Across commits — algorithms-gpu profile

Intel Core i9-9900K 8c/16t · 62.6 GiB · CoffeeLake-S GT2 [UHD Graphics 630] + Ellesmere [Radeon RX 470/480/570/570X/580/580X/590] (8.0 GiB) · machine id 5cf3f79c

5 runs, oldest first. Only runs from this machine are compared — a cross-machine line would be a hardware history wearing a performance history's clothes.

  1. 1f324088 · 2026-08-10 19:05 — round 65.9: first published run of the algorithms-gpu profile — executor cpu vs gpu per family per size on the amd gcn-4 adapter, after the 65.8 kernel perf pass
  2. 44e55dc9 · 2026-08-10 22:13 — round 65.10: executor sweep with the pageRankDense family; pageRank and hierarchical now route 'auto' to the CPU (sparse iteration / flat merges), the GPU keeping the seven families it wins
  3. c2d24c2e · 2026-08-12 23:08 — round 68 status refresh: cpu vs gpu executor sweep on the RX 580
  4. 0d3561de · 2026-08-13 01:30 — post-merge baseline at 0d3561de: executor sweep incl. the round 69/70 families
  5. 9ed49abc · 2026-09-02 01:47 — round 113 verification: executor sweep on the RX 580 at 9ed49abc — no src/algorithms change since 13 Aug; the GPU side reads 20-50% faster at n=512 (driver/box), the cpu control flat

Of the 171 rows that could be compared, none moved beyond its own measured noise in the slower direction. 31 other row(s) moved beyond ±10% but carry no evidence about their own noise — listed below, ranked nowhere.

Drift: across the 171 row measurements shared between consecutive runs, the geometric mean change is −4.5%. A row moving near that factor is the box (thermals, load, driver), not the commit — and the frozen twin beside each mover is the same control per row, since v3 (and, for the executor sweep, the cpu side) is not what changed.

Runs are grouped by machine and by harness version. The benchmark harness has been corrected several times — most consequentially round 62.5c, which equalised inline-cache state between the v3 and v4 sides and moved v4 rows 12–35% with the library untouched. Rows measured across such a change are not comparable, so this page shows a ⋮ harness break instead of a change. Within one epoch, a change is screened against the row's own run-to-run band where the run was made with --repeat.

Slower than the previous run (beyond ±10%, evidence-backed)

No v4 row moved beyond its own measured noise in the slower direction.

Faster than the previous run (beyond ±10%, evidence-backed)

No v4 row moved beyond its own measured noise in the faster direction.

Moved, but unscreened 31 row(s) — one-shot, or inside their own run-to-run band. Not evidence of anything, in either direction.
rowp50changenoise controlwhy unscreenedspan
algorithms-gpu · katzCentrality [cpu] 100 µs → 200 µs +100% no frozen twin no repeats — this row has no noise band 0d3561de → 9ed49abc
algorithms-gpu · affinityPropagation [gpu] 114 ms → 54.6 ms −51.9% +0.0% (cpu twin) no repeats — this row has no noise band 0d3561de → 9ed49abc
algorithms-gpu · affinityPropagation [gpu first call] 117 ms → 59.1 ms −49.4% +0.0% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · katzCentrality [cpu] 500 µs → 300 µs −40.0% no frozen twin no repeats — this row has no noise band 0d3561de → 9ed49abc
algorithms-gpu · markovClustering [gpu] 8.80 ms → 5.30 ms −39.8% −1.5% (cpu twin) no repeats — this row has no noise band 0d3561de → 9ed49abc
algorithms-gpu · heatKernel [gpu] 11.1 ms → 7.00 ms −36.9% −0.5% (cpu twin) no repeats — this row has no noise band 0d3561de → 9ed49abc
algorithms-gpu · markovClustering [gpu first call] 33.0 ms → 21.9 ms −33.6% −0.4% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · simRank [gpu first call] 43.3 ms → 29.7 ms −31.4% +1.9% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · kMedoids [gpu] 4.70 ms → 3.50 ms −25.5% −1.3% (cpu twin) no repeats — this row has no noise band 0d3561de → 9ed49abc
algorithms-gpu · floydWarshall [gpu] 15.3 ms → 11.6 ms −24.2% +1.1% (cpu twin) no repeats — this row has no noise band 0d3561de → 9ed49abc
algorithms-gpu · katzCentrality [gpu first call] 8.70 ms → 10.8 ms +24% −40.0% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · kMedoids [gpu] 4.20 ms → 3.20 ms −23.8% −7.6% (cpu twin) no repeats — this row has no noise band 0d3561de → 9ed49abc
algorithms-gpu · effectiveResistance [gpu first call] 18.2 ms → 14.0 ms −23.1% +1.4% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · effectiveResistance [gpu] 24.9 ms → 20.0 ms −19.7% −0.6% (cpu twin) no repeats — this row has no noise band 0d3561de → 9ed49abc
algorithms-gpu · motifCensus [gpu first call] 137 ms → 111 ms −19.1% +5.8% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · simRank [gpu first call] 15.9 ms → 13.0 ms −18.2% +0.4% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · rwrProximity [gpu] 5.80 ms → 4.80 ms −17.2% −3.0% (cpu twin) no repeats — this row has no noise band 0d3561de → 9ed49abc
algorithms-gpu · affinityPropagation [gpu] 15.4 ms → 12.8 ms −16.9% +0.0% (cpu twin) no repeats — this row has no noise band 0d3561de → 9ed49abc
algorithms-gpu · closenessCentralityNormalized [gpu first call] 33.2 ms → 27.8 ms −16.3% −0.3% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · closenessCentralityNormalized [gpu] 8.50 ms → 7.20 ms −15.3% +0.2% (cpu twin) no repeats — this row has no noise band 0d3561de → 9ed49abc
algorithms-gpu · betweennessCentrality [gpu first call] 16.6 ms → 19.1 ms +15% −0.2% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · markovClustering [gpu first call] 62.7 ms → 53.3 ms −15.0% −1.1% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · kMeans [gpu first call] 14.8 ms → 12.6 ms −14.9% +3.7% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · neighborhoodSimilarity [gpu first call] 42.7 ms → 36.4 ms −14.8% −0.0% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · floydWarshall [gpu first call] 15.9 ms → 13.7 ms −13.8% +1.1% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · floydWarshall [gpu] 43.9 ms → 37.9 ms −13.7% −1.6% (cpu twin) no repeats — this row has no noise band 0d3561de → 9ed49abc
algorithms-gpu · kMeans [gpu first call] 16.4 ms → 18.5 ms +13% −0.2% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · floydWarshall [gpu first call] 44.6 ms → 40.0 ms −10.3% −1.6% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · kMeans [gpu first call] 77.5 ms → 69.7 ms −10.1% −3.5% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · closenessCentralityNormalized [gpu first call] 9.00 ms → 8.10 ms −10.0% +0.2% (cpu twin) one-shot row (a single measurement) 0d3561de → 9ed49abc
algorithms-gpu · kMedoids [gpu] 10.0 ms → 11.0 ms +10% +6.0% (cpu twin) no repeats — this row has no noise band 0d3561de → 9ed49abc

Every row, every run

Each cell is that run's p50 with its change against the nearest earlier run that measured the same row. A dash means the run did not measure it (a --suite filter, or the row did not exist yet) — gaps are shown, never dropped.

algorithms-gpu n=256 · 24 rows
row 1f324088
2026-08-10
44e55dc9
2026-08-10
c2d24c2e
2026-08-12
0d3561de
2026-08-13
9ed49abc
2026-09-02
markovClustering [cpu]612 ms615 ms ⋮ harness625 ms ⋮ harness617 ms ⋮ harness607 ms −1.5%
markovClustering [gpu]8.50 ms8.80 ms ⋮ harness9.00 ms ⋮ harness8.80 ms ⋮ harness5.30 ms −39.8%
markovClustering [gpu first call]31.7 ms28.0 ms ⋮ harness27.4 ms ⋮ harness26.7 ms ⋮ harness25.7 ms −3.7%
floydWarshall [cpu]23.3 ms23.4 ms ⋮ harness23.3 ms ⋮ harness23.4 ms ⋮ harness23.2 ms −0.9%
floydWarshall [gpu]6.90 ms6.90 ms ⋮ harness6.90 ms ⋮ harness6.40 ms ⋮ harness6.40 ms −0.0%
floydWarshall [gpu first call]20.3 ms20.1 ms ⋮ harness29.6 ms ⋮ harness20.6 ms ⋮ harness20.6 ms +0.0%
affinityPropagation [cpu]39.8 ms39.5 ms ⋮ harness38.4 ms ⋮ harness40.6 ms ⋮ harness40.6 ms +0.0%
affinityPropagation [gpu]15.1 ms15.5 ms ⋮ harness15.8 ms ⋮ harness15.4 ms ⋮ harness12.8 ms −16.9%
affinityPropagation [gpu first call]25.5 ms27.9 ms ⋮ harness26.0 ms ⋮ harness25.5 ms ⋮ harness25.2 ms −1.2%
closenessCentralityNormalized [cpu]24.1 ms23.2 ms −3.7%
closenessCentralityNormalized [gpu]3.60 ms3.60 ms −0.0%
closenessCentralityNormalized [gpu first call]7.50 ms7.30 ms −2.7%
effectiveResistance [cpu]51.7 ms52.4 ms +1.4%
effectiveResistance [gpu]7.60 ms7.10 ms −6.6%
effectiveResistance [gpu first call]18.2 ms14.0 ms −23.1%
simRank [cpu]146 ms147 ms +0.4%
simRank [gpu]4.50 ms4.40 ms −2.2%
simRank [gpu first call]15.9 ms13.0 ms −18.2%
rwrProximity [cpu]173 ms167 ms −3.0%
rwrProximity [gpu]5.80 ms4.80 ms −17.2%
rwrProximity [gpu first call]13.5 ms14.8 ms +9.6%
heatKernel [cpu]308 ms305 ms −1.0%
heatKernel [gpu]3.70 ms4.00 ms +8.1%
heatKernel [gpu first call]14.5 ms13.2 ms −9.0%
algorithms-gpu n=512 · 45 rows
row 1f324088
2026-08-10
44e55dc9
2026-08-10
c2d24c2e
2026-08-12
0d3561de
2026-08-13
9ed49abc
2026-09-02
markovClustering [cpu]4.78 s4.79 s ⋮ harness4.89 s ⋮ harness4.77 s ⋮ harness4.75 s −0.4%
markovClustering [gpu]15.9 ms16.8 ms ⋮ harness16.8 ms ⋮ harness16.7 ms ⋮ harness15.8 ms −5.4%
markovClustering [gpu first call]25.6 ms25.8 ms ⋮ harness27.2 ms ⋮ harness33.0 ms ⋮ harness21.9 ms −33.6%
pageRank [cpu]8.00 ms400 µs ⋮ harness600 µs ⋮ harness400 µs ⋮ harness400 µs −0.0%
pageRank [gpu]6.30 ms6.90 ms ⋮ harness6.30 ms ⋮ harness6.60 ms ⋮ harness6.60 ms −0.0%
pageRank [gpu first call]13.3 ms13.5 ms ⋮ harness13.6 ms ⋮ harness13.7 ms ⋮ harness13.4 ms −2.2%
floydWarshall [cpu]179 ms179 ms ⋮ harness180 ms ⋮ harness180 ms ⋮ harness182 ms +1.1%
floydWarshall [gpu]16.1 ms15.5 ms ⋮ harness15.1 ms ⋮ harness15.3 ms ⋮ harness11.6 ms −24.2%
floydWarshall [gpu first call]18.0 ms16.7 ms ⋮ harness17.5 ms ⋮ harness15.9 ms ⋮ harness13.7 ms −13.8%
betweennessCentrality [cpu]34.6 ms36.0 ms ⋮ harness34.4 ms ⋮ harness36.4 ms ⋮ harness34.8 ms −4.4%
betweennessCentrality [gpu]9.10 ms9.00 ms ⋮ harness8.80 ms ⋮ harness9.00 ms ⋮ harness8.90 ms −1.1%
betweennessCentrality [gpu first call]19.6 ms19.3 ms ⋮ harness20.0 ms ⋮ harness19.1 ms ⋮ harness19.2 ms +0.5%
affinityPropagation [cpu]209 ms206 ms ⋮ harness209 ms ⋮ harness204 ms ⋮ harness204 ms +0.0%
affinityPropagation [gpu]112 ms112 ms ⋮ harness112 ms ⋮ harness114 ms ⋮ harness54.6 ms −51.9%
affinityPropagation [gpu first call]117 ms117 ms ⋮ harness115 ms ⋮ harness117 ms ⋮ harness59.1 ms −49.4%
pageRankDense [cpu]900 µs1.00 ms ⋮ harness1.00 ms ⋮ harness1.00 ms +0.0%
pageRankDense [gpu]6.80 ms6.90 ms ⋮ harness7.40 ms ⋮ harness6.70 ms −9.5%
pageRankDense [gpu first call]31.5 ms30.0 ms ⋮ harness27.8 ms ⋮ harness30.0 ms +7.9%
closenessCentralityNormalized [cpu]180 ms180 ms +0.2%
closenessCentralityNormalized [gpu]8.50 ms7.20 ms −15.3%
closenessCentralityNormalized [gpu first call]9.00 ms8.10 ms −10.0%
triangleCount [cpu]16.2 ms15.8 ms −2.5%
triangleCount [gpu]7.10 ms6.90 ms −2.8%
triangleCount [gpu first call]27.0 ms29.1 ms +7.8%
neighborhoodSimilarity [cpu]14.6 ms14.6 ms −0.0%
neighborhoodSimilarity [gpu]7.40 ms7.30 ms −1.4%
neighborhoodSimilarity [gpu first call]42.7 ms36.4 ms −14.8%
katzCentrality [cpu]500 µs300 µs −40.0%
katzCentrality [gpu]6.40 ms6.00 ms −6.2%
katzCentrality [gpu first call]8.70 ms10.8 ms +24%
effectiveResistance [cpu]413 ms411 ms −0.6%
effectiveResistance [gpu]24.9 ms20.0 ms −19.7%
effectiveResistance [gpu first call]20.6 ms19.3 ms −6.3%
simRank [cpu]1.05 s1.07 s +1.9%
simRank [gpu]12.0 ms11.9 ms −0.8%
simRank [gpu first call]43.3 ms29.7 ms −31.4%
rwrProximity [cpu]1.31 s1.29 s −1.3%
rwrProximity [gpu]10.5 ms10.4 ms −1.0%
rwrProximity [gpu first call]33.6 ms31.1 ms −7.4%
heatKernel [cpu]4.74 s4.72 s −0.5%
heatKernel [gpu]11.1 ms7.00 ms −36.9%
heatKernel [gpu first call]29.4 ms28.4 ms −3.4%
motifCensus [cpu]42.3 ms42.5 ms +0.5%
motifCensus [gpu]20.1 ms21.2 ms +5.5%
motifCensus [gpu first call]48.5 ms47.0 ms −3.1%
algorithms-gpu n=1024 · 51 rows
row 1f324088
2026-08-10
44e55dc9
2026-08-10
c2d24c2e
2026-08-12
0d3561de
2026-08-13
9ed49abc
2026-09-02
markovClustering [cpu]30.2 s30.3 s ⋮ harness30.3 s ⋮ harness30.4 s ⋮ harness30.0 s −1.1%
markovClustering [gpu]47.1 ms46.8 ms ⋮ harness46.9 ms ⋮ harness46.6 ms ⋮ harness46.2 ms −0.9%
markovClustering [gpu first call]62.0 ms62.1 ms ⋮ harness62.2 ms ⋮ harness62.7 ms ⋮ harness53.3 ms −15.0%
pageRank [cpu]32.1 ms100 µs ⋮ harness300 µs ⋮ harness400 µs ⋮ harness400 µs −0.0%
pageRank [gpu]19.2 ms17.8 ms ⋮ harness18.0 ms ⋮ harness18.1 ms ⋮ harness18.2 ms +0.6%
pageRank [gpu first call]21.3 ms19.8 ms ⋮ harness19.2 ms ⋮ harness20.3 ms ⋮ harness20.2 ms −0.5%
floydWarshall [cpu]1.35 s1.41 s ⋮ harness1.41 s ⋮ harness1.43 s ⋮ harness1.40 s −1.6%
floydWarshall [gpu]42.1 ms48.0 ms ⋮ harness46.6 ms ⋮ harness43.9 ms ⋮ harness37.9 ms −13.7%
floydWarshall [gpu first call]46.5 ms51.5 ms ⋮ harness51.1 ms ⋮ harness44.6 ms ⋮ harness40.0 ms −10.3%
betweennessCentrality [cpu]153 ms149 ms ⋮ harness149 ms ⋮ harness153 ms ⋮ harness152 ms −0.2%
betweennessCentrality [gpu]15.2 ms15.6 ms ⋮ harness15.4 ms ⋮ harness15.2 ms ⋮ harness15.2 ms −0.0%
betweennessCentrality [gpu first call]17.5 ms17.1 ms ⋮ harness17.5 ms ⋮ harness16.6 ms ⋮ harness19.1 ms +15%
affinityPropagation [cpu]1.49 s1.11 s ⋮ harness1.15 s ⋮ harness1.10 s ⋮ harness1.07 s −2.1%
affinityPropagation [gpu]293 ms297 ms ⋮ harness295 ms ⋮ harness296 ms ⋮ harness293 ms −0.9%
affinityPropagation [gpu first call]305 ms296 ms ⋮ harness292 ms ⋮ harness301 ms ⋮ harness291 ms −3.3%
kMedoids [cpu]78.6 ms82.5 ms ⋮ harness72.6 ms ⋮ harness80.2 ms ⋮ harness74.1 ms −7.6%
kMedoids [gpu]3.40 ms4.10 ms ⋮ harness3.20 ms ⋮ harness4.20 ms ⋮ harness3.20 ms −23.8%
kMedoids [gpu first call]12.2 ms12.7 ms ⋮ harness13.5 ms ⋮ harness13.1 ms ⋮ harness12.0 ms −8.4%
hierarchicalClustering [cpu]73.5 ms27.9 ms ⋮ harness28.0 ms ⋮ harness27.9 ms ⋮ harness27.3 ms −2.2%
hierarchicalClustering [gpu]34.8 ms27.8 ms ⋮ harness27.0 ms ⋮ harness29.5 ms ⋮ harness26.9 ms −8.8%
hierarchicalClustering [gpu first call]34.4 ms37.3 ms ⋮ harness36.0 ms ⋮ harness31.4 ms ⋮ harness32.9 ms +4.8%
pageRankDense [cpu]3.60 ms3.90 ms ⋮ harness3.80 ms ⋮ harness3.70 ms −2.6%
pageRankDense [gpu]20.5 ms20.7 ms ⋮ harness21.2 ms ⋮ harness19.3 ms −9.0%
pageRankDense [gpu first call]96.6 ms107 ms ⋮ harness105 ms ⋮ harness107 ms +1.8%
closenessCentralityNormalized [cpu]1.29 s1.29 s −0.3%
closenessCentralityNormalized [gpu]27.7 ms26.4 ms −4.7%
closenessCentralityNormalized [gpu first call]33.2 ms27.8 ms −16.3%
triangleCount [cpu]107 ms104 ms −2.3%
triangleCount [gpu]22.3 ms22.2 ms −0.4%
triangleCount [gpu first call]95.0 ms103 ms +8.2%
neighborhoodSimilarity [cpu]90.3 ms90.9 ms +0.7%
neighborhoodSimilarity [gpu]23.3 ms23.5 ms +0.9%
neighborhoodSimilarity [gpu first call]92.3 ms90.9 ms −1.5%
katzCentrality [cpu]100 µs200 µs +100%
katzCentrality [gpu]12.1 ms13.0 ms +7.4%
katzCentrality [gpu first call]13.0 ms14.0 ms +7.7%
effectiveResistance [cpu]3.44 s3.39 s −1.3%
effectiveResistance [gpu]79.5 ms79.7 ms +0.3%
effectiveResistance [gpu first call]82.0 ms81.9 ms −0.1%
simRank [cpu]8.45 s8.37 s −0.9%
simRank [gpu]57.6 ms53.7 ms −6.8%
simRank [gpu first call]142 ms129 ms −9.4%
rwrProximity [cpu]10.4 s10.3 s −0.5%
rwrProximity [gpu]36.6 ms37.2 ms +1.6%
rwrProximity [gpu first call]120 ms109 ms −9.1%
heatKernel [cpu]75.4 s74.9 s −0.7%
heatKernel [gpu]33.3 ms33.7 ms +1.2%
heatKernel [gpu first call]127 ms116 ms −8.6%
motifCensus [cpu]311 ms329 ms +5.8%
motifCensus [gpu]38.2 ms41.5 ms +8.6%
motifCensus [gpu first call]137 ms111 ms −19.1%
algorithms-gpu n=2048 · 27 rows
row 1f324088
2026-08-10
44e55dc9
2026-08-10
c2d24c2e
2026-08-12
0d3561de
2026-08-13
9ed49abc
2026-09-02
pageRank [cpu]122 ms300 µs ⋮ harness600 µs ⋮ harness600 µs ⋮ harness600 µs +0.0%
pageRank [gpu]71.9 ms73.2 ms ⋮ harness72.5 ms ⋮ harness72.4 ms ⋮ harness69.0 ms −4.7%
pageRank [gpu first call]90.7 ms87.8 ms ⋮ harness85.3 ms ⋮ harness88.0 ms ⋮ harness83.1 ms −5.6%
betweennessCentrality [cpu]642 ms650 ms ⋮ harness640 ms ⋮ harness648 ms ⋮ harness635 ms −2.0%
betweennessCentrality [gpu]35.9 ms37.5 ms ⋮ harness36.5 ms ⋮ harness35.3 ms ⋮ harness35.1 ms −0.6%
betweennessCentrality [gpu first call]41.6 ms43.7 ms ⋮ harness42.7 ms ⋮ harness42.3 ms ⋮ harness39.1 ms −7.6%
kMedoids [cpu]382 ms246 ms ⋮ harness271 ms ⋮ harness386 ms ⋮ harness381 ms −1.3%
kMedoids [gpu]3.70 ms4.60 ms ⋮ harness4.60 ms ⋮ harness4.70 ms ⋮ harness3.50 ms −25.5%
kMedoids [gpu first call]5.90 ms6.00 ms ⋮ harness5.90 ms ⋮ harness4.50 ms ⋮ harness4.60 ms +2.2%
hierarchicalClustering [cpu]307 ms99.3 ms ⋮ harness98.6 ms ⋮ harness99.7 ms ⋮ harness98.2 ms −1.5%
hierarchicalClustering [gpu]162 ms108 ms ⋮ harness108 ms ⋮ harness111 ms ⋮ harness105 ms −5.8%
hierarchicalClustering [gpu first call]168 ms103 ms ⋮ harness106 ms ⋮ harness119 ms ⋮ harness131 ms +9.7%
pageRankDense [cpu]14.8 ms15.4 ms ⋮ harness15.3 ms ⋮ harness15.4 ms +0.7%
pageRankDense [gpu]81.8 ms82.5 ms ⋮ harness80.6 ms ⋮ harness80.5 ms −0.1%
pageRankDense [gpu first call]395 ms380 ms ⋮ harness392 ms ⋮ harness397 ms +1.2%
triangleCount [cpu]769 ms765 ms −0.4%
triangleCount [gpu]89.4 ms85.1 ms −4.8%
triangleCount [gpu first call]417 ms398 ms −4.6%
neighborhoodSimilarity [cpu]630 ms602 ms −4.5%
neighborhoodSimilarity [gpu]95.0 ms91.7 ms −3.5%
neighborhoodSimilarity [gpu first call]419 ms406 ms −3.1%
katzCentrality [cpu]300 µs300 µs −0.0%
katzCentrality [gpu]36.0 ms34.9 ms −3.1%
katzCentrality [gpu first call]38.3 ms37.2 ms −2.9%
motifCensus [cpu]2.86 s2.88 s +0.6%
motifCensus [gpu]127 ms121 ms −4.0%
motifCensus [gpu first call]453 ms460 ms +1.6%
algorithms-gpu n=4096 · 12 rows
row 1f324088
2026-08-10
44e55dc9
2026-08-10
c2d24c2e
2026-08-12
0d3561de
2026-08-13
9ed49abc
2026-09-02
kMeans [cpu]31.4 ms30.5 ms ⋮ harness30.6 ms ⋮ harness29.7 ms ⋮ harness30.8 ms +3.7%
kMeans [gpu]4.00 ms4.00 ms ⋮ harness3.90 ms ⋮ harness4.00 ms ⋮ harness4.00 ms +0.0%
kMeans [gpu first call]12.2 ms14.1 ms ⋮ harness12.2 ms ⋮ harness14.8 ms ⋮ harness12.6 ms −14.9%
fuzzyCMeans [cpu]157 ms156 ms ⋮ harness152 ms ⋮ harness156 ms ⋮ harness153 ms −2.4%
fuzzyCMeans [gpu]5.00 ms5.80 ms ⋮ harness4.80 ms ⋮ harness5.60 ms ⋮ harness5.10 ms −8.9%
fuzzyCMeans [gpu first call]17.8 ms18.8 ms ⋮ harness18.0 ms ⋮ harness18.5 ms ⋮ harness17.8 ms −3.8%
kMedoids [cpu]1.74 s1.46 s ⋮ harness1.71 s ⋮ harness1.65 s ⋮ harness1.75 s +6.0%
kMedoids [gpu]10.7 ms12.0 ms ⋮ harness12.0 ms ⋮ harness10.0 ms ⋮ harness11.0 ms +10%
kMedoids [gpu first call]12.1 ms12.0 ms ⋮ harness12.0 ms ⋮ harness12.3 ms ⋮ harness12.0 ms −2.4%
hierarchicalClustering [cpu]1.24 s420 ms ⋮ harness422 ms ⋮ harness431 ms ⋮ harness422 ms −2.2%
hierarchicalClustering [gpu]714 ms452 ms ⋮ harness439 ms ⋮ harness447 ms ⋮ harness439 ms −1.8%
hierarchicalClustering [gpu first call]702 ms478 ms ⋮ harness493 ms ⋮ harness500 ms ⋮ harness469 ms −6.1%
algorithms-gpu n=16384 · 6 rows
row 1f324088
2026-08-10
44e55dc9
2026-08-10
c2d24c2e
2026-08-12
0d3561de
2026-08-13
9ed49abc
2026-09-02
kMeans [cpu]129 ms127 ms ⋮ harness124 ms ⋮ harness126 ms ⋮ harness126 ms −0.2%
kMeans [gpu]7.70 ms9.00 ms ⋮ harness7.30 ms ⋮ harness7.70 ms ⋮ harness7.50 ms −2.6%
kMeans [gpu first call]16.7 ms17.7 ms ⋮ harness17.4 ms ⋮ harness16.4 ms ⋮ harness18.5 ms +13%
fuzzyCMeans [cpu]613 ms616 ms ⋮ harness600 ms ⋮ harness612 ms ⋮ harness601 ms −1.7%
fuzzyCMeans [gpu]12.4 ms16.3 ms ⋮ harness12.3 ms ⋮ harness12.8 ms ⋮ harness11.7 ms −8.6%
fuzzyCMeans [gpu first call]21.1 ms22.8 ms ⋮ harness22.0 ms ⋮ harness22.8 ms ⋮ harness21.5 ms −5.7%
algorithms-gpu n=65536 · 6 rows
row 1f324088
2026-08-10
44e55dc9
2026-08-10
c2d24c2e
2026-08-12
0d3561de
2026-08-13
9ed49abc
2026-09-02
kMeans [cpu]628 ms613 ms ⋮ harness611 ms ⋮ harness630 ms ⋮ harness608 ms −3.5%
kMeans [gpu]27.4 ms27.7 ms ⋮ harness24.3 ms ⋮ harness28.3 ms ⋮ harness27.1 ms −4.2%
kMeans [gpu first call]68.5 ms78.4 ms ⋮ harness70.3 ms ⋮ harness77.5 ms ⋮ harness69.7 ms −10.1%
fuzzyCMeans [cpu]2.54 s2.57 s ⋮ harness2.53 s ⋮ harness2.64 s ⋮ harness2.57 s −2.6%
fuzzyCMeans [gpu]38.5 ms36.8 ms ⋮ harness42.6 ms ⋮ harness38.0 ms ⋮ harness37.2 ms −2.1%
fuzzyCMeans [gpu first call]82.5 ms84.8 ms ⋮ harness84.3 ms ⋮ harness85.4 ms ⋮ harness91.4 ms +7.0%