Top all-time · #7610
salykova
Personal blog · salykova.github.ioprogrammingaisystems
salykova is a personal blog at salykova.github.io about programming, ai and systems. Since 2024, 10 different Hacker News readers have linked it 29 times, most often “Beating cuBLAS in Single-Precision General Matrix Multiplication”. It is still writing: the last post was on 2025-09-30, and you can follow it by feed.
Too few sampled comments (3) to say how often HN recommends it.?
Why HN links it
HN readers cite salykova's posts on optimizing matrix multiplication (GEMM, GEMV) on CPUs and NVIDIA and AMD GPUs, as evidence in discussions of CUDA and BLAS performance.
Summarised by Claude from a sample of 3 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
making AI inference run really fast.
Most-linked posts?
- Beating cuBLAS in Single-Precision General Matrix Multiplication5 people · 8 mentions · first linked 2025-01-14
- Beating NumPy matrix multiplication in 150 lines of C5 people · 7 mentions · first linked 2024-07-01
- Beating OpenBLAS in FP32 Matrix Multiplication2 people · 8 mentions · first linked 2025-01-14
What Hacker News says?
I plan to write my own GEMM implementations so I can do kernel fusion. I am using GEMM implementations from Nvidia and Intel until I have my own performant substitutes. So far, my initial attempts at writing my own GEMM implementations have not been performance competitive. Intel and Nvidia use tiling algorithms. I know (unintentionally) from profiling that Intel is doing a memcpy trick that is presumably similar to…
Comment linking /matmul-cpu
on “Fast LLM Inference From Scratch (using CUDA)”Submitted Beating NumPy matrix multiplication in 150 lines of C as a story · 392 points
HN discussionThere are likely plenty of unrealized opportunities to improve mature BLAS libraries. For example, this guy who was able to outperform OpenBLAS' GEMM on Zen 4: https://salykova.github.io/matmul-cpu Concidentally, the Intel MKL also outperforms OpenBLAS, so there being room for improvement is well known. That said, I have a GEMV implementation that outperforms both the Intel MKL and OpenBLAS in my tests on Zen 3: htt…
Comment linking /matmul-cpu
on “CUDA Moat Still Alive”Submitted Matrix Core Programming on AMD GPUs as a story · 116 points
HN discussionPeople have been trying to bypass CUDA and even PTX for a long time. One long rundown of optimizing gemm on NVIDIA hardware ( https://salykova.github.io/sgemm-gpu ) mentions 'maxas' ( https://github.com/NervanaSystems/maxas/wiki/Introduction ) - which was really a step forward in this space. I still blame Intel (buying NervanaSystems) for killing it...
Comment linking /sgemm-gpu
on “Optimizing Matrix Multiplication on RDNA3”Submitted Beating cuBLAS in Single-Precision General Matrix Multiplication as a story · 98 points
HN discussionSubmitted Advanced Matrix Multiplication Optimization on Multi-Core Processors (2024) as a story · 85 points
HN discussionSubmitted Matrix Core Programming on AMD CDNA3 and CDNA4 Architecture as a story · 24 points
HN discussionSubmitted Beating NumPy's matrix multiplication in 150 lines of C code as a story · 11 points
HN discussionSubmitted Beating OpenBLAS in FP32 Matrix Multiplication as a story · 7 points
HN discussionSubmitted Beating cuBLAS in Single-Precision General Matrix Multiplication as a story · 7 points
HN discussionSubmitted Beating NumPy's matrix multiplication in 150 lines of C code as a story · 5 points
HN discussionSubmitted Beating OpenBLAS in FP32 Matrix Multiplication as a story · 4 points
HN discussionSubmitted Beating OpenBLAS in FP32 Matrix Multiplication as a story · 4 points
HN discussionSubmitted Beating cuBLAS in Single-Precision General Matrix Multiplication as a story · 4 points
HN discussionSubmitted Beating NumPy's matrix multiplication in 150 lines of C code as a story · 4 points
HN discussionSubmitted Beating cuBLAS in Single-Precision General Matrix Multiplication as a story · 3 points
HN discussionSubmitted Beating cuBLAS in Single-Precision General Matrix Multiplication as a story · 3 points
HN discussionSubmitted Advanced Matrix Multiplication Optimization on Nvidia GPUs as a story · 2 points
HN discussionSubmitted Matrix Core Programming on AMD GPUs as a story · 2 points
HN discussionSubmitted Introduction to Matrix Core Programming on AMD CDNA3 and CDNA4 Architecture as a story · 2 points
HN discussionSubmitted Beating OpenBLAS in FP32 Matrix Multiplication as a story · 2 points
HN discussionSubmitted Beating Nvidia's cuBLAS in GEMM as a story · 2 points
HN discussionSubmitted Beating OpenBLAS in FP32 Matrix Multiplication as a story · 2 points
HN discussionSubmitted Show HN: Beating cuBLAS in Single-Precision General Matrix Multiplication as a story · 2 points
HN discussionSubmitted Beating OpenBLAS in FP32 Matrix Multiplication as a story · 2 points
HN discussionSubmitted Beating NumPy's matrix multiplication in 150 lines of C code as a story · 2 points
HN discussionSubmitted Beating OpenBLAS in FP32 Matrix Multiplication as a story · 1 points
HN discussionSubmitted Beating OpenBLAS in Matrix Multiplication as a story · 1 points
HN discussion
