Skip to content

Releases

tensixinet 0.3.1: the first release

Choose a tag to compare

@tsingletaryTT tsingletaryTT released this 31 Oct 17:00
804b3e2

First release. Chip telemetry from Tenstorrent chips (or synthetic ones) played as a band of eleven synthesizers inspired by Ciat-Lonbarde instruments. Tenstorrent hardware hardly strictly required.

What is in it

  • The band: eleven devices dealt across any number of chips, trading solos, patched together with cross-modulation cables. Stereo placement, eviction, styles (chiptune, enya, blues, aphex), replayable patches.
  • The choir sings syllables: consonants (fricatives, plosives, nasals) and vowels that start on triggers from a source the patch names, a rare self-oscillating howl, and a speech-chip crush. Inspired by the ERD/WORM and the Speak & Spell (see the credits).
  • --tui: the patch layout and a colour-coded log, full screen, beside tt-toplike (--live --with-tt-toplike --tui).
  • Fast sampling: --live --interval MS reads hwmon at up to 10 Hz.
  • MIDI out and a VCV Rack voice: a MIDI file, a live port, rtpMIDI.
  • TT-Lang kernels: seven devices run on the chips (--kernels chips), or in the simulator (--kernels sim): the gong, shimmer, drone ripple, AVDog, delay mix, choir and mixer.
  • Gradio app: a live flow chart of what is playing, and a Play tab to play a device, a module or the whole system.

Quick start

git clone https://github.com/tsingletaryTT/tensixinet && cd tensixinet
./tensixinet.py --live

Known limits

  • The choir kernel is about 1.3 times slower than the CPU for a 30 s render. The point is that the chips make the sound.
  • On the box this was built on, tt-metal was rebuilt on 2026-10-30 and no TT-Lang kernel compiles with the shared toolchain, so --kernels chips could not start there. The choir kernel was checked on chips 0 and 1 in tt-orchard's pinned environment. Board 1 and the on-chip clips were not re-run.
  • The site's listening clips were made before the choir sang syllables. A seed now also draws the choir's trigger source.
  • In plain --live the TUI can run ahead of the speaker; use --player none for an exact picture.
  • Suite: 3899 passed, 63 skipped; CI is green on Python 3.10 and 3.12.

See the pull request (#2) for the full list of changes.