GlobalGrid2050

An open grid development, engineering, procurement, construction and operations platform dedicated to documenting, analysing and improving the world's electrical energy systems as they undergo rapid electrification.

VENTUS WIRING — CURRENT WORKING RELEASE

20260922-A · Simulation environment · NOT A REAL DESIGN

→ OPEN DC WIRING — 24 strings / 48 cables → OPEN AC LV — 28 inverters / 2 transformer sections Older version: solar array drawing

Disclaimer: Content provided for general technical documentation and research purposes only.

Full archive

GlobalGrid2050 explores current technology with Claude and Codex AI, and actively studies CuPy for faster independent compute on local machines. The engineering arithmetic on this site is enumerated on our own hardware rather than estimated, and no result is published until it has been verified against a reference implementation.

Taken on one machine, 20 September 2026 — a complete solar string-and-block design space of 1,020,600,000 designs: 177,289,777 cases/s on the GPU with CuPy against 12,767,553 cases/s on twelve CPU workers, a 13.89× difference, the billion run in 5.6 s. A second kernel reached 306,440,267 cases/s, 40.82× its CPU reference. Verification: 100,000 cases on a coprime stride, 0 differ between the scalar reference, the CPU vectorisation and the GPU, worst relative difference 0.00e+00 against a 1e-9 tolerance.

Conditions, because a rate without them is not evidence: one machine, one 16 GB card that was shared with other work during the runs, single and double precision both checked and agreeing on every verdict. CPU watts were never recorded — no wall meter, no RAPL read — so no efficiency ratio between processor and card is stated here or anywhere else in this work. A claim of “tens of millions a second” for a third kernel was tested and not reached, falling 3.0× short; it is recorded as not reached. Method, figures and the solar block models: Independent compute · code at gpu-drivers-for-global-grid.