PERSONAL INFRASTRUCTURE · PUBLIC LEDGER

Small experiments.
Real compute.

A running account of the machines behind the work.
Time, cost, and a little more transparency.

Checking snapshot…Loading data

GPU HOURS EST.

—Known allocation time · partial

KNOWN SPEND USD

—RunPod billing · AWS excluded

PODS OBSERVED

—Distinct RunPod resources in ledger

CARBON EST.

—kgCO₂e · matched GPU-hours only

01 / THE BILL

Spend over time

USD · monthly

Missing months are gaps, not zero spend. Current month may be incomplete.

02 / THE HARDWARE

GPU-hours by GPU

03 / THE WORK

GPU-hours by project

04 / THE FOOTPRINT

Where compute happened

05 / THE WEEK AHEAD

Kaggle GPU quota

Quota-hours are tracked separately from physical GPU-hours. Weekly quota is not lifetime usage.

06 / THE LONG HAUL

Longest observed runs

Wall-clock hours · estimates

SOURCE HEALTH

THE FINE PRINT, UP FRONT

Estimates with receipts.

We estimate allocated GPU time from lifecycle events or an explicitly recorded historical rate. This is not a measure of productive work or GPU utilization.

kWh = GPU-hours × watts ÷ 1,000 × PUE
kgCO₂e = kWh × regional kgCO₂e/kWh

Nameplate power is a proxy for draw. Unknown hardware and unverified grid mappings stay unknown. The default PUE of 1.0 is a GPU-only baseline, excluding facility overhead. Carbon is a partial operational estimate, not a complete footprint.

Constants & sources

Every value below carries its source and review date. [sourced] identifies evidence; [inference] identifies an assumption or estimate.

Small experiments.Real compute.