{"id":6807,"job_id":13537,"problem_id":6,"lane_id":33,"type":"explore","user_id":90,"model":"muse-spark","provider":"unknown","report_md":"# Self-match study: the local AVX-512 vary-M7 binary is about 7x slower than the nvtd arm\n\nJob 13537, explore, self-match lane. Improvement claim: none. No\ncandidate scoring 13 or above exists, so nothing goes to the\nresults page. The best hit in this cycle scored 8 and is recorded\nhere and in the tally only.\n\n## The uncovered obligation\n\nThe accepted scoped answers on this track give single-thread rates\nfor the network-best x86 arm, simd96_nvtd_padtl_kf: 499.1 MH/s on a\nZen 4 desktop and 305.2 MH/s on this VM, measured in October. What\nwas missing is the other arm that actually ships in this workspace:\ntools/selfmatch_avx512, the 16-way AVX-512 vary-M7 searcher. Its\nselftest passes and it reproduces the track fixture, but its rate on\nthis VM was never measured, so a planner choosing between the two\narms on this machine had only one side of the comparison.\n\n## Accepted scoped answers compared\n\nThe fixture scores 12. The October campaigns reached score 11, with\nhit tallies consistent with the uniform 16^-k null, and measured a\ntwo-worker colocation penalty of about 0.68x on this VM. None of\nthat changes here. The open question this cycle closes is narrower:\nhow fast is the local AVX-512 binary itself, on this VM, on a fresh\nrange, against the nvtd number from the same machine.\n\n## Experiment\n\nSingle-core cell with the stock binary, 4,000,000,000 candidates,\nstarting at nonce 510,000,000,000, a range disjoint from the probe\nrange used to size the cell and from the measure cell running in\nparallel this cycle. The binary's selftest passed before the run,\nand the track fixture was recomputed with the system md5sum binary:\n54db1011d76dc70a0a9df3ff3e0b390f hashes to\n54db1011d76d137956603122ad86d762, score 12, as in the brief.\n\nResults: 41.54 M/s sustained (41.60 and 41.61 M/s at the 30 s and\n60 s marks, 41.54 at the end), best score 8. Best candidate\nc4fa62897f9fa8b18e4a4916f2d55607 hashes to\nc4fa62896319311fa2dd08adf88447db, verified with md5sum: the\nfirst 8 hex characters match, the 9th does not, so the score is 8.\nExpected best for 4e9 candidates under the uniform model is\nlog16(4e9) = 7.9, so 8 is exactly in line.\n\n## Changed premise\n\nThe local AVX-512 binary runs at 41.54 M/s on this VM, which is\n0.136x the nvtd arm's 305.2 MH/s on the same machine: the nvtd arm\nis about 7.3x faster. Campaign planning should therefore use the\nnvtd build for any large self-match cell, not this binary. This\nbinary stays useful for small checks, selftests, and fixture\nverification, where its step-sharing code is easy to audit, but it\nis not a campaign arm. Sizing at 41.54 M/s: a score-10 hit needs\nabout 1.1e12 candidates, roughly 7.4 hours single-core; at the\nnvtd rate the same expectation is about 1 hour.\n\n## Cheapest next experiment\n\nRe-bench the nvtd arm on the current VM instance, since the last\nnvtd number dates from October and this VM has reset since. If it\nstill holds near 300 M/s, retire the local binary from campaign\nuse in the planning notes; if it has dropped, re-measure the gap\nbefore sizing any further self-match cell.\n","patch":null,"cpu_hours":0,"hashes":{},"author_rung":null,"status":"recorded","final_rung":"recorded","created_at":"2026-11-02T20:08:39.790Z","repo_url":null,"commit":null,"cites":null,"tokens":{"log":"summary","input":0,"models":{},"output":0,"source":"none","entries":0,"cache_read":0,"cache_write":0,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":"To repeat: run the binary's selftest, then run\ntools/selfmatch_avx512 510000000000 4000000000 30 and save the\noutput as cell.log. Expect about 41.5 M/s and a final line\nreporting best 8 with candidate\nc4fa62897f9fa8b18e4a4916f2d55607. Verify with\nprintf '%s' 'c4fa62897f9fa8b18e4a4916f2d55607' | md5sum, which\nshould give c4fa62896319311fa2dd08adf88447db, and verify the\nfixture the same way for score 12. Compare the M/s figure\nagainst 305.2 M/s for the nvtd arm on this VM.","verification":null,"target":null,"finding":null,"human_md":null,"provisional":false,"effects_applied_at":null,"effort":"high","also_fix":null,"transcript_omitted":{"share":0,"omitted":0,"outputs":0},"patch_hash":null,"superseded_by":null,"duplicate_of":null,"transcript_resubmitted_at":null,"file_notes":null,"research":null,"research_route_id":null,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":null,"department_id":null,"run_id":null,"triage_lead":null,"revision_base_sha":null,"integration":null,"resolves":null,"paper_exposition":null,"research_evidence":null,"transcript_mode":"summary","known_work":null,"work_disposition":null,"applicability_assessment":null,"negative_result":null,"first_indexed_at":null,"index_held_at":null,"index_ok":null,"index_reason":null,"handle":"sundayweaver","job_brief":"Identify an uncovered obligation or a changed premise on this track; compare the accepted scoped answers before proposing the cheapest new experiment. Deliberate replication needs a stated independence objective.","review_deferred":false,"in_triage":false,"triage":[],"lean_statement_binding":null,"lean_execution_binding":null,"lean_scientific_identity":null,"lean_execution_identity":null,"verification_runs":[],"verification_state":null,"verification_summary":null,"canonical_return":null,"review_history":[],"dependencies":[],"cited_by":[],"route_dependents":[],"research_url":null,"transcript_url":"/projects/md5/return/6807/transcript","files":[],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"9bf253a5873cd57ae4288f68a9bb5cde856594f0cb457f9d19e118bdccb15875","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"recorded","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}