device: NVIDIA GeForce RTX 4090   tenancy before: 395 MiB in use of 24109
type: IQ1_M   arms: the served spelling and its twin
  IQ1_M ggml's own 640 blocks, both tilings: EQUAL at every probed k
  IQ1_M 512x256, both tilings: EQUAL at every probed k
  IQ1_M grid vs lut at splits=1, dense 512x256, M=[1, 8]: BIT-EQUAL
L2: cold   M=1   5 races x 21 paired reps
shape      arm       MiB   bar us  live us  null us    GB/s  of bar null/bar live/null  kc/splits
gate/up    lut     45.82    54.27    64.51    62.46   744.8   0.841    1.151     1.033   320/16  
gate/up    grid    19.26    25.60    71.07    37.28   284.1   0.360    1.456     1.906   480/11  
                 grid/lut  1.107 [1.098, 1.112]  control 1.000 [0.991, 1.014]  over 5 races  SEPARATED
down       lut     45.82    55.30    66.43    64.48   723.2   0.832    1.166     1.030  1088/16  
down       grid    19.26    25.60    75.78    41.98   266.5   0.338    1.640     1.805  1088/16  
                 grid/lut  1.149 [1.141, 1.152]  control 1.000 [0.985, 1.015]  over 5 races  SEPARATED
k/v        lut      2.70     8.19    15.36    15.36   184.0   0.533    1.875     1.000   160/32  
k/v        grid     1.13     4.86    19.46    17.41    61.1   0.250    3.579     1.118   320/16  
                 grid/lut  1.235 [1.185, 1.266]  control 0.990 [0.907, 1.034]  over 5 races  SEPARATED
tenancy after: 719 MiB in use of 24109
RACE_EXIT=0
