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       0.00436    0.00462    0.00445    0.00476    0.00452    0.00478    0.00451    0.00475    0.987     1.15         1.13         4.77e-07     
32    1024      0.00456    0.00478    0.00450    0.00483    0.00452    0.00481    0.00456    0.00484    0.986     4.55         4.49         4.77e-07     
48    2304      0.00462    0.00475    0.00449    0.00480    0.00455    0.00484    0.00455    0.00482    0.988     10.26        10.14        4.77e-07     
64    4096      0.00462    0.00492    0.00448    0.00475    0.00455    0.00479    0.00454    0.00474    0.986     18.29        18.04        4.77e-07     
128   16384     0.00352    0.00363    0.00337    0.00361    0.00344    0.00359    0.00343    0.00354    0.982     97.20        95.46        4.77e-07     
256   65536     0.00570    0.00579    0.00528    0.00533    0.00409    0.00412    0.00407    0.00410    1.005     320.44       321.89       9.54e-07     
512   262144    0.01451    0.01457    0.01386    0.01390    0.00855    0.00858    0.00753    0.00756    1.136     613.17       696.67       9.54e-07     
1024  1048576   0.05183    0.05405    0.05365    0.05376    0.02731    0.02739    0.02565    0.02569    1.065     767.87       817.49       9.54e-07     
2048  4194304   0.21961    0.21993    0.20524    0.20543    0.10388    0.10394    0.09973    0.09981    1.042     807.56       841.16       9.54e-07     
4096  16777216  0.87966    0.88068    0.82923    0.84803    0.52367    0.52507    0.48058    0.48248    1.090     640.76       698.20       9.54e-07     
8192  67108864  4.47605    4.48633    4.42877    4.43232    2.31518    2.31687    2.06831    2.06879    1.119     579.73       648.92       9.54e-07     

Rule 3 Distribution Audit (Per-Round Separation Verification):
L     Tiled Rounds (ms)               Slowest Tiled  Fastest Control  Separation Pass
-------------------------------------------------------------------------------------
2048  [0.09965, 0.09973, 0.09974]     0.09989        0.10089          PASS           
4096  [0.47988, 0.48058, 0.48261]     0.48339        0.52203          PASS           
8192  [2.06788, 2.06831, 2.06832]     2.07151        2.31193          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.
