GPU Performance Details: Host PC

MATLAB Release: R2015a

These results show the performance of the GPU or host PC when calculating a matrix multiplication of two NxN real matrices. The number of operations is assumed to be 2*N^3 - N^2.

This calculation is usually compute-bound, i.e. the performance depends mainly on how fast the GPU or host PC can perform floating-point operations.

Raw data for Host PC - MTimes (double) Array size (elements) Num Operations Time (ms) GigaFLOPS 1,024 64,512 0.03 2.14 4,096 520,192 0.08 6.53 16,384 4,177,920 0.08 54.79 65,536 33,488,896 0.30 110.38 262,144 268,173,312 2.00 133.77 1,048,576 2,146,435,072 15.99 134.25 4,194,304 17,175,674,880 125.26 137.12 16,777,216 137,422,176,256 1050.56 130.81 67,108,864 1,099,444,518,912 8437.32 130.31 (N gigaflops = Nx109 operations per second)

These results show the performance of the GPU or host PC when calculating the matrix left division of an NxN matrix with an Nx1 vector. The number of operations is assumed to be 2/3*N^3 + 3/2*N^2.

This calculation is usually compute-bound, i.e. the performance depends mainly on how fast the GPU or host PC can perform floating-point operations.

Raw data for Host PC - Backslash (double) Array size (elements) Num Operations Time (ms) GigaFLOPS 1,024 23,381 0.06 0.39 4,096 180,907 0.11 1.61 16,384 1,422,677 0.24 5.99 65,536 11,283,115 0.62 18.17 262,144 89,871,701 2.13 42.18 1,048,576 717,400,747 11.68 61.40 4,194,304 5,732,914,517 73.74 77.75 16,777,216 45,838,150,315 530.23 86.45 67,108,864 366,604,539,221 3687.73 99.41 (N gigaflops = Nx109 operations per second)

These results show the performance of the GPU or host PC when calculating the Fast-Fourier-Transform of a vector of complex numbers. The number of operations for a vector of length N is assumed to be 5*N*log2(N).

This calculation is usually memory-bound, i.e. the performance depends mainly on how fast the GPU or host PC can read and write data.

Raw data for Host PC - FFT (double) Array size (elements) Num Operations Time (ms) GigaFLOPS 1,024 51,200 0.03 1.69 4,096 245,760 0.06 4.23 16,384 1,146,880 0.20 5.80 65,536 5,242,880 0.61 8.59 262,144 23,592,960 2.50 9.43 1,048,576 104,857,600 15.92 6.59 4,194,304 461,373,440 74.89 6.16 16,777,216 2,013,265,920 579.64 3.47 (N gigaflops = Nx109 operations per second)

These results show the performance of the GPU or host PC when calculating a matrix multiplication of two NxN real matrices. The number of operations is assumed to be 2*N^3 - N^2.

This calculation is usually compute-bound, i.e. the performance depends mainly on how fast the GPU or host PC can perform floating-point operations.

Raw data for Host PC - MTimes (single) Array size (elements) Num Operations Time (ms) GigaFLOPS 1,024 64,512 0.03 2.31 4,096 520,192 0.06 8.56 16,384 4,177,920 0.07 61.49 65,536 33,488,896 0.23 143.86 262,144 268,173,312 1.48 181.35 1,048,576 2,146,435,072 8.82 243.30 4,194,304 17,175,674,880 68.52 250.68 16,777,216 137,422,176,256 557.74 246.39 67,108,864 1,099,444,518,912 4570.42 240.56 (N gigaflops = Nx109 operations per second)

These results show the performance of the GPU or host PC when calculating the matrix left division of an NxN matrix with an Nx1 vector. The number of operations is assumed to be 2/3*N^3 + 3/2*N^2.

This calculation is usually compute-bound, i.e. the performance depends mainly on how fast the GPU or host PC can perform floating-point operations.

Raw data for Host PC - Backslash (single) Array size (elements) Num Operations Time (ms) GigaFLOPS 1,024 23,381 0.06 0.40 4,096 180,907 0.09 2.05 16,384 1,422,677 0.17 8.19 65,536 11,283,115 0.58 19.40 262,144 89,871,701 1.86 48.21 1,048,576 717,400,747 8.20 87.54 4,194,304 5,732,914,517 45.07 127.21 16,777,216 45,838,150,315 261.61 175.21 67,108,864 366,604,539,221 1899.99 192.95 (N gigaflops = Nx109 operations per second)

These results show the performance of the GPU or host PC when calculating the Fast-Fourier-Transform of a vector of complex numbers. The number of operations for a vector of length N is assumed to be 5*N*log2(N).

This calculation is usually memory-bound, i.e. the performance depends mainly on how fast the GPU or host PC can read and write data.

Raw data for Host PC - FFT (single) Array size (elements) Num Operations Time (ms) GigaFLOPS 1,024 51,200 0.05 1.10 4,096 245,760 0.07 3.40 16,384 1,146,880 0.16 7.33 65,536 5,242,880 0.50 10.58 262,144 23,592,960 1.64 14.43 1,048,576 104,857,600 11.62 9.03 4,194,304 461,373,440 53.49 8.63 16,777,216 2,013,265,920 275.21 7.32 (N gigaflops = Nx109 operations per second)