Overall score
Project report
μBench microcontroller benchmark.
A portable benchmark and reporting system for comparing microcontroller boards using repeatable tests, normalized scores, charts, and power data.
Current featured report
μBench microcontroller benchmark.
22
boards captured
29
deterministic tests
50
runs per board
10
test suites
Flagship project
Repeatable microcontroller benchmarks.
μBench compares microcontroller boards using one firmware codebase, calibrated work, structured serial capture, generated reports, and power measurements. The public score is normalized against an Arduino Uno R3 baseline where Uno equals 100.
AVR
Arm M0+/M4F/M7
Xtensa LX6/LX7
RISC-V
FNB58 power capture
Results snapshot
Top findings from the current data set.
| Finding | Board | Score | Energy | Usable takeaway |
|---|---|---|---|---|
| Fastest overall | Waveshare ESP32-P4-WiFi6 | 18,868 | 0.085 J | Best raw benchmark score and a leading score-per-dollar result. |
| Best score per joule | Arduino Nano ESP32 | 11,885 | 0.034 J | Completes the same work with the strongest measured energy efficiency. |
| Strong value board | ESP32 DevKit (WROOM) | 10,774 | 0.089 J | Low-cost board with a score-per-dollar nearly tied with the ESP32-P4. |
| Best per-MHz result | ST Nucleo-F746ZG | 15,810 | 0.413 J | Shows high architecture efficiency from a Cortex-M7 class board. |
| Board-level power spread | ATmega328P family | 98-100 | 0.964-3.037 J | Similar silicon can draw very different board-level power. |
Data status: captured from the internal microBench result set available on July 3, 2026. Prices are approximate single-unit USD and should be rechecked before procurement. FNB58 power numbers are useful for large comparisons, but not source-meter-grade.
Readable charts
The public view favors legible decision charts.
Energy efficiency
Best score per joule
Value
Best score per dollar
Full annotated scatter plots are useful during analysis, but they become hard to read on a public page with long board names. These summary charts keep the labels separate from the marks and preserve the key buying signals.
Reporting method
The report is built from repeatable evidence.
Identical work
Every board runs the same calibrated iteration counts, so faster boards finish the same work sooner.
Correctness checks
Checksums make optimization mistakes, miscompiles, and non-portable assumptions visible.
Raw capture first
Firmware reports raw timing and checksums; host tooling computes normalized scores afterward.
Decision views
The same capture supports overall score, per-MHz efficiency, score-per-dollar, energy, and noise views.
Reusable pattern
From bench data to readable engineering report.
Capture
Firmware, serial tooling, and power logging produce structured evidence for each board.
Normalize
Host tools join board metadata, compute scores, preserve raw runs, and regenerate tables.
Publish
Charts, caveats, and procurement-oriented summaries make the results useful outside the bench.
Other project patterns
Infrastructure, automation, and technical memory.
Infrastructure operations
Static web deployment, Caddy-hosted sites, DNS cutovers, backups, monitoring, and recovery runbooks.
Engineering knowledge systems
Markdown workspaces for decisions, requirements, test records, project notes, and reusable procedures.
Test and report automation
Scripts and pipelines that convert repeated technical work into reviewable artifacts and clear outputs.
Covector Core workflows
Multi-agent technical delivery patterns with human review gates, artifacts, and implementation handoff.
Contact
Need a benchmark, technical report, or evidence-backed workflow?
contact@covectorlabs.com
Use this address for project briefs, benchmark questions, and report automation work.