Files

Michael Chus bf6ecab4f0 Add per-precision benchmark phases, weighted TOPS scoring, and ECC tracking

- Split steady window into 6 equal slots: fp8/fp16/fp32/fp64/fp4 + combined
- Each precision phase runs bee-gpu-burn with --precision filter so PowerCVPct reflects single-kernel stability (not round-robin artifact)
- Add fp4 support in bee-gpu-stress.c for Blackwell (cc>=100) via existing CUDA_R_4F_E2M1 guard
- Weighted TOPS: fp64×2.0, fp32×1.0, fp16×0.5, fp8×0.25, fp4×0.125
- SyntheticScore = sum of weighted TOPS from per-precision phases
- MixedScore = sum from combined phase; MixedEfficiency = Mixed/Synthetic
- ComputeScore = SyntheticScore × (1 + MixedEfficiency × 0.3)
- ECC volatile counters sampled before/after each phase and overall
- DegradationReasons: ecc_uncorrected_errors, ecc_corrected_errors
- Report: per-precision stability table with ECC columns, methodology section
- Ramp-up history table redesign: GPU indices as columns, runs as rows

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

2026-04-13 10:49:49 +03:00

builder

Add per-precision benchmark phases, weighted TOPS scoring, and ECC tracking

2026-04-13 10:49:49 +03:00

overlay

Add per-precision benchmark phases, weighted TOPS scoring, and ECC tracking

2026-04-13 10:49:49 +03:00

vendor

Implement audit enrichments, TUI workflows, and production ISO scaffold

2026-03-06 11:56:26 +03:00

README.md

Add NVIDIA stress loader selection and DCGM 4 support

2026-03-31 11:15:15 +03:00

README.md

ISO Build

bee ISO is built inside a Debian 12 builder container via iso/builder/build-in-container.sh.

Requirements

Docker Desktop or another Docker-compatible container runtime
Privileged containers enabled
Enough free disk space for builder cache, Debian live-build artifacts, NVIDIA driver cache, and CUDA userspace packages

Build On macOS

From the repository root:

sh iso/builder/build-in-container.sh

The script defaults to linux/amd64 builder containers, so it works on:

Intel Mac
Apple Silicon (M1 / M2 / M3 / M4) via Docker Desktop's Linux VM

You do not need to pass --platform manually for normal ISO builds.

Useful Options

Build with explicit SSH keys baked into the ISO:

sh iso/builder/build-in-container.sh --authorized-keys ~/.ssh/id_ed25519.pub

Rebuild the builder image:

sh iso/builder/build-in-container.sh --rebuild-image

Use a custom cache directory:

sh iso/builder/build-in-container.sh --cache-dir /path/to/cache

Notes

The builder image is automatically rebuilt if the local tag exists for the wrong architecture.
The live ISO boots with Debian live-boot toram, so the read-only medium is copied into RAM during boot and the runtime no longer depends on the original USB/BMC virtual media staying present.
Target systems need enough RAM for the full compressed live medium plus normal runtime overhead, or boot may fail before reaching the TUI.
The NVIDIA variant installs DCGM 4 packages matched to the CUDA user-mode driver major version. For driver branch 590 / CUDA 13.x, the package family is datacenter-gpu-manager-4-cuda13 rather than legacy datacenter-gpu-manager.
Override the container platform only if you know why:

BEE_BUILDER_PLATFORM=linux/amd64 sh iso/builder/build-in-container.sh

The shipped ISO is still amd64.
Output ISO artifacts are written under dist/.