GPU: NVIDIA A40
Python Version: 3.12.3
PyTorch Version: 2.8.0+cu128
CUDA Version: 12.8
Driver Version: 595.91.07
L2 Cache: 6.0 MB
Theoretical Bandwidth: 696.0 GB/s

WARNING: Logical bytes != DRAM transactions. Achieved GB/s > theoretical DRAM ceiling indicates coalescing reuse. The theoretical ceiling applies to real traffic.

Quattra Tiled Benchmark (3 rounds of 20 iters per L, median & p95 reported)
L     N         Mod_med    Mod_p95    Cond_med   Cond_p95   C16_med    C16_p95    Tile_med   Tile_p95   Speedup   Naive GB/s   Tiled GB/s   Max Abs Diff 
---------------------------------------------------------------------------------------------------------------------------------------------------------
16    256       0.00319    0.00341    0.00322    0.00347    0.00325    0.00342    0.00315    0.00332    1.025     1.59         1.63         4.77e-07     
32    1024      0.00322    0.00344    0.00317    0.00324    0.00319    0.00336    0.00316    0.00326    1.003     6.46         6.48         4.77e-07     
48    2304      0.00320    0.00336    0.00315    0.00328    0.00319    0.00342    0.00320    0.00333    0.983     14.63        14.38        4.77e-07     
64    4096      0.00320    0.00333    0.00313    0.00325    0.00315    0.00328    0.00322    0.00327    0.974     26.17        25.48        4.77e-07     
128   16384     0.00352    0.00357    0.00331    0.00355    0.00335    0.00349    0.00329    0.00340    1.004     99.13        99.54        4.77e-07     
256   65536     0.00570    0.00573    0.00525    0.00529    0.00406    0.00410    0.00403    0.00405    1.009     322.65       325.47       9.54e-07     
512   262144    0.01453    0.01459    0.01389    0.01394    0.00855    0.00859    0.00755    0.00758    1.133     613.29       694.83       9.54e-07     
1024  1048576   0.05232    0.05270    0.05419    0.05459    0.02750    0.02759    0.02583    0.02589    1.065     762.53       811.82       9.54e-07     
2048  4194304   0.22181    0.22237    0.20746    0.20755    0.10495    0.10502    0.10012    0.10018    1.048     799.27       837.86       9.54e-07     
4096  16777216  0.88411    0.88469    0.83346    0.84750    0.52312    0.52421    0.48226    0.48272    1.085     641.43       695.78       9.54e-07     
8192  67108864  4.47608    4.48686    4.42950    4.43295    2.31483    2.31716    2.06830    2.07090    1.119     579.82       648.93       9.54e-07     

Rule 3 Distribution Audit (Per-Round Separation Verification):
L     Tiled Rounds (ms)               Slowest Tiled  Fastest Control  Separation Pass
-------------------------------------------------------------------------------------
2048  [0.10009, 0.10012, 0.10017]     0.10027        0.10181          PASS           
4096  [0.48048, 0.48226, 0.48261]     0.48294        0.52207          PASS           
8192  [2.06821, 2.06830, 2.06831]     2.07133        2.31236          PASS           

Pre-Registered Pass/Fail Evaluation:
Pre-Registered Rules: (1) >3% speedup vs best control, (2) wins at >=2 of {2048, 4096, 8192}, (3) slowest-of-3 tiled < fastest-of-3 control.
RESULT: PASS — skew_tiled achieved >= 3% speedup with statistically separating bounds at 3/3 large sizes.
