From: Multistep schemes for solving backward stochastic differential equations on GPU
K | N | M | \(|y_{0,0}-y_{0}^{0}|\) | \(|z_{0,0}-z_{0}^{0}|\) | \(t_{CPU}\) | \(t_{GPU}\) | speedup |
---|---|---|---|---|---|---|---|
1 | 128 | 364 | 7.85E−04 | 3.52E−03 | 0.32 | 1.17 | 0.27 |
1 | 256 | 512 | 3.77E−04 | 1.76E−03 | 0.88 | 2.13 | 0.41 |
1 | 512 | 726 | 1.85E−04 | 8.78E−04 | 2.52 | 4.04 | 0.62 |
1 | 1024 | 1024 | 9.15E−05 | 4.39E−04 | 6.98 | 7.80 | 0.89 |
2 | 128 | 1218 | 1.85E−04 | 8.37E−04 | 1.52 | 1.24 | 1.23 |
2 | 256 | 2048 | 9.13E−05 | 4.29E−04 | 5.13 | 2.51 | 2.04 |
2 | 512 | 3446 | 4.54E−05 | 2.17E−04 | 17.47 | 5.11 | 3.42 |
2 | 1024 | 5794 | 2.26E−05 | 1.09E−04 | 58.93 | 10.65 | 5.53 |
3 | 128 | 4096 | 1.92E−07 | 8.34E−07 | 6.61 | 1.53 | 4.31 |
3 | 256 | 8192 | 2.41E−08 | 1.06E−07 | 26.81 | 2.97 | 9.03 |
3 | 512 | 16,384 | 3.02E−09 | 1.33E−08 | 108.92 | 6.62 | 16.46 |
3 | 1024 | 32,768 | 3.77E−10 | 1.67E−09 | 435.23 | 18.35 | 23.71 |
4 | 128 | 4096 | 1.10E−07 | 4.86E−07 | 8.06 | 1.53 | 5.28 |
4 | 256 | 8192 | 1.42E−08 | 6.28E−08 | 32.82 | 3.02 | 10.87 |
4 | 512 | 16,384 | 1.80E−09 | 7.99E−09 | 133.26 | 6.47 | 20.61 |
4 | 1024 | 32,768 | 2.27E−10 | 1.01E−09 | 538.13 | 19.33 | 27.84 |
5 | 128 | 4096 | 1.20E−07 | 5.40E−07 | 9.48 | 1.54 | 6.14 |
5 | 256 | 8192 | 1.58E−08 | 7.04E−08 | 38.68 | 2.97 | 13.05 |
5 | 512 | 16,384 | 2.02E−09 | 8.99E−09 | 156.63 | 6.67 | 23.48 |
5 | 1024 | 32,768 | 2.55E−10 | 1.14E−09 | 635.01 | 19.55 | 32.48 |
6 | 128 | 4096 | 1.11E−07 | 5.08E−07 | 10.91 | 1.54 | 7.07 |
6 | 256 | 8192 | 1.49E−08 | 6.71E−08 | 44.77 | 3.09 | 14.48 |
6 | 512 | 16,384 | 1.93E−09 | 8.63E−09 | 182.74 | 7.15 | 25.57 |
6 | 1024 | 32,768 | 2.45E−10 | 1.09E−09 | 735.15 | 20.97 | 35.05 |