Microsoft's analog optical computer cracks 2 practical problems,shows AI promise (news.microsoft.com)

1 points by pseudolus 4m ago 0 comments

WaveBoy (waveboy.bitbybitsynths.com)

1 points by derbOac 5m ago 0 comments

The MCP Registry (blog.modelcontextprotocol.io)

2 points by aratahikaru5 10m ago 0 comments

EUV: Lithography: History, Latest Results, Technology Roadmap [video] (youtube.com)

2 points by matt_d 20m ago 0 comments

What Happens If No One Reads (thefp.com)

3 points by lermontov 20m ago 0 comments

Ask HN: How do developers learn design intuition?

2 points by mercat 21m ago 0 comments

Josh Wolfe on AI and the Breaking of Silicon Valley's Social Contract [video] (youtube.com)

1 points by mooreds 22m ago 0 comments

Ted Lasso is 1M Micro-Lessons for what Tech Leadership needs (substack.com)

2 points by mooreds 23m ago 0 comments

Anthropic reduced model output quality from Aug 5 (status.anthropic.com)

1 points by bashtoni 24m ago 0 comments

Lumi Language Model (websim.com)

1 points by eunoiaAI 25m ago 1 comments

First 'perovskite camera' can see inside the human body (news.northwestern.edu)

3 points by geox 27m ago 0 comments

Show HN: Tablemd – canvas-based Markdown table editor (tablemd.app)

1 points by albert-yu 27m ago 0 comments

Kwil: The Database for Web3 (github.com)

1 points by InitEnabler 28m ago 0 comments

Disposable masks during Covid have left chemical timebomb, research suggests (theguardian.com)

2 points by jnord 30m ago 1 comments

Space-Based Infrared System (SBIRS) (en.wikipedia.org)

1 points by areoform 30m ago 0 comments

Tech promised everything. Did it deliver? (youtube.com)

1 points by QuadrupleA 35m ago 0 comments

OpenAI's AI-Made Animated Feature Film Aims to Debut at Cannes Film Festival (wsj.com)

1 points by bookofjoe 36m ago 3 comments

Show HN: Attempt – A CLI for retrying fallible commands (github.com)

7 points by maxbond 39m ago 1 comments

Show HN: Devibe – pip lib that will remove AI comments and devibe your code (github.com)

1 points by 10000000001 39m ago 0 comments

Tesla is (still) following in Waymo's footsteps (understandingai.org)

3 points by NullHypothesist 43m ago 0 comments

Show HN: Verse – FOSS Markdown editor web app (github.com)

1 points by p4cs 43m ago 1 comments

Ask HN: Grafana or Datadog?

3 points by kvaranasi_ 50m ago 6 comments

OpenAI Executives Rattled by Campaigns to Derail For-Profit Restructuring (wsj.com)

5 points by spenvo 50m ago 0 comments

America's First Private Nuclear Fuel Recycling Facility to Open in Tennessee (gizmodo.com)

2 points by toomuchtodo 51m ago 0 comments

Airbnb Images Downloader (chromewebstore.google.com)

1 points by qwikhost 52m ago 1 comments

Show HN: Vizza – Interactive, Beautiful Simulations (github.com)

3 points by zeldahessler 55m ago 1 comments

The Evolution of Laziness [video] (youtube.com)

1 points by gmays 58m ago 0 comments

Unbound 1.23.1 and the RFC 8767 performance regression (nox.sh)

1 points by mattredact 1h ago 0 comments

No Adblocker Detected (maurycyz.com)

2 points by LorenDB 1h ago 0 comments

Workday Canvas Design System (canvas.workday.com)

1 points by mooreds 1h ago 0 comments

RU 'Doomsday Radio' speaks again;UVB-76 broadcasts names, numbers, & phrases (economictimes.indiatimes.com)

9 points by ggm 1h ago 1 comments

Former WhatsApp security boss in lawsuit likens Meta's culture to a "cult" (arstechnica.com)

4 points by Bender 1h ago 0 comments

Anthropic addresses Claude Code quality issues (twitter.com)

3 points by just_human 1h ago 2 comments

Master Foo and LLM Mountain (tusshah.github.io)

1 points by mmts 1h ago 0 comments

Load Llama-3.2 WebGPU in the browser from a local folder (simonwillison.net)

1 points by indigodaddy 1h ago 0 comments

Schlep Blindness (2012) (paulgraham.com)

1 points by ksec 1h ago 0 comments

Show HN: OSS app to find LLMs across multiple LLM providers (Azure, AWS, etc.) (github.com)

2 points by njbrake 1h ago 0 comments

Show HN: I built a way to monetize any link with a crypto paywall

2 points by allynjalford 1h ago 0 comments

Geoffrey Hinton: 'AI will make a few people much richer and most people poorer' (ft.com)

15 points by pseudolus 1h ago 8 comments

Orchestrate multiple AI agents with cagent by Docker to create coding assistant (tobiasfenster.io)

3 points by AsmodiusVI 1h ago 2 comments

Windows-Use: an AI agent that interacts with Windows at GUI layer (github.com)

2 points by djhu9 1h ago 0 comments

Anthropic is endorsing SB 53 (anthropic.com)

1 points by arroia 1h ago 0 comments

But how do AI images and videos work? (youtube.com)

1 points by tzury 1h ago 0 comments

Biz Academy aiuta imprenditrici con percorsi online (biz-academy.it)

1 points by lorisfreez 1h ago 0 comments

Internet censorship is complex, usually (en.wikipedia.org)

1 points by DaveZale 1h ago 1 comments

How Python Implements List Comprehensions (pythonkoans.substack.com)

3 points by meander_water 1h ago 0 comments

Jakub and Szymon (blog.samaltman.com)

3 points by davidbarker 2h ago 0 comments

How Big Was IBM? (thechipletter.substack.com)

3 points by chmaynard 2h ago 2 comments

It's AI all the way down as Google's AI cites web pages written by AI (theregister.com)

3 points by akyuu 2h ago 0 comments

Agentic AI Runs on Tools (simplicityissota.substack.com)

2 points by bookish 2h ago 0 comments

Ask HN: Why no inference directly from flash/SSD?

1 myrmidon 2 9/8/2025, 8:15:39 AM

My understanding is that current LLMs require a lot of space for pre-computed weights (that are constant at inference-time).

Why is it currently not feasible to just keep those in flash memory (fast PCIe SSD Raid or somesuch), and only use RAM for intermediate values/results?

Even modest success on this front seems very attractive to me, because Flash storage appears much cheaper and easier to scale than GPU memory right now.

Are there any efforts in this direction? Is this a flawed approach for some reason, or am I fundamentally misunderstanding things?

Comments (2)

sunscream89 · 16h ago

> A typical DRAM has a transfer rate of approximately 2-20GB/s, whereas typical SSDs have a transfer rate of 50MB-200MB/s. So it's one to two orders of magnitude slower.

myrmidon · 11h ago

I don't think the bandwidth gap is that big-- single WD SN8100 drives (before any potential gain from RAID) already have sequential read speed of >10GB/s for under $200 and 1TB of storage.

A GPU setup with a terabyte of video memory costs a fortune by comparison-- there has to be some kind of reason that people are not trying really hard to make this work, no?

No comments yet