Skip to content

backlog(B-0771): audio codecs working (DAW-ready) + Intel NPU/VPU exposed + ONNX as operator contract#5058

Closed
AceHack wants to merge 1 commit into
mainfrom
otto-cli/b0771-audio-codecs-and-intel-npu-daw-ai-workloads-2026-05-25
Closed

backlog(B-0771): audio codecs working (DAW-ready) + Intel NPU/VPU exposed + ONNX as operator contract#5058
AceHack wants to merge 1 commit into
mainfrom
otto-cli/b0771-audio-codecs-and-intel-npu-daw-ai-workloads-2026-05-25

Conversation

@AceHack
Copy link
Copy Markdown
Member

@AceHack AceHack commented May 26, 2026

Aaron 2026-05-25: 'i'd like the sound codecs workikng and npus for use like by daw and others' + 'OpenVINO what about ONYX?' (ONNX).

Beyond the cosmetic firmware fix in iter-3 (#5057). Full stack:

  • Audio: PipeWire + ALSA + RT priority + per-DAW recipes (Ableton, Reaper, Ardour, Bitwig)
  • NPU: intel_vpu kernel module + intel-npu-driver + OpenVINO 2024.0+ + intel-device-plugins-for-kubernetes (k8s NPU device plugin)
  • ONNX as operator contract: ONNX → ONNX Runtime → Execution Provider per hardware (OpenVINO EP for Intel NPU/GPU/CPU; CUDA+TensorRT for NVIDIA; ROCm+MIGraphX for AMD; CoreML for Apple; CPU fallback). Vendor-native runtimes available as opt-in for max perf.
  • Per-role: new workstation host config (composes B-0755); existing roles auto-enable NPU when hardware detected
  • Scheduler awareness: composes with B-0767 NPU-aware scheduler plugins

Composes with B-0754 / B-0755 / B-0759 / B-0761 / B-0763 / B-0764 / B-0767. Per B-0763 vendor swap: alternative runtimes (TVM, IREE, MLIR) + alternative NPU hardware (AMD XDNA, Apple Neural Engine, Hailo) fit the same Zeta.AI.Inference interface.

… as cluster compute resource — beyond cosmetic firmware fix

Aaron 2026-05-25 mid-iter-3-prep, extending the cosmetic firmware
fix scope: 'i'd like the sound codecs workikng and npus for use
like by daw and others' + 'OpenVINO what about ONYX?' (ONNX).

B-0754 iter-3 PR #5057 bundles hardware.enableRedistributableFirmware
to silence Intel SoF audio probe warnings. But that's only the
firmware layer. To actually use audio (DAW workloads) + NPU (AI
inference) on Zeta cluster nodes, the full stack needs config:

- Audio: PipeWire + ALSA + RT priority + per-DAW recipes
- NPU: intel_vpu kernel module + intel-npu-driver userspace +
  OpenVINO 2024.0+ runtime + intel-device-plugins-for-kubernetes
- Per-role: new workstation host config; existing roles get
  NPU auto-enable if hardware detected
- Scheduler awareness: composes with B-0767 Wave-1 NPU-aware
  scheduler plugin

ONNX is the operator-facing contract (per B-0763 + B-0765
ServiceTitan-route + B-0741 ontology negotiation):

- Model format: ONNX (operator deploys one model definition;
  runs everywhere Zeta supports)
- Runtime: ONNX Runtime with Execution Provider per hardware
  (OpenVINO EP for Intel, CUDA/TensorRT EP for NVIDIA, ROCm EP
  for AMD, CoreML EP for Apple, default CPU EP fallback)
- Operator override: vendor-native runtimes (OpenVINO IR /
  TensorRT engine / MIGraphX MXR) available for max-perf at
  cost of portability

Composes with B-0754 / B-0755 (workstation role) / B-0759
persona / B-0761 reference arch / B-0763 interfaces / B-0764
CNCF / B-0767 scheduler.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Copilot AI review requested due to automatic review settings May 26, 2026 01:00
@AceHack AceHack enabled auto-merge (squash) May 26, 2026 01:00
Copy link
Copy Markdown

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Adds a new P2 backlog row (B-0771) describing the intended full-stack delivery for DAW-ready audio on cluster nodes, Intel NPU/VPU exposure as a schedulable resource, and ONNX/ONNX Runtime as the operator-facing inference contract; updates the generated backlog index to include the new row.

Changes:

  • Add B-0771 per-row backlog document under docs/backlog/P2/.
  • Add B-0771 entry to docs/BACKLOG.md under the P2 section.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 5 comments.

File Description
docs/backlog/P2/B-0771-audio-codecs-working-plus-intel-npu-vpu-exposed-for-daw-and-ai-workloads-aaron-2026-05-25.md New backlog row capturing scope/acceptance for audio + Intel NPU/VPU + ONNX layering.
docs/BACKLOG.md Adds the new B-0771 index entry under P2.

- B-0755
- B-0759
- B-0761
- B-0763

### Capability 1: Audio

- [ ] `modules/audio-stack.nix` ships PipeWire as the default
Comment on lines +99 to +103
- [ ] `modules/intel-npu.nix` ships:
- `boot.kernelModules = [ "intel_vpu" ];` (mainline
kernel 6.5+ has the driver)
- `hardware.firmware = with pkgs; [ intel-npu-driver-firmware ];`
(Intel NPU firmware blobs)
(Intel NPU firmware blobs)
- User-namespace permission to `/dev/accel/accel0` via
udev rule + group membership
- [ ] `modules/openvino.nix` ships OpenVINO 2024.0+ runtime
Comment on lines +122 to +126
- [ ] New `nixos/hosts/workstation/configuration.nix` host
config: composes modules/audio-stack.nix + modules/
intel-npu.nix + modules/desktop-environment.nix (GNOME
or KDE Plasma) + modules/k3s-agent.nix (so workstation
is also a cluster member)
@AceHack
Copy link
Copy Markdown
Member Author

AceHack commented May 26, 2026

Closing as substrate-recoverable per pr-triage-tiers.md Tier-3. Re-landed via #5094 (byte-identical B-0771 + regenerated index + pre-emptive MD022 + MD012 fixes). Cascade audit at tick 0410Z.

@AceHack AceHack closed this May 26, 2026
auto-merge was automatically disabled May 26, 2026 04:59

Pull request was closed

AceHack added a commit that referenced this pull request May 26, 2026
…/VPU exposed + ONNX as operator contract (#5094)

* backlog(B-0771): re-land audio codecs working (DAW-ready) + Intel NPU/VPU exposed + ONNX as operator contract

Re-land of stale-DIRTY PR #5058 (Tier-3 per pr-triage-tiers).
Same B-0771 row (269 lines) from PR #5058 head cf9f8e2; BACKLOG.md
regenerated; pre-emptive MD022 fix (heading wrap across lines collapsed
to single line to satisfy blanks-around-headings).

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(B-0771): collapse MD012 double blank introduced by MD022 heading-wrap fix

Co-Authored-By: Claude <noreply@anthropic.com>

---------

Co-authored-by: Lior <lior@zeta.dev>
Co-authored-by: Claude <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants