and deep learning.
Read more →
How NVIDIA BioNeMo Breaks Through Co-F…
·2 min
How NVIDIA BioNeMo Breaks Through Co-Folding End-to-End Performance Bottlenecks
How NVIDIA BioNeMo Agent Toolkit uses agent-based orchestration to solve MSA preprocessing, pipeline scheduling, and end-to-end bottlenecks in OpenFold3 co-folding workloads for drug discovery.
Read more →
NVIDIA Ising Decoding: A 300x Reductio…
·2 min
NVIDIA Ising Decoding: A 300x Reduction in Color Code Logical Error Rates
NVIDIA applies the Ising model to color code quantum error correction decoding, achieving a 300x reduction in logical error rates via GPU parallel computing. A deep dive into the principles and strategic significance.
Read more →
Running Gemma LLM in Godot with GDScri…
·2 min
Running Gemma LLM in Godot with GDScript and Vulkan: An Experimental Deep Dive
A developer runs the Gemma LLM inside Godot 4 using only GDScript and Vulkan compute shaders — no llama.cpp or external dependencies. A technical breakdown of how it works.
Read more →
·4 min
6 Months of Self-Taught Machine Learning: A Complete Roadmap from Zero to Job-Ready
A self-learner completed a full progression from math foundations and core ML to deep learning in 6 months—hand-writing a Transformer and implementing gradient boosting from scratch. This article breaks down the highlights and blind spots of this real roadmap.
Read more →
·7 min
Breaking Down the Core Skills of an AI Application Engineer: From Beginner to Industrial-Grade Deployment
Knowing how to call an API doesn't make you an AI engineer. This article breaks down the complete skill structure of an AI application engineer, covering Python fundamentals, LLM fine-tuning, Agent development, and enterprise projects.
Read more →
·4 min
Unsloth Releases Qwen3.6 NVFP4 Quantization: 2.5x Faster with Zero Accuracy Loss
Unsloth releases NVFP4 quantization for Qwen3.6 using W4A4 true 4-bit Tensor Core computation, delivering up to 2.5x inference speedup over NVIDIA's official implementation with accuracy matching or exceeding BF16 on benchmarks like MMLU-Pro.
Read more →
·2 min
SGLang Integrates DSpark: How Confidence-Driven Speculative Decoding Cracks the High-Concurrency Failure Problem
SGLang officially integrates DSpark, solving the core pain point of speculative decoding failure under high-concurrency batches via confidence-driven variable-length verification. Supports Qwen3 and DeepSeek-V4, hitting 383.7 tok/s on B300.
Read more →
·6 min
DeepSeek Builds Its Own AI Chips: The Strategic Logic and Challenges Behind Compute Independence
DeepSeek is entering AI chip development, targeting compute autonomy. This article analyzes its motivations, software-hardware synergy, chip R&D challenges, and impact on China's AI vertical integration.
Read more →
·7 min
Practical RL Compute Optimization: Core Strategies to Boost CPU and GPU Utilization
An in-depth look at why CPU and GPU utilization is low in RL training, covering vectorized environment parallelism, distributed Actor-Learner architectures, GPU-side simulation (Isaac Gym/Brax), and Ray RLlib practice.
Read more →