Ash Vardanian
ashvardanian.com · Personal blog · on HN since 2023
Ash Vardanian, creator of USearch, writes about low-level performance work: SIMD, GPUs, string processing, vector search, and C++ and Rust parallelism.
What HN saysHN links it for performance deep dives on Graviton SVE, trigonometry speedups, and USearch molecule embeddings; one commenter reported a broken table-of-contents link.
25HN people shared it
6linked it in comments
59stories submitted
813story points
8 months agolast shared
2023people per year2026
Most-shared posts
- 1Processing Strings 109x Faster Than Nvidia on H1004 people · 268 points · first shared 2025 · 26 comments on HN
- 2Show HN: I've Published 28B Molecule Embeddings on AWS Open Data4 people · 16 points · first shared 2023 · discussion on HN
- 3Beyond OpenMP in C++ and Rust: Taskflow, Rayon, Fork Union3 people · 144 points · first shared 2025 · 31 comments on HN
- 4Show HN: Python, C, Assembly – 2'500x Faster Cosine Similarity3 people · 14 points · first shared 2023 · discussion on HN
- 5Before AI's Kepler Moment – Are LLMs the Epicycles of Intelligence?3 people · 12 points · first shared 2025 · 1 comments on HN
- 6NumPy vs. BLAS: Losing 90% of Throughput3 people · 8 points · first shared 2024 · 4 comments on HN
- 72x Faster Hashes on AWS Graviton: Neon → SVE23 people · 7 points · first shared 2025 · 2 comments on HN
- 8Mastering C++ with Google Benchmark3 people · 6 points · first shared 2024 · discussion on HN
- 9Binding a C++ Library to 10 Programming Languages3 people · 6 points · first shared 2023 · 1 comments on HN
- 10Abusing Vector Search for Texts, Maps, and Chess ♟2 people · 18 points · first shared 2023 · 1 comments on HN
- 11Show HN: SimSIMD vs. SciPy: How AVX-512 and SVE make SIMD cleaner and ML faster2 people · 12 points · first shared 2023 · 1 comments on HN
- 12Stringzilla – Stupid Heuristics to Search and Sort Strings 5-10x Faster2 people · 10 points · first shared 2023 · discussion on HN
- 13The Longest Nvidia PTX Instruction2 people · 10 points · first shared 2025 · discussion on HN
- 14JavaScript for AI: Array or TypedArray for Vector Search2 people · 4 points · first shared 2023 · discussion on HN
- 15The Painful Pitfalls of C++ STL Strings2 people · 3 points · first shared 2024 · discussion on HN
What HN people wrote when they linked it
jasonjmcghee · 2025-06-27
Nice project! A regular on HN and creator of usearch built an embedding search for the same dataset and did a write up which is a great read.
https://ashvardanian.com/posts/usearch-molecules/
ashvardanian · 2025-03-10
It’s a nice post, but “using array methods” probably shouldn’t be placed in the “Efficient Implementation” section. As often happens with high-level languages, a single plain old loop is faster than three array methods.
Similarly, if you plan to query those vectors in search, you should consider continuous `TypedArray` types and smaller scalars than the double precision `number`.
I know very little about JS, but some of the amazing HackerNews community members have previously helped port SimSIMD to JavaScript (https://github.com/ashvardanian/SimSIMD), and I wrote a blog post covering some of those JS/TS-specifics, NumJS, and MathJS in 2023 (https://ashvardanian.com/posts/javascript-ai-vector-search/).
Hopefully, it should help unlock another 10-100x in performance.
jsheard · 2026-02-18
ARMs Neoverse IP does support SVE, so it's at least already relevant in cloud applications. Apparently AWS Graviton3 has 256bit vectors, but Graviton4 regressed back to 128bit vectors for some reason?
https://ashvardanian.com/posts/aws-graviton-checksums-on-neon-vs-sve/
Labo333 · 2025-02-01
To accelerate the analytical solution, there are more obvious techniques, like replacing trigonometric opetations with series. This is a 10x improvement in https://ashvardanian.com/posts/google-benchmark/#slow-trigonometry-in-…
jonesnc · 2023-12-18
If the author of the article sees this, this link in the table of contents doesn't work:
https://ashvardanian.com/posts/python-c-assembly-comparison/#vnni
cosmicgadget · 2025-04-25
Three fundamental flaws of SIMD ISA:s - Bits'n'Bites (https://news.ycombinator.com/item?id=43783416)
1. Snell-Pym » Processor architecture (https://www.snell-pym.org.uk/archives/2017/03/17/processor-architectur…
2. Fun with vectors in the Raspberry Pi 1 - Part 9 (https://thinkingeek.com/2021/08/22/raspberry-vectors-part-9/)
I think we have enough pieces of machinery working already that we can start with the most exciting part of this journey: autovectorisation!
3. Efficiently compiling efficient query plans for modern hardware | the morning paper (https://blog.acolyer.org/2016/05/23/efficiently-compiling-efficient-qu…
4. Understanding SIMD: Infinite Complexity of Trivial Problems ?? | Ash's Blog (https://ashvardanian.com/posts/understanding-simd-complexity/)
5. SIMD in the GPU world - RasterGrid (https://www.rastergrid.com/blog/gpu-tech/2022/02/simd-in-the-gpu-world…