Skip to main content

Table 5 Workload of three main steps for point sets derived from 3D models on GTX 660M

From: CudaChain: an alternative algorithm for finding 2D convex hulls on the GPU

3D Model

Size

Running time (/ms)

Percentage of running time

First step

Second step

Third step

Total

First step

Second step

Third step

Armadillo

172k

4.3

33.3

2.1

39.7

10.83

83.88

5.29

Angel

237k

4.8

33.9

2.9

41.6

11.54

81.49

6.97

Skeleton hand

327k

4.9

36.6

3.9

45.4

10.79

80.62

8.59

Dragon

437k

6.6

47.3

5.9

59.8

11.04

79.10

9.87

Happy Buddha

543k

7.3

52.3

9.1

68.7

10.63

76.13

13.25

Turbine blade

882k

8.4

59.1

6.4

73.9

11.37

79.97

8.66

Vellum manuscript

2M

11.7

75.2

3.7

90.6

12.91

83.00

4.08

Asian Dragon

3M

16.4

81.5

3.8

101.7

16.13

80.14

3.74

Thai statue

5M

21.0

81.4

3.6

106.0

19.81

76.79

3.40

Lucy

14M

60.7

180.2

4.3

245.2

24.76

73.49

1.75

  1. (1) The first step is the first round of discarding on the GPU. The second step is the second round of discarding on the GPU. The third step is the finalization of computing the convex hull on the CPU. (2) The running time of the first step includes the overhead of transferring data from the host side to the device side; and the running time of the second step includes the overhead of transferring data from the device side back to the host side