[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
  s27    128x128x128 1x4 st2 sw8         128   199   49152      2        8 128
  s110   128x128x128 1x4 st2 sw8         128   254   49152      2        8 128   <-- SPILLS
  s111   128x128x128 1x4 st2 sw8 gs128   128   255   50176      2        8 128x2

  q_proj k=5120 n=12288 M=2048   burst>= 50 ms, reps=5
    f111      870.04 us/call  inner=60     burst=   52.2 ms
    f27       798.20 us/call  inner=66     burst=   52.7 ms
    g111      661.83 us/call  inner=79     burst=   52.3 ms
    g27       584.56 us/call  inner=87     burst=   50.9 ms
    legA     2603.89 us/call  inner=21     burst=   54.7 ms
    legB     1206.56 us/call  inner=43     burst=   51.9 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s111 gs128                                             gemm/s27 1.1460x  chain/s27 1.0714x
      BEST grouped s111 vs BEST per-token s27: gemm 1.1460x [1.0750-1.2057]  chain 1.0714x [1.0117-1.2240]

  q_proj k=5120 n=12288 M=8192   burst>= 50 ms, reps=5
    f111     3609.36 us/call  inner=14     burst=   50.5 ms
    f27      3357.49 us/call  inner=15     burst=   50.4 ms
    g111     2579.31 us/call  inner=21     burst=   54.2 ms
    g27      2314.86 us/call  inner=21     burst=   48.6 ms
    legA    10157.44 us/call  inner=8      burst=   81.3 ms
    legB     4216.17 us/call  inner=12     burst=   50.6 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s111 gs128                                             gemm/s27 1.1093x  chain/s27 1.0773x
      BEST grouped s111 vs BEST per-token s27: gemm 1.1093x [0.9925-1.2856]  chain 1.0773x [0.9969-1.1528]

  k_proj k=5120 n=1024 M=2048   burst>= 50 ms, reps=5
    f111      177.47 us/call  inner=266    burst=   47.2 ms
    f27       152.01 us/call  inner=310    burst=   47.1 ms
    g111       67.97 us/call  inner=680    burst=   46.2 ms
    g27        59.88 us/call  inner=779    burst=   46.6 ms
    legA      838.19 us/call  inner=61     burst=   51.1 ms
    legB      107.04 us/call  inner=481    burst=   51.5 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s111 gs128                                             gemm/s27 1.1506x  chain/s27 1.1742x
      BEST grouped s111 vs BEST per-token s27: gemm 1.1506x [1.1290-1.1608]  chain 1.1742x [1.0968-1.3020]

  k_proj k=5120 n=1024 M=8192   burst>= 50 ms, reps=5
    f111      842.97 us/call  inner=61     burst=   51.4 ms
    f27       788.81 us/call  inner=65     burst=   51.3 ms
    g111      213.67 us/call  inner=241    burst=   51.5 ms
    g27       193.01 us/call  inner=261    burst=   50.4 ms
    legA     3540.72 us/call  inner=15     burst=   53.1 ms
    legB      369.84 us/call  inner=136    burst=   50.3 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s111 gs128                                             gemm/s27 1.1235x  chain/s27 1.0681x
      BEST grouped s111 vs BEST per-token s27: gemm 1.1235x [0.9210-1.1336]  chain 1.0681x [1.0003-1.0916]

  v_proj k=5120 n=1024 M=2048   burst>= 50 ms, reps=5
    f111      178.01 us/call  inner=283    burst=   50.4 ms
    f27       156.33 us/call  inner=330    burst=   51.6 ms
    g111       68.38 us/call  inner=732    burst=   50.1 ms
    g27        59.28 us/call  inner=835    burst=   49.5 ms
    legA      839.36 us/call  inner=60     burst=   50.4 ms
    legB      106.39 us/call  inner=464    burst=   49.4 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s111 gs128                                             gemm/s27 1.1524x  chain/s27 1.1637x
      BEST grouped s111 vs BEST per-token s27: gemm 1.1524x [1.1319-1.1652]  chain 1.1637x [1.1261-1.1745]

  v_proj k=5120 n=1024 M=8192   burst>= 50 ms, reps=5
    f111      837.11 us/call  inner=61     burst=   51.1 ms
    f27       794.88 us/call  inner=65     burst=   51.7 ms
    g111      216.84 us/call  inner=242    burst=   52.5 ms
    g27       208.29 us/call  inner=262    burst=   54.6 ms
    legA     3536.15 us/call  inner=15     burst=   53.0 ms
    legB      367.34 us/call  inner=137    burst=   50.3 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s111 gs128                                             gemm/s27 1.0051x  chain/s27 1.0543x
      BEST grouped s111 vs BEST per-token s27: gemm 1.0051x [0.9774-1.1287]  chain 1.0543x [0.9928-1.0837]

  o_proj k=6144 n=5120 M=2048   burst>= 50 ms, reps=5
    f111      492.38 us/call  inner=105    burst=   51.7 ms
    f27       440.04 us/call  inner=112    burst=   49.3 ms
    g111      327.92 us/call  inner=149    burst=   48.9 ms
    g27       298.45 us/call  inner=167    burst=   49.8 ms
    legA     1621.15 us/call  inner=33     burst=   53.5 ms
    legB      612.52 us/call  inner=84     burst=   51.5 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s111 gs128                                             gemm/s27 1.1037x  chain/s27 1.1359x
      BEST grouped s111 vs BEST per-token s27: gemm 1.1037x [1.0666-1.2298]  chain 1.1359x [1.0106-1.2441]

  o_proj k=6144 n=5120 M=8192   burst>= 50 ms, reps=5
    f111     2176.34 us/call  inner=24     burst=   52.2 ms
    f27      2168.60 us/call  inner=26     burst=   56.4 ms
    g111     1270.98 us/call  inner=41     burst=   52.1 ms
    g27      1159.12 us/call  inner=44     burst=   51.0 ms
    legA     6180.10 us/call  inner=8      burst=   49.4 ms
    legB     2069.14 us/call  inner=25     burst=   51.7 ms
    per-arm ratios, paired within a rep:
      s27  per-token                                         gemm/s27 1.0000x  chain/s27 1.0000x
      s111 gs128                                             gemm/s27 1.0965x  chain/s27 1.0036x
      BEST grouped s111 vs BEST per-token s27: gemm 1.0965x [1.0613-1.2743]  chain 1.0036x [0.9767-1.1084]

[sampler] 574 samples over 28.7 s of process life; 12.6 s of that was inside a timed block

[under load] sm clock median 210 MHz (min 210, max 210)  power median 34 W  sw_power_cap 0/251 samples  hw_slowdown 0/251
[tenancy after ] (none)   (this process is pid 1025)
DONE
