EXP-084REJECTED
Canonical GPU engineering / SHA-256d

P16 half-batch synchronization every two launches

Halving launch size and synchronizing every two launches preserves ~20 ms observability with less overhead than synchronizing every launch.

Claim statusNO END-TO-END MINING ADVANTAGE DEMONSTRATED

Diagnostic work, exact algebra and local capabilities are not treated as end-to-end mining advantage.

Date2026-08-16
HardwareNVIDIA GeForce RTX 5080 · exact CUDA SHA-256d verifier
Run scopePreregistered paired campaign with sealed validation and audit
ReproducibilityRepeat unchanged with 67,108,864×64 audit work and require all binaries to recover nonce 3,842,969,478 before reading timing.
01 / Question & hypothesis

What was tested?

Halving launch size and synchronizing every two launches preserves ~20 ms observability with less overhead than synchronizing every launch.

02 / Scientific basis

Why the test is meaningful

Performance evidence is interpretable only if the exactness audit covers the complete registered nonce domain and includes its known target hit.

H_audit=67,108,864×32=2^31known B32 nonce=3,842,969,478>2^31−1valid inference requires H_audit=2^32
P16 half-batch synchronization every two launchesVisual reading of the published metrics and gates for EXP-084; it summarizes the registered result, not a mining advantage.EXP-084 / P16 HALF-BATCH SYNCHRONIZATION EVERY TWO LAUNCHESAUDIT COVERAGE2^31 / 2^32KNOWN HITOUT OF RANGEVERDICTINVALID
FIGURE / RESULT READINGVisual reading of the published metrics and gates for EXP-084; it summarizes the registered result, not a mining advantage.
03 / Method

How it was tested

Compare P16 no-sync, sync-every-launch and sync-every-two across 24 benchmarks, batch 67,108,864, 192 launches per measurement and one copy per 16.

04 / Observed result

What happened

Audit coverage2^31 / 2^32
Known hitOUT OF RANGE
VerdictINVALID

The candidate appeared 1.000108× versus sync-one (CI95 0.999717–1.000499), retained 0.999761 of no-sync and observed at 0.020276 s. But no binary found B32 because the audit retained only 32 repetitions after halving the batch.

05 / Validation

Exactness and statistical controls

Benchmarks, factors, spills and observed comparisons passed, but the registered known hit lay outside the 2^31-word audit range. The campaign is technically invalid, not a mechanism rejection.

06 / Interpretation

What the result means

No performance conclusion is allowed. The exact design and gates must be repeated with 64 audit repetitions covering 2^32 words.

Limitations

  • The exactness audit covered only half the nonce domain.
  • Apparently favorable timing is scientifically unusable.
  • No promotion or negative mechanism claim follows.
07 / Reproduction

Evidence trail

Repeat unchanged with 67,108,864×64 audit work and require all binaries to recover nonce 3,842,969,478 before reading timing.

Canonical variants

CANONICAL-EXP-160PREREGISTERED-CAMPAIGNSEALED-AUDIT

Source: internally audited canonical reports. Local filesystem structure, private headers and operational identifiers are excluded from publication.