ashvardanian
Re:Performance of WebAssembly Runtimes in 2026
on Saturday September 12, 2026 @12:52PM (#49674366)
One fairly revealing microbenchmark for WASM runtimes is `int8` dot products & angular/cosine distances.
(My) NumKong [1] has implementations targeting both vanilla AVX2/Haswell and AVX2-VNNI/Alder Lake, which makes it easy to see where runtimes and code generators leave performance on the table.
I started a few Wasmtime/Cranelift PRs around this, but didn’t manage to finish them :facepalm: Might be a fun weekend project for someone interested in backend/codegen work.
Bridging Models' Internal States
Re:Tell HN: GitHub Is Experiencing Degraded Performance
on Monday August 17, 2026 @05:42PM (#49338078)
If only we had a myriad of heavily funded, founder-led AI-for-coding startups that could afford to store a few source files and need all that code for training anyway :)
PS: Saw Cursor’s Origin announcement a second later.
Using AI to build your own software
Show HN: ForkUnion v3 – Faster OpenMP-style thread-pool for C, C++, Rust, & Zig
https://github.com/ashvardanian/ForkUnion
Hi community! Wanted to share a project of mine that is now enabling parallelism across many of my open-source projects: ForkUnion.The absolute majority of "thread pool" projects on GitHub are `std::deque<std::shared_ptr<task_t>>` task queues guarded by a `std::mutex`. I'm not sure which part of that equation I'm more allergic to, but I knew I had to find an alternative. It probably...
AI Model Co-Design: Hardware-Friendly LLM Design
Re:Claude Sonnet 5
on Tuesday June 30, 2026 @07:50PM (#48740737)
Got really excited for this model and asked my Opus planners in 3 pretty different projects to use Sonnets instead of Opus subagents to help me experiment on HPC kernels faster. Not one of them ended up writing a single line of code... Sonnets just kept spinning, wasting tokens. Can't remember the last time it happened with Opus in my codebases. Reverting back.
KinetIQ Ascend: Toward 100% Reliable Manipulation and Superhuman Speed
Nvidia CUDA Python 1.0 and CUDA 13.3 Release
Re:Cloudflare Flagship
on Wednesday May 27, 2026 @10:42AM (#48295153)
I really like the speed at which Cloudflare is executing toward becoming a critical infrastructure player with all of those new product offerings. That said, not everything needs to be serverless. Their Gen 13 hardware looks impressive, and it’s a pity you can’t rent it by the hour like AWS EC2 Metal instances.
Laguna XS.2 and Laguna M.1 by Poolside
FP8 Search and KV-Caching in USearch
Escaping the Fork: How Meta Modernized WebRTC Across 50 Use Cases
Porting Go's io package to C
Schema as the Core of Reliability in AI Memory
Re:NumKong: 2'000 Mixed Precision Kernels for All
on Saturday March 21, 2026 @01:51PM (#47469400)
I'm not aware of that, but it would likely be a great application area for SME!
Re:NumKong: 2'000 Mixed Precision Kernels for All
on Saturday March 21, 2026 @01:56AM (#47464359)
The README was written by a human. I’ve used models extensively to refine the content, but never accepted more than a couple of lines of edits at a time.
NumKong: 2'000 Mixed Precision Kernels for All
The State of Allocators in 2026
Re:What makes Intel Optane stand out (2023)
on Sunday March 15, 2026 @11:46AM (#47388532)
I don't have the inside scoop on Intel's current mess, but they definitely have a habit of killing off their coolest projects.
Re:Meta's Race to Scale AI Chips for Billions: Four Chips in Two Years
on Wednesday March 11, 2026 @11:37AM (#47337005)
Would it be accurate to say that Meta currently produces more RISC-V chips than other vendors? The specs for those chips look extremely interesting and seem much more programmable than Google's TPUs. It would be cool to see Meta making them available to third parties.
Recraft V4
Nebius to buy AI agent search company Tavily for 275M
Re:Nebius to buy AI agent search company Tavily for 275M
on Sunday February 15, 2026 @09:58AM (#47024217)
Re:zvec: embedded vector database
on Saturday February 14, 2026 @07:44PM (#47020007)
8K QPS is probably quite trivial on their setup and a 10M dataset. I rarely use comparably small instances & datasets in my benchmarks, but on 100M-1B datasets on a larger dual-socket server, 100K QPS was easily achievable in 2023: https://www.unum.cloud/blog/2023-11-07-scaling-vector-search... ;)
Typically, the recipe is to keep the hot parts of the data structure in SRAM in CPU caches and a lot of SIMD. At the time of those measurements, USearch used ~100 custom kernels for different data types, similarity metrics, and hardware platforms. The upcoming release of the underlying SimSIMD micro-kernels project will push this number beyond 1000. So we should be able to squeeze a lot more performance later this year.