NODELIFT / PUBLIC REFERENCE
Measurements with their boundaries.
Read actual reports, test conditions and unsuccessful outcomes. A file-transfer result, a CPU qualification and a language-model inference result answer different questions.
No universal speedup, pooled-memory result or whole-system energy saving follows from these reports. Each result retains its hardware, model, connection and observation window.
01
Published experiment reports
AModel-file distribution, 16 November 2026Verified transfers, cache reuse, transfer time and scoped energy evidence. File distribution does not measure LLM acceleration.BHive LLM execution, 15 November 2026Complete-model dispatch and worker/Hive timings; model partitioning is a separate qualification.CEarlier engineering evidenceThe technical archive preserves earlier charts, comparisons, negative outcomes and qualification notes.
02