1. X
  2. Ash Vardanian
Log inSign up
Ash Vardanian
1,895 posts
Image
user avatar
Ash Vardanian
@ashvardanian
Building @unum_cloud since 2015 · Investing @aal_vc · author of USearch, StringZilla, NumKong & other open-source infra
London, San Francisco, Yerevan
ashvardanian.com
Joined June 2012
462
Following
3,237
Followers
RepliesRepliesMediaMedia

New to X?

Sign up now to get your own personalized timeline!

Create account

By signing up, you agree to the Terms of Service and Privacy Policy, including Cookie Use.

Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
Don't miss what's happening
People on X are the first to know.
Log inSign up
  • user avatar
    Ash Vardanian
    @ashvardanian
    Apr 18, 2025
    So glad to be on the front page of HN with this 😊 A decade of low-level HPC + AI tricks packed into 3 repos: C++, Rust, Python. SIMD vs ASM, Nvidia TCs (Volta → Hopper), async I/O, and more. If you’re into ML infra, HFT, or databases — dig in! github.com/ashvardanian/l…
    Image
    15K
  • user avatar
    Ash Vardanian
    @ashvardanian
    Sep 13, 2025
    The 26-year-old Apache Lucene is my worst nightmare. It’s arguably one of the world’s top 3 search engines — powering Elastic, Solr, MongoDB Atlas, AWS OpenSearch, and Azure Cognitive Search. Alongside Meta’s FAISS and (my) @unum_cloud’s USearch — the younger successors. So yes,
    Image
    26K
  • user avatar
    Ash Vardanian
    @ashvardanian
    Nov 7, 2025
    SIMD throughput on AMD Zen5 is impressive... even on tiny, irregularly sized inputs! Computing word hashes for dictionary lookups: • Rust standard lib: 333 MiB/s • xxHash3: 998 MiB/s • aHash: 1'159 MiB/s • StringZilla: 2'241 MiB/s 🦖 Finishing StringWars / StringZilla
    Image
    Image
    00:20
    user avatar
    AMD EPYC
    AMD
    @AMDServer
    Nov 6, 2025
    AWS has announced the general availability of its new Amazon EC2 R8a memory-optimized instances, powered by 5th Gen AMD EPYC processors with clock speeds up to 4.5 GHz. This means more speed and efficiency at a lower total cost of ownership, without having to change their cloud
    29K
  • user avatar
    Ash Vardanian
    @ashvardanian
    Nov 11, 2025
    It's remarkable how significant the gap is between SIMD and auto-vectorized performance on @AMDServer Zen5, even for a dense dot-product of two vanilla 1536-dimensional vectors, matching the OpenAI API output. - Dot Product: 16 GB/s in autovec vs 31 GB/s in SimSIMD - Cosine: 9
    Image
    Image
    16K
  • user avatar
    Ash Vardanian
    @ashvardanian
    Nov 11, 2025
    AVX-512 wins on Zen5 for dense & sparse mixed precision math dot_u8_serial - 12 M/s dot_u8_haswell - 34 M/s dot_u8_ice - 42 M/s The primary reason the Haswell-generation AVX2 is slow for 8-bit math is the amount of extraction/unpacking that must be performed for 16-bit
    Image
    GitHub - ashvardanian/NumKong: SIMD-accelerated distances, dot products, matrix ops, geospatial &...
    From github.com
    15K
  • user avatar
    Ash Vardanian
    @ashvardanian
    Jan 3, 2025
    Multiplying 3x3x3 float matrices may take 70% more time on modern CPUs than 4x4x4, despite involving 60% fewer arithmetic operations (45 vs 112) 🤯 This was probably the weirdest artifact I've encountered in 2024 - now part of `less_slow.cpp` 🤗
    Image
    GitHub - ashvardanian/less_slow.cpp: Playing around "Less Slow" coding practices in C++ 20, C,...
    From github.com
    8.3K
  • user avatar
    Ash Vardanian
    @ashvardanian
    Oct 30, 2023
    Cosine Similarity: - 2'500x Faster than 🐍 - 12x faster than C I've combined AVX-512 VNNI extensions with BMI2 and masked loads, all of those seem to be surprisingly rare in open-source
    Image
    Python, C, Assembly - 2'500x Faster Cosine Similarity 📐
    From ashvardanian.com
    31K
  • user avatar
    Ash Vardanian
    @ashvardanian
    Oct 30, 2025
    Switching from Nvidia H100 → B200: +8% Tested on integer-heavy combinatorics/bioinformatics workloads — was hoping for +20%. Right now, a single @intel Granite Rapids socket runs ≈1.5× faster than B200. Even in linear algebra, the gap isn’t what most expect.
    Image
    15K
  • user avatar
    Ash Vardanian
    @ashvardanian
    Feb 25, 2025
    DataStax → @ScyllaDB Redis → @dragonflydbio Confluent → @redpandadata Snowflake → @ClickHouseDB Neo4J → @memgraphdb Thank me later 😉
    10K
  • user avatar
    Ash Vardanian
    @ashvardanian
    Oct 28, 2025
    Scaling Elections with GPUs and Mojo 🔥 I love occasionally optimizing obscure algorithms no one cares about. One of those, also having O(n³) complexity, like matrix multiplication, is the Schulze method for ranked-choice voting. So last year, at one of the AGI hackathons I
    Image
    Scaling Elections with GPUs and Mojo across Nvidia and AMD 🔥
    From ashvardanian.com
    37K
  • user avatar
    Ash Vardanian
    @ashvardanian
    Oct 2, 2024
    Exciting News – I’m Back in the Investing Game! For the past 20 years, most of my time has been spent writing code. But, funnily enough, that’s not how I’ve put food on the table — at least not for the last decade. When I was a teen, I dragged my mother to a bank to open my
    Image
    9.3K
  • user avatar
    Ash Vardanian
    @ashvardanian
    Feb 15, 2024
    We are releasing a new tiny VLM 🎉 The model is smaller and much more accurate than our previous @unum_cloud "uform-gen" downloaded 100K times/mo The new decoder is only 0.5 billion parameters and the model is already available on @huggingface 🤗 huggingface.co/unum-cloud/ufo… 🧵
    Image
    29K
  • user avatar
    Ash Vardanian
    @ashvardanian
    Sep 15, 2025
    StringWa.rs on GPUs: Databases & Bioinformatics 🦠 Levenshtein & NW/SW scores at 500+ GigCUPS MinHash sketching in 52-bit integers at 300+ MB/s More CPU-port-level optimizations & SVE2 A massive post on my largest open-source release of '25 ashvardanian.com/posts/stringwa…
    16K
  • user avatar
    Ash Vardanian
    @ashvardanian
    May 7, 2024
    How did I miss this?! @duckdb is now also using USearch by @unum_cloud for its Vector Search extension 🦆🚀 DuckDB blog post: duckdb.org/2024/05/03/vec… USearch repository & documentation: github.com/unum-cloud/use…
    Image
    5.1K
Advertisement
Advertisement