frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

GateGPT: 56k tokens per second Transformer (KV cache) on FPGA at 80 MHz

https://twitter.com/fguzmanai/status/2065832668172845209
21•laxmena•1h ago

Comments

amelius•1h ago
See also:

https://rits.shanghai.nyu.edu/ai/karpathys-microgpt-on-fpga-...

TL;DR: The CPU implementation was 71x faster than the FPGA.

Note: model has only 4192 parameters.

cyanydeez•37m ago
yeah, then theres prompt loading too.

but anyone who can fit QWEN-3.6 35B with a sustained ~30 token/s and ~100k context with cache could print money as a hardware vendor.

wmf•21m ago
That just sounds like a 3090.
hedgehog•11m ago
That post is uninteresting both because they miss the point, and it's not clear a human was even involved to perceive a point to miss. Sure, with an unlimited transistor budget, power budget, and a design clocked at 4GHz fabbed on 5nm one of the best CPU design teams in the world can make a thing that is straight line faster than a one-person project running at 80MHz on a 20 year old 65nm FPGA. Any other answer would be extremely surprising.

Now, there are a bunch of interesting things about this project. Seeing the example of a tiny transformer running on FPGA is informative, and that it was apparently a pretty quick project for one person + robot assistance. Probably some transferable lessons for anyone else doing robo-FPGA development.

https://github.com/fguzman82/gateGPT/tree/main/

genxy•38m ago
The context window is 16 characters. Talking about tokens per second is meaningless.
dominotw•4m ago
its not meaningless. there could be usecases like spell correction.
cadamsdotcom•17m ago
Transformers scale poorly vs. context window size and parameter count.

Which means really impressive when those N’s are small!

I’m but a pundit in this area so don’t know much. But one wonders if there’s a future in burning larger models to FPGAs - whether big enough FPGAs exist (or can be built), and whether locating specialized compute right with the memory it needs can speed things up.

Likely would need a lot of algorithm parallelism work that’d translate back to CPUs/GPUs.

Simdjson: Parsing gigabytes of JSON per second

https://github.com/simdjson/simdjson
1•saikatsg•1m ago•0 comments

A satellite just learned to find things on its own – here's what that means

https://techcrunch.com/2026/06/15/a-satellite-just-learned-to-find-things-on-its-own-heres-what-t...
1•speckx•1m ago•0 comments

Microsoft turns to Amazon for help with GitHub's AI-driven capacity issues

https://www.businessinsider.com/microsoft-github-amazon-ai-cloud-capacity-2026-6
1•otterley•2m ago•0 comments

Third SAIR competition: inverse Galois challenge

https://terrytao.wordpress.com/2026/06/16/third-sair-competition-inverse-galois-challenge/
1•jjgreen•2m ago•0 comments

At first, it does sound crazy: meet the scientists trying to refreeze the Arctic

https://www.theguardian.com/environment/2026/jun/16/arctic-sea-ice-rethickening-climate-geoengine...
1•robaato•4m ago•0 comments

Claude: Elevated errors across many models

https://status.claude.com/incidents/xmhsglsz3h3w
18•forks•5m ago•0 comments

Logical Ways to Track AI Agent Lineage and State in Code Development

https://davenporter.substack.com/p/how-to-track-ai-agent-lineage-and
1•davenportjw•5m ago•0 comments

Making things: interview series on creativity

https://digitalseams.com/blog/making-things-interview-series
1•bobbiechen•9m ago•0 comments

How to Use an Nvidia EGPU with Your Mac for Local AI in 2026

https://www.compute-market.com/blog/nvidia-egpu-mac-local-ai-setup-2026
2•falava•9m ago•0 comments

Show HN: VulnFeed – 9 security tools your AI agent can call (MCP server)

https://vulnfeed.novadyne.ai/
1•ngburke•9m ago•1 comments

They made a Pokemon TCG AI Battle Challenge with a $290k prize pool

https://www.shanethegamer.com/esports-news/pokemon-tcg-ai-battle-challenge/
1•misbloss•9m ago•0 comments

The octopus architecture for AI agents

https://blog.goodman.dev/blog/octopus-agent-architecture/
2•joshbetz•10m ago•0 comments

Russian warship 'fires warning shot at a British yacht in English Channel'

https://www.dailymail.com/news/article-15904823/Russian-warship-fires-warning-shot-yacht-English-...
2•Bender•10m ago•0 comments

Show HN: I built an AI that calls you, interviews you and publishes your content

https://heybono.ai/sms
1•zinxor•10m ago•0 comments

Show HN: Skill Atlas – Local, visual IDE for Agentic Skills (BYOK, no back end)

https://github.com/revanthpobala/skill-atlas
1•revanth1108•11m ago•0 comments

German court holds Google liable for fake AI answers

https://www.dw.com/en/german-court-holds-google-liable-for-fake-ai-answers/a-77527661
2•sergdigon•15m ago•0 comments

Using OxCaml to implement type-safe reference counting between OCaml and Python

https://blog.janestreet.com/oxcaml-typesafe-reference-counting-python/
1•pjmlp•15m ago•0 comments

Student Writes 'Not Interested in Working for a Jew' on Handshake

https://www.cornellsun.com/article/2026/06/student-writes-not-interested-in-working-for-a-jew-on-...
1•pinewurst•16m ago•0 comments

A Git forge for the agentic era

https://cursor.com/origin
1•jeremy_k•17m ago•0 comments

Scientists Have Developed a New Technology to Identify Forgeries

https://www.artnews.com/art-news/news/french-scientists-new-technology-identify-forged-artwork-au...
2•ohjeez•18m ago•0 comments

American250 Time Capsule revealed – CA submitted Claude's future prediction

https://america250.org/time-capsule/contents/
1•mcchen51•19m ago•0 comments

ShinyHunters has leaked the data of multiple companies

https://xcancel.com/DarkWebInformer/status/2066906568101081220#m
3•Cider9986•19m ago•0 comments

Sors: a Rust proxy that reorders prompts to maximize vLLM prefix cache hits

https://github.com/flouthoc/sors
2•flaccount•20m ago•0 comments

Gamers beware: malicious wallpapers on Steam found stealing accounts

https://securelist.com/dozens-of-malicious-wallpapers-found-on-steam-workshop/120186/
11•speckx•20m ago•0 comments

San Francisco's Patchwork Streets

https://www.nasa.gov/image-article/san-franciscos-patchwork-streets/
2•MehrdadKhnzd•21m ago•0 comments

Reimagining Xenon II (a 1989 classic game) using Fable and Ghidra [video]

https://www.youtube.com/watch?v=n3EKR58-T1U
1•forelle2•21m ago•1 comments

#44 Travel, enthusiasm and history: an interview with Don and Silke Zagier

https://www.newton.ac.uk/media/podcasts/post/44-travel-enthusiasm-and-history-an-interview-with-d...
1•paulpauper•22m ago•0 comments

Nanogram – Private social media from your Raspberry Pi

5•smalltorch•22m ago•0 comments

Show HN: A policy gate that runs before your AI coding agent's tool calls

https://sigmashake.com
1•cavalrytactics•23m ago•0 comments

Has AI already killed self-help nonfiction books?

https://tim.blog/2026/06/12/has-ai-already-killed-nonfiction/
4•imakwana•24m ago•1 comments