[tenancy before] (none)

  shape  geometry                        thr  regs    smem blk/SM warps/SM   bank
  s24    128x128x128 2x4 st2 sw8         256   123   49152      2       16 64
  s27    128x128x128 1x4 st2 sw8         128   199   49152      2        8 128
  s83    128x128x128 2x8 st2 sw8 gs128   512   128   49152      1       16 32x2
  s84    128x128x128 2x8 st2 sw8         512    64   49152      2       32 32   <-- SPILLS
  s85    64x128x128 1x8 st2 sw8 gs128    256   128   32768      2       16 32x2
  s86    64x128x128 1x8 st2 sw8          256    78   32768      3       24 32
  s89    64x128x128 1x8 st2 sw8 gs256    256   126   32768      2       16 32x2
  s94    32x128x128 1x8 st2 sw8 gs128    256    99   24576      2       16 16x2
  s95    32x128x128 1x8 st2 sw8          256    64   24576      4       32 16
  s98    16x128x128 1x8 st2 sw8 gs128    256    63   20480      4       32 8x2
  s99    16x128x128 1x8 st2 sw8          256    63   20480      4       32 8
  s101   32x128x128 1x8 st2 sw8 gs256    256    78   24576      3       24 16x2
  s102   16x128x128 1x8 st2 sw8 gs256    256    63   20480      4       32 8x2
[l2 scrub] mode='read' -- see the module docstring for the bias it carries

  out_proj k=6144 n=5120 M=48   burst>= 50 ms, reps=7
    f101       64.48 us/call  inner=664    burst=   42.8 ms
    f102       53.93 us/call  inner=787    burst=   42.4 ms
    f24       100.45 us/call  inner=449    burst=   45.1 ms
    f27        95.80 us/call  inner=450    burst=   43.1 ms
    f83       105.43 us/call  inner=391    burst=   41.2 ms
    f84       101.83 us/call  inner=436    burst=   44.4 ms
    f85        75.62 us/call  inner=545    burst=   41.2 ms
    f86        68.36 us/call  inner=587    burst=   40.1 ms
    f89        70.62 us/call  inner=608    burst=   42.9 ms
    f94        65.47 us/call  inner=645    burst=   42.2 ms
    f95        59.39 us/call  inner=749    burst=   44.5 ms
    f98        57.19 us/call  inner=750    burst=   42.9 ms
    f99        54.56 us/call  inner=806    burst=   44.0 ms
    g101       38.34 us/call  inner=1184   burst=   45.4 ms
    g102       31.68 us/call  inner=1421   burst=   45.0 ms
    g24        73.16 us/call  inner=626    burst=   45.8 ms
    g27        68.23 us/call  inner=646    burst=   44.1 ms
    g83        79.19 us/call  inner=583    burst=   46.2 ms
    g84        74.73 us/call  inner=615    burst=   46.0 ms
    g85        49.72 us/call  inner=889    burst=   44.2 ms
    g86        41.76 us/call  inner=1077   burst=   45.0 ms
    g89        44.80 us/call  inner=1017   burst=   45.6 ms
    g94        39.55 us/call  inner=1126   burst=   44.5 ms
    g95        32.71 us/call  inner=1392   burst=   45.5 ms
    g98        34.61 us/call  inner=1296   burst=   44.9 ms
    g99        28.00 us/call  inner=1595   burst=   44.7 ms
    legB      119.17 us/call  inner=405    burst=   48.3 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s24  per-token                                         gemm/s27 1.0724x  chain/s27 1.0489x
      s84  per-token                                         gemm/s27 1.0953x  chain/s27 1.0630x
      s83  gs128      bank vs s84  1.0597x [1.0594-1.0709]  gemm/s27 1.1623x  chain/s27 1.0990x
      s86  per-token                                         gemm/s27 0.6120x  chain/s27 0.7128x
      s85  gs128      bank vs s86  1.1904x [1.1687-1.2799]  gemm/s27 0.7286x  chain/s27 0.7890x
      s89  gs256      bank vs s86  1.0712x [1.0556-1.0950]  gemm/s27 0.6579x  chain/s27 0.7368x
      s95  per-token                                         gemm/s27 0.4794x  chain/s27 0.6201x
      s94  gs128      bank vs s95  1.2088x [1.1269-1.2170]  gemm/s27 0.5798x  chain/s27 0.6833x
      s101 gs256      bank vs s95  1.1733x [1.0916-1.1740]  gemm/s27 0.5624x  chain/s27 0.6736x
      s99  per-token                                         gemm/s27 0.4104x  chain/s27 0.5694x
      s98  gs128      bank vs s99  1.2362x [1.2272-1.2403]  gemm/s27 0.5073x  chain/s27 0.5959x
      s102 gs256      bank vs s99  1.1324x [1.1228-1.1354]  gemm/s27 0.4648x  chain/s27 0.5623x
      BEST grouped s102 vs BEST per-token s99: gemm 1.1324x [1.1228-1.1354]  chain 0.9867x [0.9759-0.9937]

  out_proj k=6144 n=5120 M=64   burst>= 50 ms, reps=7
    f101       61.90 us/call  inner=729    burst=   45.1 ms
    f102       73.38 us/call  inner=640    burst=   47.0 ms
    f24       101.31 us/call  inner=482    burst=   48.8 ms
    f27        96.95 us/call  inner=503    burst=   48.8 ms
    f83       106.50 us/call  inner=441    burst=   47.0 ms
    f84       102.90 us/call  inner=475    burst=   48.9 ms
    f85        73.95 us/call  inner=619    burst=   45.8 ms
    f86        69.50 us/call  inner=690    burst=   48.0 ms
    f89        68.25 us/call  inner=673    burst=   45.9 ms
    f94        63.26 us/call  inner=723    burst=   45.7 ms
    f95        60.14 us/call  inner=787    burst=   47.3 ms
    f98        74.92 us/call  inner=618    burst=   46.3 ms
    f99        73.27 us/call  inner=646    burst=   47.3 ms
    g101       38.50 us/call  inner=1273   burst=   49.0 ms
    g102       49.91 us/call  inner=989    burst=   49.4 ms
    g24        73.36 us/call  inner=676    burst=   49.6 ms
    g27        68.61 us/call  inner=720    burst=   49.4 ms
    g83        79.31 us/call  inner=624    burst=   49.5 ms
    g84        74.95 us/call  inner=659    burst=   49.4 ms
    g85        50.22 us/call  inner=942    burst=   47.3 ms
    g86        42.11 us/call  inner=1170   burst=   49.3 ms
    g89        44.84 us/call  inner=1092   burst=   49.0 ms
    g94        39.51 us/call  inner=1210   burst=   47.8 ms
    g95        32.74 us/call  inner=1491   burst=   48.8 ms
    g98        51.43 us/call  inner=960    burst=   49.4 ms
    g99        45.76 us/call  inner=1077   burst=   49.3 ms
    legB      120.88 us/call  inner=417    burst=   50.4 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s24  per-token                                         gemm/s27 1.0692x  chain/s27 1.0449x
      s84  per-token                                         gemm/s27 1.0923x  chain/s27 1.0613x
      s83  gs128      bank vs s84  1.0582x [1.0000-1.0587]  gemm/s27 1.1559x  chain/s27 1.0971x
      s86  per-token                                         gemm/s27 0.6137x  chain/s27 0.7169x
      s85  gs128      bank vs s86  1.1929x [1.1925-1.2056]  gemm/s27 0.7320x  chain/s27 0.7627x
      s89  gs256      bank vs s86  1.0649x [1.0631-1.0936]  gemm/s27 0.6535x  chain/s27 0.7038x
      s95  per-token                                         gemm/s27 0.4770x  chain/s27 0.6200x
      s94  gs128      bank vs s95  1.2075x [1.2056-1.2601]  gemm/s27 0.5759x  chain/s27 0.6525x
      s101 gs256      bank vs s95  1.1768x [1.1725-1.1794]  gemm/s27 0.5612x  chain/s27 0.6375x
      s99  per-token                                         gemm/s27 0.6668x  chain/s27 0.7556x
      s98  gs128      bank vs s99  1.1237x [1.1230-1.1242]  gemm/s27 0.7494x  chain/s27 0.7723x
      s102 gs256      bank vs s99  1.0905x [1.0891-1.1417]  gemm/s27 0.7276x  chain/s27 0.7563x
      BEST grouped s101 vs BEST per-token s95: gemm 1.1768x [1.1725-1.1794]  chain 1.0286x [1.0244-1.0323]

  out_proj k=6144 n=5120 M=2048   burst>= 50 ms, reps=7
    f101      772.82 us/call  inner=68     burst=   52.6 ms
    f102     1160.40 us/call  inner=43     burst=   49.9 ms
    f24       507.99 us/call  inner=103    burst=   52.3 ms
    f27       480.01 us/call  inner=114    burst=   54.7 ms
    f83       640.39 us/call  inner=85     burst=   54.4 ms
    f84       531.35 us/call  inner=98     burst=   52.1 ms
    f85       613.12 us/call  inner=84     burst=   51.5 ms
    f86       521.23 us/call  inner=99     burst=   51.6 ms
    f89       570.47 us/call  inner=89     burst=   50.8 ms
    f94       833.78 us/call  inner=64     burst=   53.4 ms
    f95       718.02 us/call  inner=74     burst=   53.1 ms
    f98      1225.10 us/call  inner=41     burst=   50.2 ms
    f99      1092.19 us/call  inner=47     burst=   51.3 ms
    g101      614.26 us/call  inner=87     burst=   53.4 ms
    g102     1013.02 us/call  inner=50     burst=   50.7 ms
    g24       358.66 us/call  inner=148    burst=   53.1 ms
    g27       317.27 us/call  inner=170    burst=   53.9 ms
    g83       471.88 us/call  inner=117    burst=   55.2 ms
    g84       376.22 us/call  inner=136    burst=   51.2 ms
    g85       452.18 us/call  inner=120    burst=   54.3 ms
    g86       365.18 us/call  inner=140    burst=   51.1 ms
    g89       422.83 us/call  inner=126    burst=   53.3 ms
    g94       634.21 us/call  inner=82     burst=   52.0 ms
    g95       541.59 us/call  inner=94     burst=   50.9 ms
    g98      1038.95 us/call  inner=48     burst=   49.9 ms
    g99       955.83 us/call  inner=56     burst=   53.5 ms
    legB      644.37 us/call  inner=80     burst=   51.5 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s24  per-token                                         gemm/s27 1.1232x  chain/s27 1.0928x
      s84  per-token                                         gemm/s27 1.2115x  chain/s27 1.1319x
      s83  gs128      bank vs s84  1.2082x [1.1647-1.3083]  gemm/s27 1.4710x  chain/s27 1.3588x
      s86  per-token                                         gemm/s27 1.1473x  chain/s27 1.1118x
      s85  gs128      bank vs s86  1.2382x [1.1612-1.2763]  gemm/s27 1.4320x  chain/s27 1.2768x
      s89  gs256      bank vs s86  1.1611x [1.1238-1.1742]  gemm/s27 1.3327x  chain/s27 1.1935x
      s95  per-token                                         gemm/s27 1.7155x  chain/s27 1.4961x
      s94  gs128      bank vs s95  1.1721x [1.0468-1.2401]  gemm/s27 1.9980x  chain/s27 1.7539x
      s101 gs256      bank vs s95  1.1128x [1.0176-1.1860]  gemm/s27 1.9089x  chain/s27 1.6163x
      s99  per-token                                         gemm/s27 2.9569x  chain/s27 2.3087x
      s98  gs128      bank vs s99  1.0809x [1.0557-1.1634]  gemm/s27 3.3421x  chain/s27 2.5431x
      s102 gs256      bank vs s99  1.0512x [1.0198-1.0905]  gemm/s27 3.1624x  chain/s27 2.4543x
      BEST grouped s89 vs BEST per-token s27: gemm 1.3327x [1.2374-1.4112]  chain 1.1935x [1.1736-1.3055]

  out_proj k=6144 n=5120 M=8192   burst>= 50 ms, reps=7
    f101     3216.45 us/call  inner=15     burst=   48.2 ms
    f102     5003.82 us/call  inner=11     burst=   55.0 ms
    f24      2250.44 us/call  inner=23     burst=   51.8 ms
    f27      2070.53 us/call  inner=25     burst=   51.8 ms
    f83      2713.87 us/call  inner=19     burst=   51.6 ms
    f84      2332.95 us/call  inner=22     burst=   51.3 ms
    f85      2664.83 us/call  inner=19     burst=   50.6 ms
    f86      2308.31 us/call  inner=22     burst=   50.8 ms
    f89      2485.14 us/call  inner=20     burst=   49.7 ms
    f94      3433.06 us/call  inner=15     burst=   51.5 ms
    f95      3041.82 us/call  inner=17     burst=   51.7 ms
    f98      5098.97 us/call  inner=10     burst=   51.0 ms
    f99      4573.71 us/call  inner=11     burst=   50.3 ms
    g101     2418.32 us/call  inner=22     burst=   53.2 ms
    g102     4126.55 us/call  inner=13     burst=   53.6 ms
    g24      1393.53 us/call  inner=38     burst=   53.0 ms
    g27      1247.99 us/call  inner=42     burst=   52.4 ms
    g83      1889.86 us/call  inner=28     burst=   52.9 ms
    g84      1524.00 us/call  inner=35     burst=   53.3 ms
    g85      1749.84 us/call  inner=29     burst=   50.7 ms
    g86      1478.59 us/call  inner=36     burst=   53.2 ms
    g89      1632.38 us/call  inner=32     burst=   52.2 ms
    g94      2660.51 us/call  inner=20     burst=   53.2 ms
    g95      2231.55 us/call  inner=24     burst=   53.6 ms
    g98      4286.81 us/call  inner=12     burst=   51.4 ms
    g99      3847.31 us/call  inner=14     burst=   53.9 ms
    legB     2202.11 us/call  inner=24     burst=   52.9 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s24  per-token                                         gemm/s27 1.1005x  chain/s27 1.0863x
      s84  per-token                                         gemm/s27 1.2112x  chain/s27 1.1406x
      s83  gs128      bank vs s84  1.1856x [1.1528-1.2848]  gemm/s27 1.4544x  chain/s27 1.3207x
      s86  per-token                                         gemm/s27 1.1810x  chain/s27 1.1139x
      s85  gs128      bank vs s86  1.1816x [1.1460-1.3467]  gemm/s27 1.4021x  chain/s27 1.2996x
      s89  gs256      bank vs s86  1.1040x [1.0478-1.2148]  gemm/s27 1.3220x  chain/s27 1.2002x
      s95  per-token                                         gemm/s27 1.7881x  chain/s27 1.4716x
      s94  gs128      bank vs s95  1.1571x [1.1249-1.3173]  gemm/s27 2.0872x  chain/s27 1.6581x
      s101 gs256      bank vs s95  1.0813x [1.0609-1.1325]  gemm/s27 1.9378x  chain/s27 1.5799x
      s99  per-token                                         gemm/s27 3.0953x  chain/s27 2.2156x
      s98  gs128      bank vs s99  1.1170x [0.9897-1.1566]  gemm/s27 3.3100x  chain/s27 2.4811x
      s102 gs256      bank vs s99  1.0647x [0.9803-1.2051]  gemm/s27 3.2708x  chain/s27 2.4043x
      BEST grouped s89 vs BEST per-token s27: gemm 1.3220x [1.2824-1.4347]  chain 1.2002x [1.1843-1.2371]

  gate_proj k=5120 n=17408 M=48   burst>= 50 ms, reps=7
    f101      100.39 us/call  inner=470    burst=   47.2 ms
    f102      109.51 us/call  inner=443    burst=   48.5 ms
    f24       136.31 us/call  inner=357    burst=   48.7 ms
    f27       125.89 us/call  inner=382    burst=   48.1 ms
    f83       162.41 us/call  inner=294    burst=   47.7 ms
    f84       143.79 us/call  inner=332    burst=   47.7 ms
    f85        94.86 us/call  inner=491    burst=   46.6 ms
    f86        90.78 us/call  inner=551    burst=   50.0 ms
    f89        93.36 us/call  inner=504    burst=   47.1 ms
    f94       106.66 us/call  inner=442    burst=   47.1 ms
    f95        93.33 us/call  inner=517    burst=   48.3 ms
    f98       111.79 us/call  inner=434    burst=   48.5 ms
    f99       104.98 us/call  inner=464    burst=   48.7 ms
    g101       76.70 us/call  inner=655    burst=   50.2 ms
    g102       87.31 us/call  inner=575    burst=   50.2 ms
    g24       106.40 us/call  inner=468    burst=   49.8 ms
    g27        95.30 us/call  inner=521    burst=   49.6 ms
    g83       134.60 us/call  inner=367    burst=   49.4 ms
    g84       113.69 us/call  inner=434    burst=   49.3 ms
    g85        68.78 us/call  inner=724    burst=   49.8 ms
    g86        59.00 us/call  inner=846    burst=   49.9 ms
    g89        65.62 us/call  inner=769    burst=   50.5 ms
    g94        82.13 us/call  inner=590    burst=   48.5 ms
    g95        70.15 us/call  inner=702    burst=   49.2 ms
    g98        91.29 us/call  inner=549    burst=   50.1 ms
    g99        82.14 us/call  inner=607    burst=   49.9 ms
    legB      485.53 us/call  inner=104    burst=   50.5 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s24  per-token                                         gemm/s27 1.1165x  chain/s27 1.0736x
      s84  per-token                                         gemm/s27 1.1929x  chain/s27 1.1399x
      s83  gs128      bank vs s84  1.1847x [1.1463-1.2005]  gemm/s27 1.4121x  chain/s27 1.3030x
      s86  per-token                                         gemm/s27 0.6191x  chain/s27 0.7041x
      s85  gs128      bank vs s86  1.1711x [1.1306-1.2274]  gemm/s27 0.7156x  chain/s27 0.7555x
      s89  gs256      bank vs s86  1.1061x [1.0797-1.1907]  gemm/s27 0.6885x  chain/s27 0.7388x
      s95  per-token                                         gemm/s27 0.7361x  chain/s27 0.7492x
      s94  gs128      bank vs s95  1.1665x [1.1484-1.1973]  gemm/s27 0.8608x  chain/s27 0.8472x
      s101 gs256      bank vs s95  1.0877x [1.0748-1.1239]  gemm/s27 0.8016x  chain/s27 0.7974x
      s99  per-token                                         gemm/s27 0.8583x  chain/s27 0.8339x
      s98  gs128      bank vs s99  1.1091x [1.0892-1.1364]  gemm/s27 0.9519x  chain/s27 0.8903x
      s102 gs256      bank vs s99  1.0652x [1.0484-1.0694]  gemm/s27 0.9154x  chain/s27 0.8687x
      BEST grouped s89 vs BEST per-token s86: gemm 1.1061x [1.0797-1.1907]  chain 1.0621x [1.0089-1.0895]

  gate_proj k=5120 n=17408 M=64   burst>= 50 ms, reps=7
    f101      100.41 us/call  inner=480    burst=   48.2 ms
    f102      130.65 us/call  inner=372    burst=   48.6 ms
    f24       138.63 us/call  inner=348    burst=   48.2 ms
    f27       127.36 us/call  inner=378    burst=   48.1 ms
    f83       166.97 us/call  inner=282    burst=   47.1 ms
    f84       151.21 us/call  inner=333    burst=   50.4 ms
    f85        96.49 us/call  inner=508    burst=   49.0 ms
    f86        91.30 us/call  inner=533    burst=   48.7 ms
    f89        90.79 us/call  inner=530    burst=   48.1 ms
    f94       106.50 us/call  inner=450    burst=   47.9 ms
    f95        95.56 us/call  inner=508    burst=   48.5 ms
    f98       134.53 us/call  inner=362    burst=   48.7 ms
    f99       126.11 us/call  inner=392    burst=   49.4 ms
    g101       77.58 us/call  inner=644    burst=   50.0 ms
    g102      109.63 us/call  inner=451    burst=   49.4 ms
    g24       107.57 us/call  inner=452    burst=   48.6 ms
    g27        95.88 us/call  inner=506    burst=   48.5 ms
    g83       135.15 us/call  inner=356    burst=   48.1 ms
    g84       114.53 us/call  inner=425    burst=   48.7 ms
    g85        70.86 us/call  inner=709    burst=   50.2 ms
    g86        59.44 us/call  inner=824    burst=   49.0 ms
    g89        64.34 us/call  inner=777    burst=   50.0 ms
    g94        83.66 us/call  inner=587    burst=   49.1 ms
    g95        72.32 us/call  inner=699    burst=   50.5 ms
    g98       113.00 us/call  inner=436    burst=   49.3 ms
    g99       102.20 us/call  inner=486    burst=   49.7 ms
    legB      484.67 us/call  inner=104    burst=   50.4 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s24  per-token                                         gemm/s27 1.1201x  chain/s27 1.0922x
      s84  per-token                                         gemm/s27 1.1940x  chain/s27 1.1721x
      s83  gs128      bank vs s84  1.1797x [1.0732-1.2228]  gemm/s27 1.4098x  chain/s27 1.3157x
      s86  per-token                                         gemm/s27 0.6198x  chain/s27 0.7142x
      s85  gs128      bank vs s86  1.1969x [1.1119-1.2486]  gemm/s27 0.7392x  chain/s27 0.7567x
      s89  gs256      bank vs s86  1.0824x [1.0439-1.1294]  gemm/s27 0.6694x  chain/s27 0.7138x
      s95  per-token                                         gemm/s27 0.7472x  chain/s27 0.7604x
      s94  gs128      bank vs s95  1.1549x [1.1398-1.1679]  gemm/s27 0.8671x  chain/s27 0.8435x
      s101 gs256      bank vs s95  1.0729x [1.0540-1.1251]  gemm/s27 0.8040x  chain/s27 0.7923x
      s99  per-token                                         gemm/s27 1.0641x  chain/s27 0.9947x
      s98  gs128      bank vs s99  1.1077x [1.0979-1.1286]  gemm/s27 1.1787x  chain/s27 1.0605x
      s102 gs256      bank vs s99  1.0741x [1.0585-1.0886]  gemm/s27 1.1426x  chain/s27 1.0297x
      BEST grouped s89 vs BEST per-token s86: gemm 1.0824x [1.0439-1.1294]  chain 1.0117x [0.9329-1.0803]

  gate_proj k=5120 n=17408 M=2048   burst>= 50 ms, reps=7
    f101     2049.54 us/call  inner=26     burst=   53.3 ms
    f102     3197.36 us/call  inner=16     burst=   51.2 ms
    f24      1283.10 us/call  inner=42     burst=   53.9 ms
    f27      1135.17 us/call  inner=46     burst=   52.2 ms
    f83      1625.99 us/call  inner=31     burst=   50.4 ms
    f84      1356.26 us/call  inner=38     burst=   51.5 ms
    f85      1544.44 us/call  inner=33     burst=   51.0 ms
    f86      1299.37 us/call  inner=40     burst=   52.0 ms
    f89      1456.92 us/call  inner=36     burst=   52.4 ms
    f94      2177.66 us/call  inner=24     burst=   52.3 ms
    f95      1813.17 us/call  inner=28     burst=   50.8 ms
    f98      3340.48 us/call  inner=16     burst=   53.4 ms
    f99      2988.22 us/call  inner=18     burst=   53.8 ms
    g101     1764.86 us/call  inner=30     burst=   52.9 ms
    g102     2950.20 us/call  inner=18     burst=   53.1 ms
    g24      1029.67 us/call  inner=52     burst=   53.5 ms
    g27       909.31 us/call  inner=57     burst=   51.8 ms
    g83      1368.06 us/call  inner=37     burst=   50.6 ms
    g84      1102.15 us/call  inner=47     burst=   51.8 ms
    g85      1266.46 us/call  inner=41     burst=   51.9 ms
    g86      1065.54 us/call  inner=49     burst=   52.2 ms
    g89      1208.99 us/call  inner=43     burst=   52.0 ms
    g94      1879.12 us/call  inner=27     burst=   50.7 ms
    g95      1567.80 us/call  inner=33     burst=   51.7 ms
    g98      3063.23 us/call  inner=17     burst=   52.1 ms
    g99      2799.94 us/call  inner=19     burst=   53.2 ms
    legB     1797.87 us/call  inner=29     burst=   52.1 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s24  per-token                                         gemm/s27 1.1284x  chain/s27 1.1315x
      s84  per-token                                         gemm/s27 1.2444x  chain/s27 1.1776x
      s83  gs128      bank vs s84  1.2374x [1.1647-1.3427]  gemm/s27 1.5236x  chain/s27 1.4391x
      s86  per-token                                         gemm/s27 1.1827x  chain/s27 1.1377x
      s85  gs128      bank vs s86  1.2049x [1.1154-1.2397]  gemm/s27 1.4022x  chain/s27 1.3600x
      s89  gs256      bank vs s86  1.1310x [1.0822-1.1854]  gemm/s27 1.3296x  chain/s27 1.2873x
      s95  per-token                                         gemm/s27 1.7548x  chain/s27 1.5883x
      s94  gs128      bank vs s95  1.1758x [1.1160-1.2377]  gemm/s27 2.0999x  chain/s27 1.8860x
      s101 gs256      bank vs s95  1.1176x [0.9991-1.1648]  gemm/s27 1.9421x  chain/s27 1.7605x
      s99  per-token                                         gemm/s27 3.1035x  chain/s27 2.6440x
      s98  gs128      bank vs s99  1.1158x [1.0257-1.1609]  gemm/s27 3.3906x  chain/s27 2.9668x
      s102 gs256      bank vs s99  1.0853x [0.9862-1.1246]  gemm/s27 3.2575x  chain/s27 2.7998x
      BEST grouped s89 vs BEST per-token s27: gemm 1.3296x [1.2744-1.4248]  chain 1.2873x [1.2434-1.3523]

  gate_proj k=5120 n=17408 M=8192   burst>= 50 ms, reps=7
    f101     8159.74 us/call  inner=8      burst=   65.3 ms
    f102    12806.78 us/call  inner=8      burst=  102.5 ms
    f24      5194.55 us/call  inner=10     burst=   51.9 ms
    f27      4851.22 us/call  inner=11     burst=   53.4 ms
    f83      6611.30 us/call  inner=8      burst=   52.9 ms
    f84      5566.73 us/call  inner=9      burst=   50.1 ms
    f85      6264.98 us/call  inner=9      burst=   56.4 ms
    f86      5290.91 us/call  inner=10     burst=   52.9 ms
    f89      5926.12 us/call  inner=9      burst=   53.3 ms
    f94      8671.45 us/call  inner=8      burst=   69.4 ms
    f95      7345.79 us/call  inner=8      burst=   58.8 ms
    f98     13376.13 us/call  inner=8      burst=  107.0 ms
    f99     12097.15 us/call  inner=8      burst=   96.8 ms
    g101     6940.80 us/call  inner=8      burst=   55.5 ms
    g102    11879.12 us/call  inner=8      burst=   95.0 ms
    g24      4129.37 us/call  inner=13     burst=   53.7 ms
    g27      3602.15 us/call  inner=15     burst=   54.0 ms
    g83      5419.44 us/call  inner=10     burst=   54.2 ms
    g84      4424.94 us/call  inner=12     burst=   53.1 ms
    g85      5133.31 us/call  inner=10     burst=   51.3 ms
    g86      4224.34 us/call  inner=12     burst=   50.7 ms
    g89      4824.61 us/call  inner=11     burst=   53.1 ms
    g94      7435.26 us/call  inner=8      burst=   59.5 ms
    g95      6366.56 us/call  inner=8      burst=   50.9 ms
    g98     12439.48 us/call  inner=8      burst=   99.5 ms
    g99     10979.20 us/call  inner=8      burst=   87.8 ms
    legB     6362.75 us/call  inner=8      burst=   50.9 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s24  per-token                                         gemm/s27 1.1680x  chain/s27 1.0708x
      s84  per-token                                         gemm/s27 1.2320x  chain/s27 1.1484x
      s83  gs128      bank vs s84  1.2212x [1.1707-1.3989]  gemm/s27 1.5197x  chain/s27 1.3714x
      s86  per-token                                         gemm/s27 1.1796x  chain/s27 1.0946x
      s85  gs128      bank vs s86  1.1926x [1.1646-1.4522]  gemm/s27 1.4251x  chain/s27 1.3176x
      s89  gs256      bank vs s86  1.1584x [1.1023-1.1764]  gemm/s27 1.3342x  chain/s27 1.2658x
      s95  per-token                                         gemm/s27 1.7765x  chain/s27 1.5142x
      s94  gs128      bank vs s95  1.1818x [1.1004-1.3174]  gemm/s27 2.1000x  chain/s27 1.8644x
      s101 gs256      bank vs s95  1.1039x [1.0397-1.1757]  gemm/s27 1.9411x  chain/s27 1.7264x
      s99  per-token                                         gemm/s27 3.0117x  chain/s27 2.5087x
      s98  gs128      bank vs s99  1.1231x [1.1003-1.1588]  gemm/s27 3.4112x  chain/s27 2.7396x
      s102 gs256      bank vs s99  1.0820x [1.0197-1.1011]  gemm/s27 3.2821x  chain/s27 2.6399x
      BEST grouped s89 vs BEST per-token s27: gemm 1.3342x [1.2208-1.4215]  chain 1.2658x [1.1065-1.2788]

  down_proj k=17408 n=5120 M=48   burst>= 50 ms, reps=7
    f101      135.95 us/call  inner=356    burst=   48.4 ms
    f102      113.42 us/call  inner=426    burst=   48.3 ms
    f24       230.16 us/call  inner=216    burst=   49.7 ms
    f27       215.72 us/call  inner=229    burst=   49.4 ms
    f83       253.20 us/call  inner=194    burst=   49.1 ms
    f84       233.73 us/call  inner=212    burst=   49.6 ms
    f85       171.40 us/call  inner=290    burst=   49.7 ms
    f86       139.35 us/call  inner=352    burst=   49.1 ms
    f89       154.43 us/call  inner=314    burst=   48.5 ms
    f94       139.58 us/call  inner=348    burst=   48.6 ms
    f95       114.16 us/call  inner=429    burst=   49.0 ms
    f98       122.80 us/call  inner=392    burst=   48.1 ms
    f99       103.44 us/call  inner=474    burst=   49.0 ms
    g101      105.56 us/call  inner=469    burst=   49.5 ms
    g102       88.39 us/call  inner=555    burst=   49.1 ms
    g24       202.95 us/call  inner=247    burst=   50.1 ms
    g27       188.52 us/call  inner=265    burst=   50.0 ms
    g83       220.69 us/call  inner=227    burst=   50.1 ms
    g84       206.75 us/call  inner=242    burst=   50.0 ms
    g85       137.42 us/call  inner=354    burst=   48.6 ms
    g86       113.82 us/call  inner=438    burst=   49.9 ms
    g89       123.61 us/call  inner=400    burst=   49.4 ms
    g94       109.93 us/call  inner=446    burst=   49.0 ms
    g95        88.96 us/call  inner=560    burst=   49.8 ms
    g98        96.52 us/call  inner=506    burst=   48.8 ms
    g99        81.23 us/call  inner=614    burst=   49.9 ms
    legB      489.58 us/call  inner=103    burst=   50.4 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s24  per-token                                         gemm/s27 1.0765x  chain/s27 1.0667x
      s84  per-token                                         gemm/s27 1.0967x  chain/s27 1.0823x
      s83  gs128      bank vs s84  1.0664x [1.0655-1.0943]  gemm/s27 1.1705x  chain/s27 1.1714x
      s86  per-token                                         gemm/s27 0.6037x  chain/s27 0.6456x
      s85  gs128      bank vs s86  1.2073x [1.1656-1.2814]  gemm/s27 0.7283x  chain/s27 0.7815x
      s89  gs256      bank vs s86  1.0853x [1.0730-1.1019]  gemm/s27 0.6552x  chain/s27 0.7160x
      s95  per-token                                         gemm/s27 0.4719x  chain/s27 0.5298x
      s94  gs128      bank vs s95  1.2212x [1.1576-1.2470]  gemm/s27 0.5762x  chain/s27 0.6446x
      s101 gs256      bank vs s95  1.1866x [1.1246-1.2633]  gemm/s27 0.5593x  chain/s27 0.6299x
      s99  per-token                                         gemm/s27 0.4291x  chain/s27 0.4796x
      s98  gs128      bank vs s99  1.1916x [1.1681-1.1932]  gemm/s27 0.5111x  chain/s27 0.5688x
      s102 gs256      bank vs s99  1.0896x [1.0797-1.1172]  gemm/s27 0.4676x  chain/s27 0.5235x
      BEST grouped s102 vs BEST per-token s99: gemm 1.0896x [1.0797-1.1172]  chain 1.0995x [1.0680-1.2264]

  down_proj k=17408 n=5120 M=64   burst>= 50 ms, reps=7
    f101      135.60 us/call  inner=359    burst=   48.7 ms
    f102      169.44 us/call  inner=284    burst=   48.1 ms
    f24       231.41 us/call  inner=214    burst=   49.5 ms
    f27       217.02 us/call  inner=227    burst=   49.3 ms
    f83       255.20 us/call  inner=192    burst=   49.0 ms
    f84       234.95 us/call  inner=211    burst=   49.6 ms
    f85       169.08 us/call  inner=289    burst=   48.9 ms
    f86       141.22 us/call  inner=349    burst=   49.3 ms
    f89       153.70 us/call  inner=316    burst=   48.6 ms
    f94       138.31 us/call  inner=345    burst=   47.7 ms
    f95       116.34 us/call  inner=422    burst=   49.1 ms
    f98       173.43 us/call  inner=283    burst=   49.1 ms
    f99       154.51 us/call  inner=315    burst=   48.7 ms
    g101      105.41 us/call  inner=468    burst=   49.3 ms
    g102      140.00 us/call  inner=358    burst=   50.1 ms
    g24       203.19 us/call  inner=244    burst=   49.6 ms
    g27       189.01 us/call  inner=264    burst=   49.9 ms
    g83       220.62 us/call  inner=226    burst=   49.9 ms
    g84       207.06 us/call  inner=241    burst=   49.9 ms
    g85       139.02 us/call  inner=342    burst=   47.5 ms
    g86       114.10 us/call  inner=436    burst=   49.7 ms
    g89       123.76 us/call  inner=400    burst=   49.5 ms
    g94       108.64 us/call  inner=444    burst=   48.2 ms
    g95        89.07 us/call  inner=560    burst=   49.9 ms
    g98       144.29 us/call  inner=347    burst=   50.1 ms
    g99       129.00 us/call  inner=389    burst=   50.2 ms
    legB      490.01 us/call  inner=103    burst=   50.5 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s24  per-token                                         gemm/s27 1.0755x  chain/s27 1.0653x
      s84  per-token                                         gemm/s27 1.0962x  chain/s27 1.0819x
      s83  gs128      bank vs s84  1.0655x [1.0578-1.0671]  gemm/s27 1.1679x  chain/s27 1.1719x
      s86  per-token                                         gemm/s27 0.6043x  chain/s27 0.6500x
      s85  gs128      bank vs s86  1.2182x [1.2093-1.2212]  gemm/s27 0.7359x  chain/s27 0.7783x
      s89  gs256      bank vs s86  1.0840x [1.0754-1.1070]  gemm/s27 0.6549x  chain/s27 0.7083x
      s95  per-token                                         gemm/s27 0.4713x  chain/s27 0.5358x
      s94  gs128      bank vs s95  1.2198x [1.1931-1.2440]  gemm/s27 0.5748x  chain/s27 0.6359x
      s101 gs256      bank vs s95  1.1857x [1.1575-1.1960]  gemm/s27 0.5582x  chain/s27 0.6230x
      s99  per-token                                         gemm/s27 0.6824x  chain/s27 0.7120x
      s98  gs128      bank vs s99  1.1205x [1.1067-1.1265]  gemm/s27 0.7633x  chain/s27 0.7981x
      s102 gs256      bank vs s99  1.0858x [1.0760-1.0875]  gemm/s27 0.7409x  chain/s27 0.7800x
      BEST grouped s101 vs BEST per-token s95: gemm 1.1857x [1.1575-1.1960]  chain 1.1600x [1.1537-1.1745]

  down_proj k=17408 n=5120 M=2048   burst>= 50 ms, reps=7
    f101     2226.49 us/call  inner=23     burst=   51.2 ms
    f102     3421.52 us/call  inner=15     burst=   51.3 ms
    f24      1515.61 us/call  inner=34     burst=   51.5 ms
    f27      1399.32 us/call  inner=36     burst=   50.4 ms
    f83      1847.04 us/call  inner=28     burst=   51.7 ms
    f84      1609.32 us/call  inner=33     burst=   53.1 ms
    f85      1776.38 us/call  inner=29     burst=   51.5 ms
    f86      1557.81 us/call  inner=33     burst=   51.4 ms
    f89      1663.01 us/call  inner=31     burst=   51.6 ms
    f94      2394.81 us/call  inner=22     burst=   52.7 ms
    f95      2045.71 us/call  inner=25     burst=   51.1 ms
    f98      3553.05 us/call  inner=15     burst=   53.3 ms
    f99      3194.45 us/call  inner=17     burst=   54.3 ms
    g101     1701.89 us/call  inner=31     burst=   52.8 ms
    g102     2940.25 us/call  inner=18     burst=   52.9 ms
    g24      1009.17 us/call  inner=52     burst=   52.5 ms
    g27       895.50 us/call  inner=59     burst=   52.8 ms
    g83      1296.25 us/call  inner=39     burst=   50.6 ms
    g84      1070.08 us/call  inner=49     burst=   52.4 ms
    g85      1262.34 us/call  inner=40     burst=   50.5 ms
    g86      1064.96 us/call  inner=50     burst=   53.2 ms
    g89      1205.11 us/call  inner=44     burst=   53.0 ms
    g94      1851.68 us/call  inner=28     burst=   51.8 ms
    g95      1583.54 us/call  inner=33     burst=   52.3 ms
    g98      3034.85 us/call  inner=18     burst=   54.6 ms
    g99      2689.43 us/call  inner=20     burst=   53.8 ms
    legB     1685.54 us/call  inner=31     burst=   52.3 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s24  per-token                                         gemm/s27 1.1402x  chain/s27 1.0795x
      s84  per-token                                         gemm/s27 1.1891x  chain/s27 1.1466x
      s83  gs128      bank vs s84  1.2013x [1.1790-1.2591]  gemm/s27 1.4793x  chain/s27 1.3200x
      s86  per-token                                         gemm/s27 1.1817x  chain/s27 1.1184x
      s85  gs128      bank vs s86  1.1976x [1.1416-1.3431]  gemm/s27 1.4552x  chain/s27 1.2722x
      s89  gs256      bank vs s86  1.1497x [1.0943-1.2189]  gemm/s27 1.3375x  chain/s27 1.1884x
      s95  per-token                                         gemm/s27 1.8109x  chain/s27 1.4592x
      s94  gs128      bank vs s95  1.1566x [1.1203-1.2455]  gemm/s27 2.0457x  chain/s27 1.6967x
      s101 gs256      bank vs s95  1.0755x [1.0209-1.1433]  gemm/s27 1.9187x  chain/s27 1.5851x
      s99  per-token                                         gemm/s27 3.0285x  chain/s27 2.3010x
      s98  gs128      bank vs s99  1.1185x [1.0945-1.2055]  gemm/s27 3.3873x  chain/s27 2.4941x
      s102 gs256      bank vs s99  1.1116x [1.0689-1.1598]  gemm/s27 3.3933x  chain/s27 2.4250x
      BEST grouped s89 vs BEST per-token s27: gemm 1.3375x [1.2660-1.4839]  chain 1.1884x [1.1475-1.2375]

  down_proj k=17408 n=5120 M=8192   burst>= 50 ms, reps=7
    f101     9055.36 us/call  inner=8      burst=   72.4 ms
    f102    13706.37 us/call  inner=8      burst=  109.7 ms
    f24      6125.57 us/call  inner=9      burst=   55.1 ms
    f27      5575.69 us/call  inner=10     burst=   55.8 ms
    f83      7434.24 us/call  inner=8      burst=   59.5 ms
    f84      6488.32 us/call  inner=8      burst=   51.9 ms
    f85      7133.26 us/call  inner=8      burst=   57.1 ms
    f86      6288.00 us/call  inner=8      burst=   50.3 ms
    f89      6866.20 us/call  inner=8      burst=   54.9 ms
    f94      9408.51 us/call  inner=8      burst=   75.3 ms
    f95      8319.21 us/call  inner=8      burst=   66.6 ms
    f98     14151.76 us/call  inner=8      burst=  113.2 ms
    f99     12876.03 us/call  inner=8      burst=  103.0 ms
    g101     6823.42 us/call  inner=8      burst=   54.6 ms
    g102    11704.58 us/call  inner=8      burst=   93.6 ms
    g24      4024.18 us/call  inner=13     burst=   52.3 ms
    g27      3468.15 us/call  inner=15     burst=   52.0 ms
    g83      5214.94 us/call  inner=10     burst=   52.1 ms
    g84      4262.57 us/call  inner=12     burst=   51.2 ms
    g85      4931.58 us/call  inner=10     burst=   49.3 ms
    g86      4160.51 us/call  inner=13     burst=   54.1 ms
    g89      4774.35 us/call  inner=11     burst=   52.5 ms
    g94      7423.44 us/call  inner=8      burst=   59.4 ms
    g95      6325.25 us/call  inner=8      burst=   50.6 ms
    g98     12375.81 us/call  inner=8      burst=   99.0 ms
    g99     11060.22 us/call  inner=8      burst=   88.5 ms
    legB     5535.40 us/call  inner=9      burst=   49.8 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s24  per-token                                         gemm/s27 1.1514x  chain/s27 1.0874x
      s84  per-token                                         gemm/s27 1.2230x  chain/s27 1.1655x
      s83  gs128      bank vs s84  1.2132x [1.1838-1.2921]  gemm/s27 1.5043x  chain/s27 1.3157x
      s86  per-token                                         gemm/s27 1.2029x  chain/s27 1.1295x
      s85  gs128      bank vs s86  1.1774x [1.1280-1.2082]  gemm/s27 1.4298x  chain/s27 1.2760x
      s89  gs256      bank vs s86  1.1349x [1.1131-1.1669]  gemm/s27 1.3847x  chain/s27 1.2275x
      s95  per-token                                         gemm/s27 1.7823x  chain/s27 1.4862x
      s94  gs128      bank vs s95  1.1742x [1.1389-1.2169]  gemm/s27 2.1199x  chain/s27 1.6874x
      s101 gs256      bank vs s95  1.0813x [1.0565-1.1278]  gemm/s27 1.9675x  chain/s27 1.6241x
      s99  per-token                                         gemm/s27 3.1753x  chain/s27 2.2777x
      s98  gs128      bank vs s99  1.1171x [1.0448-1.1713]  gemm/s27 3.5558x  chain/s27 2.5263x
      s102 gs256      bank vs s99  1.0627x [1.0092-1.0922]  gemm/s27 3.3246x  chain/s27 2.4398x
      BEST grouped s89 vs BEST per-token s27: gemm 1.3847x [1.3141-1.4356]  chain 1.2275x [1.2050-1.2831]

[sampler] 2561 samples over 128.3 s of process life; 120.0 s of that was inside a timed block

[under load] sm clock median 2400 MHz (min 1890, max 2715)  power median 445 W  sw_power_cap 1644/2394 samples  hw_slowdown 0/2394
[tenancy after ] 2218273, 10154 MiB   (this process is pid 59)
