Skip to content
8BraidCreators of
8DB

Engineering note · Energy per gigabyte measured, April 2026, rerun scheduled · Daily figures are estimates with published assumption sets

Cut Storage Energy to a Quarter of SQLite's per Gigabyte Written, and Give the Battery Back to Your Users

Measured on an M1 with the machine's own power sensors, a photo-ingest workload through 8DB used about 34 joules per gigabyte written against about 130 through SQLite. The reasons are structural, not tuning: a write-optimised tree writes fewer bytes, indexes built at write time remove the background re-indexing daemons, and relevance-gated sync moves only what another device needs. Here is the measurement, the mechanisms, and what they project to for an iPhone's and a Watch's day, with every assumption in the open.

Published
Reading time6 minutes
Database battery life mobileEnergy efficiency storage engineiPhone battery databaseWrite amplification flash energyBackground indexing powerSync efficiency batteryApple Watch storage energyPost-quantum database

The most valuable thing a storage engine can give a phone is hours. A phone spends a surprising share of its energy on data plumbing that the user never sees: writing more bytes than the data needs, re-reading everything it just wrote so that a search index can catch up, and pushing every change to every other device whether or not anyone there will ever ask for it. 8DB removes each of those by construction rather than by tuning, and on the one workload we have instrumented so far the difference is a factor of about four.

The measurement

In April 2026, on an M1 MacBook Air with a 49.9 watt-hour battery, we ran the same photo-ingest workload through 8DB and through SQLite and read energy from the system management controller's power sensors. 8DB consumed about 34 joules per gigabyte written. SQLite consumed about 130. On that workload the same charge ingests about 3.8 times as many photos. The figure is dated and a rerun on the current build is scheduled, because the engine has changed since April.

Why it is structural

Fewer bytes reach the flash. 8DB's tree is write-optimised: updates are buffered and flushed in large sorted runs, so the bytes actually written approach the bytes of data rather than a multiple of them. Flash writes are among the most expensive operations a phone performs per byte, and write amplification is paid in energy and in the lifetime of the flash itself.

Indexes are built at write time. In a conventional stack the row lands first and a background daemon later re-reads it to extract metadata, compute embeddings, cluster faces and update a search index. That is the same data read again, sometimes several times, at an energy cost that appears on nobody's benchmark because it happens when the phone is idle and plugged in. In 8DB the text index, the perceptual hash, the embedding match and the governance state are produced by the write itself, at a measured 35.7 µs per row batched on the M1, and there is nothing left for a daemon to do.

Sync moves only what matters. Every atom carries a confidence vector and a relevance signal, so device-to-device sync can send the small fraction of changes another device will actually query rather than replicating everything. Radio time is the other large energy line on a phone, and the fraction sent is the lever.

What that projects to

The daily figures below are estimates, and each rests on three inputs: published radio and processor energy constants from the engineering literature, a stated usage model, and a community-sourced figure for what a background indexing daemon draws. We publish conservative, central and aggressive values, and the central ones are these.

DeviceEstimated daily savingShare of batteryLargest contributor
Current-generation large iPhoneAbout 131 mAhAbout 2.8 percent, roughly 18 minutes of screen timeElimination of background search indexing, about 35 mAh
Apple Watch Series 10About 21 mAhAbout 6.5 percent, roughly 1.7 hoursHealth time-series ingest and sync
AirPods ProNegligibleUnder 0.1 mAhAudio streaming dominates; storage is not the lever

The Watch figure is the interesting one. A small battery makes the storage layer a much larger share of the day, and health data is a continuous time series, which is exactly the write pattern the tree is best at. The iPhone figure is dominated by one line, background indexing, and that line rests on a daemon power draw we did not measure ourselves; if the draw is lower than the community figure, the saving shrinks with it, and we say so in the assumption set.

The baseline in every percentage is Apple's published battery capacity or battery life. Reviewer-measured figures, which run longer, are cited for context and never used as a denominator.

The rule we hold energy claims to

Before an energy figure appears anywhere external it is either a measurement with a host, sensor, workload and date, or an estimate with its assumption set and a range. It is a per-day realistic-usage figure or it is labelled illustrative. And it has survived one adversarial read by someone who wants it to be wrong. That discipline is why this article has one measured number and a table of labelled estimates rather than a headline, and it is why the number will still be true when you check it.

What to bring us

  • Device power telemetry. The estimates become measurements the day the workload runs on an instrumented iPhone or Watch. The workload and the assumption set are ready.
  • Your write pattern. Continuous time series, bursty media, chatty sync. Tell us the shape and we will say which of the three mechanisms applies and by how much on the M1 harness.
  • A measured daemon draw. If you have instrumented background indexing on a device, the largest line in the model is waiting for your number.

Write to hello@8braid.com with the subject "Storage energy: measure our workload". You will hear back from an engineer.

Evidence notes

The energy measurement is a single April 2026 run on one M1 MacBook Air using system sensors; a rerun on the current build is scheduled. Every daily figure above is an estimate, not a measurement, and its inputs are listed. Per-sync-cycle figures that appeared in an early internal report are illustrative only and are not used here. A separate thermal-performance property of the retrieval engine, in which accuracy on hard multi-anchor queries rises with an algorithmic bath temperature, was tested on the M1 in May 2026, initially reported as absent, and then confirmed in the calibrated range after the first benchmark was found to have run outside it; that parameter is algorithmic and is not the physical temperature of the processor.

Sources and further reading

The measured energy figure was taken by 8Braid on one M1 MacBook Air in April 2026 using system sensors and describes 8DB and SQLite as run by us; it is not a measurement of Apple's software. Daily milliamp-hour figures are estimates resting on published constants and a stated usage model. Battery capacities and battery-life baselines are Apple-published values. Apple, iPhone, Apple Watch, AirPods, M1, MacBook Air and Spotlight are trademarks of Apple Inc. SQLite is a trademark of Hipp, Wyrick and Company, Inc. 8Braid is not affiliated with or endorsed by Apple Inc.

Continue the technical conversation

Where could this help your work?

Bring a research question, a database workload or an application you want to build. Let’s connect the ideas in this article to an evaluation that matters to your team.

Discuss this work

Continue reading

The next layer of the argument.

Idea

8DB on Apple Silicon

Private On-Device AI Needs a Store That Never Phones Home: Provenance on Every Fact, Clearance Enforced at Every Hop

The promise of intelligence that never leaves the phone is only as good as the data layer under it. A model retrieving from a store that cannot say where a fact came from, who may see it, or how sure to be will give private answers that cannot be trusted. 8DB is the retrieval layer built for that setting: multi-anchor search that returned 3.6 times the precision of full-text search on the questions people actually ask, traversal that enforces clearance on every hop, and a privacy registry that decides at write time what may ever leave the device.

5 min readMulti-anchor retrieval measured on M1, May 2026 · Governed traversal integration-tested · Registry and workloads exercised on synthetic dataRead article
Research

8DB on Apple Silicon

One Store Under Every App: 8DB on Apple Silicon, by the Numbers

Every app on an Apple device keeps its data in the same embedded engine, behind a C interface that has not changed in decades. Put 8DB underneath it and the device gets a store that is faster than SQLite on its drop-in write path, carries provenance, confidence and governance as storage columns, answers graph, search, spatial and face-match queries from one substrate, and runs at the same speed on an M1 as on a desktop chip with six times the cache. Here are the measurements, with the host and date on every one.

6 min readCriterion runs on one M1 host with 95% intervals · Estimates labelled with their assumption setsRead article
Benchmark

8DB on Apple Silicon

Faster Than SQLite on the Writes That Matter, With Provenance and Category 5 Sealing SQLite Cannot Give You

8DB gives an application three ways to write. The drop-in path speaks SQLite's C interface, preserves read-your-own-writes, and inserted about 2.9 times faster than SQLite on the same M1 in the same run. The native path inserted in 99 nanoseconds, about fifteen times faster. The third path costs more per call because it computes confidence, provenance, governance, a perceptual hash and cross-modal indexes on every write, replacing a pipeline of separate passes with one. Here is the benchmark and the trade each path makes.

5 min readsqlite_comparison_bench, Criterion, one M1 host, 20 May 2026 · 95% intervals · SQLite measured in the same runRead article