frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Show HN: Avoiding the Memory Wall by computing LLM inference directly inside RAM

1•pcdeni•37m ago
The excitement surrounding PrismML’s 1-bit/ternary Bonsai models has the industry closely watching how smartphone giants, particularly Apple, will implement LLMs on edge devices. Moving AI on-device is a brilliant and necessary strategy. It ensures absolute user privacy in alignment with EU regulations, fundamentally shifts the economics away from costly cloud inference, and paves the way for a significant hardware upgrade supercycle as users seek true AI-capable silicon.

To create a smart on-device "Semantic Router," models need to reach the 27B+ parameter scale. Achieving this on a phone requires extreme quantization, such as PrismML’s ternary weights.

However, a critical hardware reality often overlooked by the software world is that fitting the weights in RAM is not equivalent to moving them. Running a 27B ternary model on standard LPDDR encounters a significant memory bandwidth limitation. Transferring gigabytes of data across the SoC bus for each token generation can lead to thermal throttling of the NPU and excessive battery drain.

This raises an important question: why are we still transferring data to the compute? Why not execute AI inference natively within the memory?

Frustrated with academic PIM simulations that overlook bare-metal physics, I developed CaSA, an architecture that performs ternary LLM inference directly inside COTS DRAM through charge-sharing, completely bypassing the memory bus.

Software quantization is a great initial step, and CaSA provides the physical hardware substrate needed to complete the bridge: https://github.com/pcdeni/CaSA

Camelid – Local AI inference in Rust with desktop, web, terminal, API interfaces

https://github.com/timtoole02/Camelid
1•Tooleman•27s ago•0 comments

The Men Who Believed Some Men Are More Equal Than Others

https://www.nytimes.com/2026/07/22/books/review/country-of-lords-kim-phillips-fein.html
1•mitchbob•1m ago•1 comments

I Reported on Flock's Cameras. Now I'm One of the System's Mistakes

https://www.rascal.news/drivethrurpg-delisted-a-tabletop-game-about-revolutions-over-hateful-poli...
1•healsdata•1m ago•0 comments

Ask HN: $100 to Facebook employee for multiple sclerosis help

1•jph•1m ago•0 comments

Encryption and Globalization 15 Years Later: E2EE and the "Going Dark" Debate

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6959699
1•iamnothere•3m ago•0 comments

Still Looking

https://stilllooking.co/
1•anonguySL•3m ago•0 comments

Milgram Experiment

https://en.wikipedia.org/wiki/Milgram_experiment
1•thunderbong•3m ago•0 comments

Learn WebGPU for C++

https://eliemichel.github.io/LearnWebGPU/
1•ibobev•3m ago•0 comments

The challenges of building a reliable PR review agent

https://charlielabs.ai/blog/how-to-build-a-pr-review-agent/
1•mrbbk•4m ago•0 comments

Vulkan Tutorial

https://vulkan-tutorial.com/
2•ibobev•5m ago•0 comments

Job Lock and Victor Fuchs – American Healthcare Chronicles

https://alearningaday.blog/2026/07/22/job-lock-and-victor-fuchs-american-healthcare-chronicles/
1•herbertl•5m ago•0 comments

Learn OpenGL, extensive tutorial resource for learning Modern OpenGL

https://learnopengl.com/
2•ibobev•5m ago•0 comments

One (strange) particle could soon rewrite our laws of gravity

https://www.sciencefocus.com/space/how-one-particle-could-soon-rewrite-our-laws-of-gravity
1•bryanrasmussen•5m ago•0 comments

California's Privacy Obstacle Course

https://pluralistic.net/2026/07/23/drop-a-dime/
1•hn_acker•5m ago•0 comments

Show HN: Plox – Lox compiler in ~600 LOC from the 'Crafting Interpreters' book

https://github.com/eliasdejong/plox
1•eliasdejong•6m ago•0 comments

I regret to report that the AIs are weebs, too

https://www.pcgamer.com/software/ai/i-regret-to-report-that-the-ais-are-weebs-too/
1•Brajeshwar•6m ago•0 comments

Show HN: GapQuery – find app-market gaps from 35,600 apps and their reviews

https://www.gapquery.com
1•northify•6m ago•0 comments

Snapshots, copy-on-write, and the economics of agent sandboxes

https://builders.cortex.io/blog/sandboxing-agents-part-2/
4•nikhilunni•7m ago•0 comments

Show HN: BatchEdits – Batch AI video editor to edit content in volume

https://batchedits.com
1•shayannadeem321•7m ago•1 comments

PasteFlow

https://chromewebstore.google.com/detail/pasteflow/paenffoomjmkonbgkmfdbnfaljoiilgm
1•rbediner•8m ago•0 comments

Sunglasses: An open-source input scanner for AI agents

https://github.com/sunglasses-dev/sunglasses
1•azrollin•10m ago•0 comments

Robots.txt – Just Crawl It

https://www.nike.com/robots.txt
1•bfoks•10m ago•0 comments

Show HN: Carta – An open-source pandoc reimplementation in Rust

https://github.com/mfkrause/carta
1•mfkrause•11m ago•0 comments

Arctic: a lock-free concurrent ordered map for Rust

https://old.reddit.com/r/rust/comments/1v4f05y/announcing_arctic_a_lockfree_concurrent_ordered/
1•arto•12m ago•0 comments

OHMG: One Hot Modular Garbling

https://eprint.iacr.org/2025/2338
1•wslh•13m ago•0 comments

Show HN: 5dive – Run a Company of Claude Code/Codex Agents (Written in Bash)

https://github.com/5dive-ai/5dive
1•lodar•13m ago•0 comments

Thoughts on Integers

https://blog.xoria.org/integers/
2•vinhnx•13m ago•0 comments

Hosted PaddleOCR-VL-1.6 API

4•TimurKramar•14m ago•0 comments

Even camels can't cope: Africa's ships of the desert hit by rising temperatures

https://www.bbc.com/news/articles/cqlxvvw06pqo
1•throw0101d•15m ago•0 comments

My Agentic Coding Setup, July 2026

https://domenic.me/agentic-coding-setup/
2•domenicd•15m ago•0 comments