Small costs pile up.
One conversion may feel instant. Repeat it thousands of times as a scene moves or devices exchange updates, and it becomes work the computer has to keep doing.
01 / Try it on this device
Same work.
Less waiting.
Computers waste time turning numbers into text and back again. Our approach keeps the numbers ready to use. Press Run to see what that changes on your device.
Two workers evolve an identical cellular field. One round-trips the state through JSON at every step; the other updates resident typed arrays. The comparison measures processing time in this browser.
01 A living field
The same little world, following the same rules. Black rewrites its data. Blue keeps it ready.
Processing ratio: — · measured on this device
Methodology
Both paths perform identical logical updates. The JSON path encodes and decodes text; the resident path writes and reads a fixed-size binary buffer. Reported medians and p95 come from the samples collected in this session. Checksums verify both paths produced the same logical state.
02 / Turn it. Move closer.
Now give it a whole world.
A lighthouse built from thousands of soft 3D points. Drag either view to look around; both cameras move together. Run the test to move the same points using each approach.
A real 3D Gaussian scene with linked cameras and equal deformation work. Choose 32K, 65K, or the full 131,078 Gaussians. Compare worker processing separately from presented updates.
Faster processing can leave more room for a busy scene. If both views already keep up with your screen, they can look equally smooth. The numbers below each view show what happened here.
03 / What is actually different?
Keep the work.
Lose the busywork.
Imagine rewriting your shopping list in another language before reading every item. A computer can do a version of that when it turns its working data into text and back. Keeping the numbers ready skips that extra step.
More room to do useful things.
Less time preparing data can mean quicker updates or room for a more complex scene. These tests measure that processing cost. They do not measure battery life, electricity, or water use.
The bigger idea is keeping it ready.
Binary data and typed arrays are established tools. Our research asks how far the same idea can go across parts of a system. This page tests one piece of that argument; it does not prove every larger claim.
04 / Bring another screen
How does your household compare?
Invite your phone, laptop, tablet, or a TV with a compatible browser. Each one runs the test itself, then shares its result in a private session.
Being on the same Wi-Fi does not let a web page run code on every appliance. Devices join by opening the link and choosing to share. A detected device is never counted as a completed test.
PUBLIC RUNS
Devices that have run this test
Loading real counts…
- Devices
- —
- Valid passes
- —
- Completed runs
- —
Curious about the bigger argument?
01 / Research loom
Five papers. One connected argument.
Hover a node to preview its claim.
Memory → communication → energy → water
Conceptual dependencies · assumptions in each paper
Read the evidence at your own pace. Browse the papers
Selected
The argument, paper by paper.
00 / The foundation
Capability-addressed binary state
From the manuscript
03 / The derivation
From bytes to a larger question.
Projected Enterprise Scale
bytes/step × steps/s × servers × J/byte × PUE × hours × 3600 ÷ 3,600,000 J/kWh × WUE ÷ 3.785411784 = US gallons
- bytes/step
- —
- bytes/s
- —
- IT watts
- —
- Facility watts
- —
- kWh
- —
- Liters
- —
- US gallons
- —
Run the comparison to provide bytes/step.
WUE on a facility-energy basis is an explicit chosen convention; another convention would remove PUE. This is a scenario model, not measured water saved.