[tenancy before] (none)
[l2 scrub] mode='read' -- see the module docstring for the bias it carries

  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
  s80    128x128x128 2x4 st2 sw8 gs128   256   222   49152      1        8 64x2
  s110   128x128x128 1x4 st2 sw8         128   254   49152      2        8 128   <-- SPILLS
  s111   128x128x128 1x4 st2 sw8 gs128   128   255   49152      2        8 128x2   <-- SPILLS

  out_proj k=6144 n=5120 M=2048   burst>= 50 ms, reps=7
    f110      470.55 us/call  inner=113    burst=   53.2 ms
    f111      511.01 us/call  inner=95     burst=   48.5 ms
    f27       448.44 us/call  inner=121    burst=   54.3 ms
    f80       627.52 us/call  inner=83     burst=   52.1 ms
    g110      320.00 us/call  inner=172    burst=   55.0 ms
    g111      360.84 us/call  inner=132    burst=   47.6 ms
    g27       304.92 us/call  inner=180    burst=   54.9 ms
    g80       450.20 us/call  inner=113    burst=   50.9 ms
    legA     1519.75 us/call  inner=17     burst=   25.8 ms
    legB      582.01 us/call  inner=43     burst=   25.0 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s110 per-token                                         gemm/s27 1.0808x  chain/s27 1.0493x
      s111 gs128      bank vs s110 1.1455x [0.8659-1.2075]  gemm/s27 1.2337x  chain/s27 1.1785x
      s80  gs128                                             gemm/s27 1.4780x  chain/s27 1.4175x
      BEST grouped s111 vs BEST per-token s27: gemm 1.2337x [0.9675-1.5406]  chain 1.1785x [1.1045-1.9721]

  out_proj k=6144 n=5120 M=4096   burst>= 50 ms, reps=7
    f110      980.91 us/call  inner=52     burst=   51.0 ms
    f111     1127.96 us/call  inner=46     burst=   51.9 ms
    f27       957.54 us/call  inner=54     burst=   51.7 ms
    f80      1312.32 us/call  inner=36     burst=   47.2 ms
    g110      610.03 us/call  inner=87     burst=   53.1 ms
    g111      782.77 us/call  inner=70     burst=   54.8 ms
    g27       566.02 us/call  inner=92     burst=   52.1 ms
    g80       927.98 us/call  inner=56     burst=   52.0 ms
    legA     3010.32 us/call  inner=17     burst=   51.2 ms
    legB     1094.63 us/call  inner=48     burst=   52.5 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s110 per-token                                         gemm/s27 1.0684x  chain/s27 1.0265x
      s111 gs128      bank vs s110 1.1691x [1.0295-2.5641]  gemm/s27 1.2656x  chain/s27 1.1658x
      s80  gs128                                             gemm/s27 1.5735x  chain/s27 1.3667x
      BEST grouped s111 vs BEST per-token s27: gemm 1.2656x [1.2294-2.6350]  chain 1.1658x [0.8784-1.2558]

  out_proj k=6144 n=5120 M=8192   burst>= 50 ms, reps=7
    f110     2192.83 us/call  inner=25     burst=   54.8 ms
    f111     2630.75 us/call  inner=11     burst=   28.9 ms
    f27      2043.12 us/call  inner=26     burst=   53.1 ms
    f80      3155.58 us/call  inner=10     burst=   31.6 ms
    g110     1996.84 us/call  inner=45     burst=   89.9 ms
    g111     1506.32 us/call  inner=19     burst=   28.6 ms
    g27      1107.83 us/call  inner=23     burst=   25.5 ms
    g80      2092.78 us/call  inner=15     burst=   31.4 ms
    legA     5991.80 us/call  inner=8      burst=   47.9 ms
    legB     2038.94 us/call  inner=13     burst=   26.5 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s110 per-token                                         gemm/s27 1.3002x  chain/s27 1.0342x
      s111 gs128      bank vs s110 1.1204x [0.6575-1.6268]  gemm/s27 1.2887x  chain/s27 1.1728x
      s80  gs128                                             gemm/s27 1.5844x  chain/s27 1.3334x
      BEST grouped s111 vs BEST per-token s27: gemm 1.2887x [1.0805-2.1152]  chain 1.1728x [1.0388-1.5278]

  gate_proj k=5120 n=17408 M=2048   burst>= 50 ms, reps=7
    f110     1059.70 us/call  inner=45     burst=   47.7 ms
    f111     1215.60 us/call  inner=39     burst=   47.4 ms
    f27      1016.14 us/call  inner=50     burst=   50.8 ms
    f80      1587.85 us/call  inner=19     burst=   30.2 ms
    g110      856.21 us/call  inner=57     burst=   48.8 ms
    g111      984.99 us/call  inner=48     burst=   47.3 ms
    g27       823.96 us/call  inner=65     burst=   53.6 ms
    g80      1403.03 us/call  inner=27     burst=   37.9 ms
    legA     3548.06 us/call  inner=16     burst=   56.8 ms
    legB     1652.46 us/call  inner=31     burst=   51.2 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s110 per-token                                         gemm/s27 1.0331x  chain/s27 1.0452x
      s111 gs128      bank vs s110 1.1663x [1.1183-1.1913]  gemm/s27 1.2067x  chain/s27 1.1860x
      s80  gs128                                             gemm/s27 1.6089x  chain/s27 1.5691x
      BEST grouped s111 vs BEST per-token s27: gemm 1.2067x [1.0118-1.2691]  chain 1.1860x [1.1428-1.2756]

  gate_proj k=5120 n=17408 M=4096   burst>= 50 ms, reps=7
    f110     2206.44 us/call  inner=23     burst=   50.7 ms
    f111     2463.33 us/call  inner=20     burst=   49.3 ms
    f27      2149.93 us/call  inner=24     burst=   51.6 ms
    f80      3295.80 us/call  inner=16     burst=   52.7 ms
    g110     1687.75 us/call  inner=31     burst=   52.3 ms
    g111     1906.53 us/call  inner=26     burst=   49.6 ms
    g27      1611.50 us/call  inner=33     burst=   53.2 ms
    g80      2637.77 us/call  inner=20     burst=   52.8 ms
    legA     6842.65 us/call  inner=8      burst=   54.7 ms
    legB     3028.99 us/call  inner=17     burst=   51.5 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s110 per-token                                         gemm/s27 1.0295x  chain/s27 1.0329x
      s111 gs128      bank vs s110 1.1484x [0.9752-1.2161]  gemm/s27 1.1829x  chain/s27 1.1520x
      s80  gs128                                             gemm/s27 1.6134x  chain/s27 1.5186x
      BEST grouped s111 vs BEST per-token s27: gemm 1.1829x [1.0219-1.3061]  chain 1.1520x [0.9905-1.2041]

  gate_proj k=5120 n=17408 M=8192   burst>= 50 ms, reps=7
    f110     4580.82 us/call  inner=11     burst=   50.4 ms
    f111     4967.43 us/call  inner=10     burst=   49.7 ms
    f27      4313.93 us/call  inner=12     burst=   51.8 ms
    f80      6543.35 us/call  inner=9      burst=   58.9 ms
    g110     3373.83 us/call  inner=15     burst=   50.6 ms
    g111     3971.41 us/call  inner=13     burst=   51.6 ms
    g27      3189.07 us/call  inner=17     burst=   54.2 ms
    g80      5489.85 us/call  inner=10     burst=   54.9 ms
    legA    13349.12 us/call  inner=8      burst=  106.8 ms
    legB     5758.08 us/call  inner=8      burst=   46.1 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s110 per-token                                         gemm/s27 1.0492x  chain/s27 1.0524x
      s111 gs128      bank vs s110 1.1748x [1.1063-1.2993]  gemm/s27 1.2512x  chain/s27 1.1549x
      s80  gs128                                             gemm/s27 1.6960x  chain/s27 1.4904x
      BEST grouped s111 vs BEST per-token s27: gemm 1.2512x [1.1418-1.3412]  chain 1.1549x [1.1155-1.2141]

  down_proj k=17408 n=5120 M=2048   burst>= 50 ms, reps=7
    f110     1356.12 us/call  inner=38     burst=   51.5 ms
    f111     1704.55 us/call  inner=30     burst=   51.1 ms
    f27      1283.10 us/call  inner=39     burst=   50.0 ms
    f80      1750.77 us/call  inner=29     burst=   50.8 ms
    g110      839.63 us/call  inner=60     burst=   50.4 ms
    g111     1177.42 us/call  inner=42     burst=   49.5 ms
    g27       803.17 us/call  inner=64     burst=   51.4 ms
    g80      1241.08 us/call  inner=41     burst=   50.9 ms
    legA     3603.39 us/call  inner=15     burst=   54.1 ms
    legB     1532.24 us/call  inner=34     burst=   52.1 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s110 per-token                                         gemm/s27 1.0477x  chain/s27 1.0564x
      s111 gs128      bank vs s110 1.3939x [1.2968-1.4477]  gemm/s27 1.4596x  chain/s27 1.3324x
      s80  gs128                                             gemm/s27 1.5570x  chain/s27 1.3483x
      BEST grouped s111 vs BEST per-token s27: gemm 1.4596x [1.4087-1.5117]  chain 1.3324x [1.2130-1.3421]

  down_proj k=17408 n=5120 M=4096   burst>= 50 ms, reps=7
    f110     2739.00 us/call  inner=19     burst=   52.0 ms
    f111     3111.68 us/call  inner=17     burst=   52.9 ms
    f27      2642.73 us/call  inner=19     burst=   50.2 ms
    f80      3636.06 us/call  inner=14     burst=   50.9 ms
    g110     1665.18 us/call  inner=31     burst=   51.6 ms
    g111     2004.83 us/call  inner=25     burst=   50.1 ms
    g27      1569.14 us/call  inner=33     burst=   51.8 ms
    g80      2478.37 us/call  inner=21     burst=   52.0 ms
    legA     7415.04 us/call  inner=8      burst=   59.3 ms
    legB     2727.60 us/call  inner=19     burst=   51.8 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s110 per-token                                         gemm/s27 1.0596x  chain/s27 1.0364x
      s111 gs128      bank vs s110 1.2040x [1.1137-1.2429]  gemm/s27 1.2707x  chain/s27 1.1730x
      s80  gs128                                             gemm/s27 1.5769x  chain/s27 1.3621x
      BEST grouped s111 vs BEST per-token s27: gemm 1.2707x [1.2551-1.3347]  chain 1.1730x [1.1252-1.2427]

  down_proj k=17408 n=5120 M=8192   burst>= 50 ms, reps=7
    f110     5560.54 us/call  inner=10     burst=   55.6 ms
    f111     6059.24 us/call  inner=9      burst=   54.5 ms
    f27      5269.91 us/call  inner=10     burst=   52.7 ms
    f80      7147.52 us/call  inner=8      burst=   57.2 ms
    g110     3357.55 us/call  inner=16     burst=   53.7 ms
    g111     3841.73 us/call  inner=13     burst=   49.9 ms
    g27      3142.22 us/call  inner=17     burst=   53.4 ms
    g80      4967.01 us/call  inner=10     burst=   49.7 ms
    legA    14150.27 us/call  inner=8      burst=  113.2 ms
    legB     5205.88 us/call  inner=10     burst=   52.1 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s110 per-token                                         gemm/s27 1.0396x  chain/s27 1.0551x
      s111 gs128      bank vs s110 1.1397x [1.0587-1.2346]  gemm/s27 1.2188x  chain/s27 1.1450x
      s80  gs128                                             gemm/s27 1.5667x  chain/s27 1.3563x
      BEST grouped s111 vs BEST per-token s27: gemm 1.2188x [1.1174-1.2901]  chain 1.1450x [1.1339-1.2491]

[sampler] 914 samples over 45.8 s of process life; 34.6 s of that was inside a timed block

[under load] sm clock median 2670 MHz (min 1890, max 2850)  power median 442 W  sw_power_cap 609/690 samples  hw_slowdown 0/690
[tenancy after ] 145, 7054 MiB   (this process is pid 145)
