== 64k repair threads=4 CPU-s: median of both copies of each arm (n legs); A/A = worst |X - X2| / min over the reps; verdict clears the A/A floor or is UNRESOLVED m | fold force F/T | aa_F aa_T floor | verdict | wF wT wF/wT | foreign max | peak MB force/fold 96 | 10.74 13.97 0.769 | 6.4% 43.5% 43.5% | unres | 2.06 3.03 0.681 | 8% (8 legs) | 1084/592 128 | 10.88 13.66 0.796 | 3.3% 2.5% 3.3% | fold | 2.24 2.91 0.770 | 8% (8 legs) | 1083/614 160 | 12.00 14.38 0.834 | 14.4% 2.0% 14.4% | fold | 2.34 3.15 0.742 | 8% (8 legs) | 1089/602 192 | 13.17 14.02 0.940 | 7.6% 2.8% 7.6% | unres | 2.69 3.03 0.885 | 8% (8 legs) | 1091/575 224 | 14.18 14.50 0.978 | 7.2% 30.5% 30.5% | unres | 2.94 3.14 0.936 | 8% (8 legs) | 1102/587 256 | 15.54 14.85 1.047 | 2.3% 8.2% 8.2% | unres | 3.23 3.25 0.995 | 8% (8 legs) | 1100/562 288 | 16.14 14.19 1.138 | 2.7% 24.5% 24.5% | unres | 3.46 3.13 1.104 | 8% (8 legs) | 1100/593 320 | 18.23 14.93 1.220 | 24.6% 6.3% 24.6% | unres | 3.92 3.29 1.194 | 7% (8 legs) | 1105/569 384 | 19.22 14.94 1.286 | 29.4% 9.4% 29.4% | unres | 4.23 3.26 1.298 | 7% (8 legs) | 1103/602 512 | 24.48 15.28 1.602 | 6.2% 27.3% 27.3% | ntt | 5.61 3.40 1.648 | 7% (8 legs) | 1108/608 crossover (log-interpolated, median): CPU m ~ 234 wall m ~ 257 foreign CPU over 80 legs: median 6% of a core, max 8% (after-leg median 6%) == 64k repair threads=8 CPU-s: median of both copies of each arm (n legs); A/A = worst |X - X2| / min over the reps; verdict clears the A/A floor or is UNRESOLVED m | fold force F/T | aa_F aa_T floor | verdict | wF wT wF/wT | foreign max | peak MB force/fold 96 | 10.54 13.26 0.795 | 6.2% 10.5% 10.5% | fold | 1.55 1.96 0.795 | 8% (8 legs) | 1112/487 128 | 11.63 13.93 0.835 | 26.2% 54.3% 54.3% | unres | 1.73 2.04 0.847 | 8% (8 legs) | 1112/515 160 | 13.16 15.03 0.875 | 8.8% 15.9% 15.9% | unres | 1.96 2.18 0.897 | 8% (8 legs) | 1116/523 192 | 13.61 13.66 0.996 | 5.5% 1.7% 5.5% | unres | 1.96 1.98 0.991 | 17% (8 legs) | 1121/516 224 | 14.82 14.70 1.008 | 2.3% 31.9% 31.9% | unres | 2.13 2.14 0.995 | 8% (8 legs) | 1124/504 256 | 15.84 14.49 1.093 | 2.2% 6.6% 6.6% | ntt | 2.25 2.09 1.077 | 8% (8 legs) | 1123/513 288 | 16.54 14.19 1.166 | 8.8% 5.8% 8.8% | ntt | 2.29 2.06 1.113 | 9% (8 legs) | 1125/492 320 | 18.23 15.02 1.214 | 6.6% 5.5% 6.6% | ntt | 2.53 2.19 1.156 | 12% (8 legs) | 1132/491 384 | 20.89 15.35 1.361 | 8.8% 11.4% 11.4% | ntt | 2.88 2.26 1.274 | 7% (8 legs) | 1134/491 512 | 26.45 15.31 1.727 | 5.5% 1.6% 5.5% | ntt | 3.66 2.25 1.622 | 8% (8 legs) | 1149/509 crossover (log-interpolated, median): CPU m ~ 202 wall m ~ 226 foreign CPU over 80 legs: median 7% of a core, max 17% (after-leg median 6%) == 64k threads=4 (k phase; every cell the minimum over reps) fold m=192:13.26 m=512:25.07 m=1024:42.30 m=2048:69.78 m=4096:115.94 c_f = 1.581e-06 CPU-s per source-row per 64 KiB (slope 0.0259 over n=16384, block 65536 B, intercept 12.48, worst residual 4.24) c_f FITTED OVER RUNGS m = 192,512,1024,2048,4096 (every fold rung in the log) - a c_f quoted without this cannot be compared with another forcep big m=256 cpu= 13.51 windows=1 slabs=1 w=65536 combine/win=0.140 c_w=5.469e-04 c_l=4.098e-04 forcep big m=1024 cpu= 15.73 windows=1 slabs=1 w=65536 combine/win=0.380 c_w=3.711e-04 c_l=4.570e-04 forcep big m=4096 cpu= 18.43 windows=1 slabs=1 w=65536 combine/win=1.470 c_w=3.589e-04 c_l=4.753e-04 forcep 128 m=256 cpu= 14.39 windows=8 slabs=1 w=65536 combine/win=0.091 c_w=3.564e-04 forcep 128 m=1024 cpu= 18.84 windows=8 slabs=2 w=32768 combine/win=0.185 c_w=3.613e-04 forcep 128 m=4096 cpu= 36.21 windows=8 slabs=4 w=16384 combine/win=0.371 c_w=3.625e-04 c_w median over 6 cells = 3.619e-04 (range 3.564e-04 .. 5.469e-04) k = c_w / c_f = 229 [c_f rungs m = 192,512,1024,2048,4096] c_l median = 4.570e-04; c_l / c_f = 289 rows == 64k threads=8 (k phase; every cell the minimum over reps) fold m=192:13.59 m=512:24.98 m=1024:44.16 m=2048:72.32 m=4096:122.01 c_f = 1.674e-06 CPU-s per source-row per 64 KiB (slope 0.0274 over n=16384, block 65536 B, intercept 12.22, worst residual 3.92) c_f FITTED OVER RUNGS m = 192,512,1024,2048,4096 (every fold rung in the log) - a c_f quoted without this cannot be compared with another forcep big m=256 cpu= 15.06 windows=1 slabs=1 w=65536 combine/win=0.180 c_w=7.031e-04 c_l=4.818e-04 forcep big m=1024 cpu= 17.04 windows=1 slabs=1 w=65536 combine/win=0.410 c_w=4.004e-04 c_l=4.746e-04 forcep big m=4096 cpu= 18.34 windows=1 slabs=1 w=65536 combine/win=1.560 c_w=3.809e-04 c_l=4.736e-04 forcep 128 m=256 cpu= 15.70 windows=8 slabs=1 w=65536 combine/win=0.121 c_w=4.736e-04 forcep 128 m=1024 cpu= 17.64 windows=8 slabs=2 w=32768 combine/win=0.207 c_w=4.053e-04 forcep 128 m=4096 cpu= 41.37 windows=8 slabs=4 w=16384 combine/win=0.446 c_w=4.358e-04 c_w median over 6 cells = 4.205e-04 (range 3.809e-04 .. 7.031e-04) k = c_w / c_f = 251 [c_f rungs m = 192,512,1024,2048,4096] c_l median = 4.746e-04; c_l / c_f = 283 rows