n,d,k,platform,backend,e2e_ms,notes
2,64,15,macOS,cpu,0.001,
4,64,15,macOS,cpu,0.001,
8,64,15,macOS,cpu,0.002,
16,64,15,macOS,cpu,0.007,
32,64,15,macOS,cpu,0.019,
64,64,15,macOS,cpu,0.093,
128,64,15,macOS,cpu,0.238,
256,64,15,macOS,cpu,0.667,
512,64,15,macOS,cpu,1.663,
1024,64,15,macOS,cpu,6.049,
2048,64,15,macOS,cpu,21.699,
4096,64,15,macOS,cpu,88.363,
2,64,15,macOS,metal,0.448,
4,64,15,macOS,metal,0.393,
8,64,15,macOS,metal,0.366,
16,64,15,macOS,metal,0.387,
32,64,15,macOS,metal,0.374,
64,64,15,macOS,metal,0.521,
128,64,15,macOS,metal,0.542,
256,64,15,macOS,metal,0.698,
512,64,15,macOS,metal,1.31,
1024,64,15,macOS,metal,2.396,
2048,64,15,macOS,metal,3.41,
4096,64,15,macOS,metal,10.599,
2,64,15,macOS,mlx,0.212,
4,64,15,macOS,mlx,0.241,
8,64,15,macOS,mlx,0.212,
16,64,15,macOS,mlx,0.223,
32,64,15,macOS,mlx,0.218,
64,64,15,macOS,mlx,0.37,
128,64,15,macOS,mlx,0.404,
256,64,15,macOS,mlx,0.513,
512,64,15,macOS,mlx,0.651,
1024,64,15,macOS,mlx,1.21,
2048,64,15,macOS,mlx,2.644,
4096,64,15,macOS,mlx,11.2,
2,64,15,macOS,wgpu,7.867,
4,64,15,macOS,wgpu,7.889,
8,64,15,macOS,wgpu,7.914,
16,64,15,macOS,wgpu,7.93,
32,64,15,macOS,wgpu,7.946,
64,64,15,macOS,wgpu,8.163,
128,64,15,macOS,wgpu,8.197,
256,64,15,macOS,wgpu,8.402,
512,64,15,macOS,wgpu,9.489,
1024,64,15,macOS,wgpu,9.219,
2048,64,15,macOS,wgpu,13.031,
4096,64,15,macOS,wgpu,34.661,
2,64,15,Win,cpu,0.003,
4,64,15,Win,cpu,0.006,
8,64,15,Win,cpu,0.013,
16,64,15,Win,cpu,0.027,
32,64,15,Win,cpu,0.076,
64,64,15,Win,cpu,0.345,
128,64,15,Win,cpu,0.818,
256,64,15,Win,cpu,2.558,
512,64,15,Win,cpu,7.782,
1024,64,15,Win,cpu,28.609,
2048,64,15,Win,cpu,110.133,
4096,64,15,Win,cpu,436.52,
2,64,15,Win,wgpu,0.814,
4,64,15,Win,wgpu,0.867,
8,64,15,Win,wgpu,0.838,
16,64,15,Win,wgpu,0.895,
32,64,15,Win,wgpu,0.949,
64,64,15,Win,wgpu,1.28,
128,64,15,Win,wgpu,1.545,
256,64,15,Win,wgpu,1.613,
512,64,15,Win,wgpu,2.487,
1024,64,15,Win,wgpu,3.986,
2048,64,15,Win,wgpu,7.204,
4096,64,15,Win,wgpu,14.01,
2,64,15,Win,cuda,0.264,
4,64,15,Win,cuda,0.244,
8,64,15,Win,cuda,0.253,
16,64,15,Win,cuda,0.28,
32,64,15,Win,cuda,0.268,
64,64,15,Win,cuda,0.417,
128,64,15,Win,cuda,0.537,
256,64,15,Win,cuda,0.795,
512,64,15,Win,cuda,1.754,
1024,64,15,Win,cuda,3.283,
2048,64,15,Win,cuda,9.818,
4096,64,15,Win,cuda,34.834,
2,64,15,WSL,cpu,0.003,
4,64,15,WSL,cpu,0.006,
8,64,15,WSL,cpu,0.014,
16,64,15,WSL,cpu,0.03,
32,64,15,WSL,cpu,0.086,
64,64,15,WSL,cpu,1.687,
128,64,15,WSL,cpu,2.724,
256,64,15,WSL,cpu,18.083,
512,64,15,WSL,cpu,38.712,
1024,64,15,WSL,cpu,164.392,
2048,64,15,WSL,cpu,248.084,
4096,64,15,WSL,cpu,592.185,
2,64,15,WSL,wgpu,4.496,
4,64,15,WSL,wgpu,4.109,
8,64,15,WSL,wgpu,4.23,
16,64,15,WSL,wgpu,4.644,
32,64,15,WSL,wgpu,4.761,
64,64,15,WSL,wgpu,6.964,
128,64,15,WSL,wgpu,8.344,
256,64,15,WSL,wgpu,10.957,
512,64,15,WSL,wgpu,21.094,
1024,64,15,WSL,wgpu,57.568,
2048,64,15,WSL,wgpu,209.278,
4096,64,15,WSL,wgpu,,wgpu arena > 128MiB binding cap at n=4096
2,64,15,WSL,cuda,0.433,
4,64,15,WSL,cuda,0.397,
8,64,15,WSL,cuda,0.448,
16,64,15,WSL,cuda,0.423,
32,64,15,WSL,cuda,1.355,
64,64,15,WSL,cuda,2.514,
128,64,15,WSL,cuda,1.792,
256,64,15,WSL,cuda,2.862,
512,64,15,WSL,cuda,3.596,
1024,64,15,WSL,cuda,7.745,
2048,64,15,WSL,cuda,21.337,
4096,64,15,WSL,cuda,75.022,
