HPC-B1 Throughput Benchmark Summary
===========================================================================

GPU: NVIDIA GeForce RTX 4060 Laptop GPU
Warmup runs: 1, Timed runs: 3 (best-of)
CPU n<=10000 skipped (direct KDE path, not FFT path)


Device: cpu
   K         n   serial_ms    batch_ms   speedup    serial_dps     batch_dps
---------------------------------------------------------------------------
   1      1000        SKIP        SKIP                                      
   2      1000        SKIP        SKIP                                      
   4      1000        SKIP        SKIP                                      
   8      1000        SKIP        SKIP                                      
  16      1000        SKIP        SKIP                                      
  32      1000        SKIP        SKIP                                      
  64      1000        SKIP        SKIP                                      
   1     10000        SKIP        SKIP                                      
   2     10000        SKIP        SKIP                                      
   4     10000        SKIP        SKIP                                      
   8     10000        SKIP        SKIP                                      
  16     10000        SKIP        SKIP                                      
  32     10000        SKIP        SKIP                                      
  64     10000        SKIP        SKIP                                      
   1     50000     519.666     200.378    2.5934           1.9           5.0
   2     50000    1864.281     220.505    8.4546           1.1           9.1
   4     50000    3751.847     328.462   11.4225           1.1          12.2
   8     50000    7813.191     550.754   14.1864           1.0          14.5
  16     50000   18666.098    1262.318   14.7872           0.9          12.7
  32     50000   39388.179    2645.835   14.8869           0.8          12.1
  64     50000   78703.638    5125.212   15.3562           0.8          12.5
   1    100000     509.559      99.068    5.1435           2.0          10.1
   2    100000    1210.856     126.474    9.5739           1.7          15.8
   4    100000    2252.694     166.822   13.5036           1.8          24.0
   8    100000    4693.741     463.512   10.1265           1.7          17.3
  16    100000    9567.514     884.372   10.8184           1.7          18.1
  32    100000   19845.336    1857.276   10.6852           1.6          17.2
  64    100000   39461.609    2782.602   14.1815           1.6          23.0

Device: cuda
   K         n   serial_ms    batch_ms   speedup    serial_dps     batch_dps
---------------------------------------------------------------------------
   1      1000      55.809      57.360    0.9730          17.9          17.4
   2      1000     116.125     122.963    0.9444          17.2          16.3
   4      1000     248.114     232.290    1.0681          16.1          17.2
   8      1000     474.366     496.243    0.9559          16.9          16.1
  16      1000     932.836     899.027    1.0376          17.2          17.8
  32      1000    1921.913    1963.217    0.9790          16.7          16.3
  64      1000    3914.391    3948.316    0.9914          16.3          16.2
   1     10000     170.594     174.341    0.9785           5.9           5.7
   2     10000     335.612     335.966    0.9989           6.0           6.0
   4     10000     654.171     653.939    1.0004           6.1           6.1
   8     10000    1342.542    1314.134    1.0216           6.0           6.1
  16     10000    2766.897    2751.178    1.0057           5.8           5.8
  32     10000    5433.207    5446.712    0.9975           5.9           5.9
  64     10000   11078.021   11016.118    1.0056           5.8           5.8
   1     50000     199.843      70.753    2.8245           5.0          14.1
   2     50000     391.037      86.172    4.5378           5.1          23.2
   4     50000     827.826     107.248    7.7188           4.8          37.3
   8     50000    1591.533     187.621    8.4827           5.0          42.6
  16     50000    3202.705     300.646   10.6527           5.0          53.2
  32     50000    6448.490     579.610   11.1256           5.0          55.2
  64     50000   12928.529    1167.859   11.0703           5.0          54.8
   1    100000     144.476      78.873    1.8318           6.9          12.7
   2    100000     267.629      89.490    2.9906           7.5          22.3
   4    100000     548.709     125.825    4.3609           7.3          31.8
   8    100000    1105.306     189.824    5.8228           7.2          42.1
  16    100000    2187.897     312.520    7.0008           7.3          51.2
  32    100000    4489.215     590.102    7.6075           7.1          54.2
  64    100000    8903.811    1062.572    8.3795           7.2          60.2


Peak speedup analysis (best speedup per device/n, FFT path only)
===========================================================================

Device: cpu
  n=   1000: SKIPPED
  n=  10000: SKIPPED
  n=  50000: best speedup=15.36x at K=64  batch_dps=12.5
  n= 100000: best speedup=14.18x at K=64  batch_dps=23.0

Device: cuda
  n=   1000: best speedup=1.07x at K=4  batch_dps=17.2
  n=  10000: best speedup=1.02x at K=8  batch_dps=6.1
  n=  50000: best speedup=11.13x at K=32  batch_dps=55.2
  n= 100000: best speedup=8.38x at K=64  batch_dps=60.2


Key table: CUDA n=50000, all K values
===========================================================================
   K   serial_ms    batch_ms   speedup    serial_dps     batch_dps
------------------------------------------------------------
   1     199.843      70.753      2.82x           5.0          14.1
   2     391.037      86.172      4.54x           5.1          23.2
   4     827.826     107.248      7.72x           4.8          37.3
   8    1591.533     187.621      8.48x           5.0          42.6
  16    3202.705     300.646     10.65x           5.0          53.2
  32    6448.490     579.610     11.13x           5.0          55.2
  64   12928.529    1167.859     11.07x           5.0          54.8
