EXP-059VALIDATED
Canonical GPU engineering / SHA-256d

Static Blackwell SASS audit

The sm_120 compiler already maps the external SHA-256d arithmetic to Blackwell-native shift, Boolean and add instructions.

Claim statusNO END-TO-END MINING ADVANTAGE DEMONSTRATED

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

Date2026-08-14
HardwareNVIDIA GeForce RTX 5080 · exact SHA-256d verifier
Run scopePreregistered campaign with sealed result and audit
ReproducibilityDisassemble the sealed binaries with matching CUDA tools and reproduce family, register and spill counts.
01 / Question & hypothesis

What was tested?

The sm_120 compiler already maps the external SHA-256d arithmetic to Blackwell-native shift, Boolean and add instructions.

02 / Scientific basis

Why the test is meaningful

Static disassembly can verify instruction selection and memory behavior even when performance counters are unavailable.

native share=(SHF+LOP3+IADD3)/all SASShot-path global loads/stores=0spill count=0
Static Blackwell SASS auditVisual reading of the published metrics and gates for EXP-059; it summarizes the registered result, not a mining advantage.EXP-059 / STATIC BLACKWELL SASS AUDITNATIVE INTEGER SHARE88.10%BINARIES PASSING9 / 9HOT-PATH SPILLS0
FIGURE / RESULT READINGVisual reading of the published metrics and gates for EXP-059; it summarizes the registered result, not a mining advantage.
03 / Method

How it was tested

Disassemble the nine exact EXP-133 binaries with cuobjdump and count instruction families, registers, local memory and hot-path memory operations.

04 / Observed result

What happened

Native integer share88.10%
Binaries passing9 / 9
Hot-path spills0

Each binary emitted 2,512 SASS instructions: 1,176 SHF, 629–630 LOP3 and 408 IADD3. Those families represented 88.097–88.137% of static code. The hot body had no global loads/stores, stack use or local memory; all 9/9 gates passed.

05 / Validation

Exactness and statistical controls

The audit used already frozen binaries and did not substitute a new benchmark. Counts were repeated across all geometries.

06 / Interpretation

What the result means

Blind intrinsic replacement and blind geometry tuning are closed: the compiler already emits the expected native integer forms.

Limitations

  • Static counts do not reveal runtime issue stalls.
  • Instruction mix is architecture-specific.
  • Native mapping alone is not a mining advantage.
07 / Reproduction

Evidence trail

Disassemble the sealed binaries with matching CUDA tools and reproduce family, register and spill counts.

Canonical variants

CANONICAL-EXP-135PREREGISTERED-CAMPAIGNSEALED-AUDIT

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