#CUDA

8 articles tagged with CUDA

FlashAttention-3 on NVIDIA Hopper/Blackwell: FP8 Asynchronous Tensor-Core Warps

A deep CUDA GPU kernel engineering guide to FlashAttention-3. We dissect warp-specialized asynchronous Tensor Core operations, block-sparse softmax quantization, overlapping GMEM-SMEM transfers via TMA, and achieving 850 TFLOPs/sec.

9/2/202624 min read

GPU vs CPU Explained: The Heavy Calculator vs Massive Orchestra (2026)

9/12/202615 min read

Speculative Decoding & Medusa Multi-Head Verification: 3x LLM Inference Speedup (2026)

9/12/202617 min read

Fine-Tuning with Unsloth & QLoRA: Training 70B LLMs on Consumer GPUs (2026)

9/12/202617 min read

vLLM PagedAttention v3 vs TensorRT-LLM: KV Cache Optimization Deep Dive (2026)

9/12/202618 min read

FlashAttention-3 & FP8 GEMM Acceleration: Asynchronous Tensor Cores on Hopper/Blackwell (2026)

9/12/20265 min read

Self-Hosted Embeddings at Scale: Hugging Face TEI (Text Embeddings Inference) in Rust & CUDA (2026)

9/12/20265 min read

EAGLE-2 & Speculative Decoding in vLLM: Tree Attention Verification & 3.5x Speedups (2026)

9/12/20265 min read