ps2-llm

AI

language model inference

SHOW HN2026

46points
12comments
#1,281of 4,366 in 2026
Top 30%of 2026 by points

How HN took it

LovedMost comments that take a side are positive

4 positive, 6 neutral or asking, 0 mixed, 0 critical, out of 10 comments about the project. Thread score +0.27.

Performance: 3 for, 0 against

The thread, comment by comment

Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.

positive neutral or mixed critical about another commenter or something else the maker

What HN said

What people liked

  • Streaming weights from CD-ROM is a clever constraint hack

    From 2 comments
    • Love this project. The CD streaming trick is such a smart constraint hack, and honestly the best part is you trained the model for the hardware instead of forcing a desktop…
      SilentEditor on HN
    • The CD-ROM streaming approach is the real insight here, keeping only activations and KV cache in RAM and streaming weights one matrix at a time sidesteps the 32MB constraint…
      SachitRafa on HN
  • Shipping on real hardware is impressive

    From 1 comment
    • Love this project. The CD streaming trick is such a smart constraint hack, and honestly the best part is you trained the model for the hardware instead of forcing a desktop…
      SilentEditor on HN

Objections

  • Inference speed is unclear; tok/sec counter reported zero

    From 1 comment
    • It didn't had any quality loss, since the PSNT as quantization it's mainly to convert the model over the console constraints (you can convert any model you want, even when i…
      xaskasdf on HN

Requests

  • Share per-token latency and per-stage timings

    From 2 comments
    • Love this project. The CD streaming trick is such a smart constraint hack, and honestly the best part is you trained the model for the hardware instead of forcing a desktop…
      SilentEditor on HN
    • Thats super helpful, thanks for the details. Makes sense now that PSNT is more of a transport/runtime format for the PS2 constraints than a quality hack. Very cool that it…
      SilentEditor on HN
  • Faster storage such as HDD or network

    From 2 comments
    • Neat! While the physicality of having the CD spin while running inference is undeniably cool, I wonder if you could run larger models at higher speeds through the PS2 HDD…
      randkyp on HN
    • The network port is faster than the CDVD drive or any of those accessories with the exception of the HDD. The ethernet PHY links at 100Mbit, but the processors inside the PS2 are…
      LocalH on HN

Questions

  • What is the per-token latency on real hardware

    From 2 comments
    • Love this project. The CD streaming trick is such a smart constraint hack, and honestly the best part is you trained the model for the hardware instead of forcing a desktop…
      SilentEditor on HN
    • How many tok/hr?
      mememememememo on HN
  • How much quality is lost from quantization

    From 1 comment
    • Love this project. The CD streaming trick is such a smart constraint hack, and honestly the best part is you trained the model for the hardware instead of forcing a desktop…
      SilentEditor on HN
  • Why a custom format instead of GGUF

    From 1 comment
    • The CD-ROM streaming approach is the real insight here, keeping only activations and KV cache in RAM and streaming weights one matrix at a time sidesteps the 32MB constraint…
      SachitRafa on HN
  • Were the EE vector units used

    From 1 comment
    • IIRC the EE had some interesting hardware with vector units. Were these of any use/benefit here?
      pooparse on HN

Category, keywords, summaries and labels on this page were made by AI (Thread Reader v2) and not checked by hand. Each point links to the comments it came from.

Add a badge to your READMEShown on HackerShow
[![Shown on HackerShow](https://hackershow.org/badge.svg)](https://hackershow.org/show/47470405)