GPU Performance Details: Tesla K20m
System Configuration
 |
Note that this is previously stored data and does not reflect your system configuration. |
MATLAB Release: R2016a
|
Host
| Name | Intel(R) Xeon(R) CPU E5-2660 0 @ 2.20GHz |
|---|
| Clock | 2201 MHz |
|---|
| Cache | 2048 KB |
|---|
| NumProcessors | 16 |
|---|
| OSType | Windows |
|---|
| OSVersion | Microsoft Windows 7 Enterprise |
|---|
|
GPU
| Name | Tesla K20m |
|---|
| Clock | 7.055000e+02 MHz |
|---|
| NumProcessors | 13 |
|---|
| ComputeCapability | 3.5 |
|---|
| TotalMemory | 4.69 GB |
|---|
| CUDAVersion | 7.5 |
|---|
| DriverVersion | 8.17.13.5390 (353.90) |
|---|
|
Results for MTimes (double)
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 Tesla K20m - MTimes (double)
|
Array size
(elements) | Num
Operations | Time
(ms) | GigaFLOPS |
|---|
| 1,024 | 64,512 | 0.10 | 0.68 |
| 4,096 | 520,192 | 0.09 | 5.73 |
| 16,384 | 4,177,920 | 0.13 | 31.94 |
| 65,536 | 33,488,896 | 0.22 | 154.40 |
| 262,144 | 268,173,312 | 0.54 | 500.15 |
| 1,048,576 | 2,146,435,072 | 2.58 | 831.82 |
| 4,194,304 | 17,175,674,880 | 18.07 | 950.30 |
| 16,777,216 | 137,422,176,256 | 138.29 | 993.74 |
| 67,108,864 | 1,099,444,518,912 | 1095.15 | 1003.93 |
(N gigaflops = Nx109 operations per second)
|
|
|
Results for Backslash (double)
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 Tesla K20m - Backslash (double)
|
Array size
(elements) | Num
Operations | Time
(ms) | GigaFLOPS |
|---|
| 1,024 | 23,381 | 0.53 | 0.04 |
| 4,096 | 180,907 | 0.86 | 0.21 |
| 16,384 | 1,422,677 | 1.24 | 1.15 |
| 65,536 | 11,283,115 | 4.11 | 2.74 |
| 262,144 | 89,871,701 | 6.44 | 13.95 |
| 1,048,576 | 717,400,747 | 12.68 | 56.60 |
| 4,194,304 | 5,732,914,517 | 37.29 | 153.72 |
| 16,777,216 | 45,838,150,315 | 122.70 | 373.57 |
| 67,108,864 | 366,604,539,221 | 633.05 | 579.11 |
(N gigaflops = Nx109 operations per second)
|
|
|
Results for FFT (double)
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 Tesla K20m - FFT (double)
|
Array size
(elements) | Num
Operations | Time
(ms) | GigaFLOPS |
|---|
| 1,024 | 51,200 | 0.28 | 0.18 |
| 4,096 | 245,760 | 0.52 | 0.47 |
| 16,384 | 1,146,880 | 0.36 | 3.21 |
| 65,536 | 5,242,880 | 0.33 | 15.82 |
| 262,144 | 23,592,960 | 0.58 | 40.52 |
| 1,048,576 | 104,857,600 | 1.31 | 79.92 |
| 4,194,304 | 461,373,440 | 4.52 | 102.11 |
| 16,777,216 | 2,013,265,920 | 20.44 | 98.52 |
(N gigaflops = Nx109 operations per second)
|
|
|
Results for MTimes (single)
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 Tesla K20m - MTimes (single)
|
Array size
(elements) | Num
Operations | Time
(ms) | GigaFLOPS |
|---|
| 1,024 | 64,512 | 0.35 | 0.18 |
| 4,096 | 520,192 | 0.23 | 2.29 |
| 16,384 | 4,177,920 | 0.11 | 38.03 |
| 65,536 | 33,488,896 | 0.37 | 89.83 |
| 262,144 | 268,173,312 | 0.46 | 579.65 |
| 1,048,576 | 2,146,435,072 | 1.37 | 1570.18 |
| 4,194,304 | 17,175,674,880 | 7.57 | 2269.35 |
| 16,777,216 | 137,422,176,256 | 52.95 | 2595.38 |
| 67,108,864 | 1,099,444,518,912 | 413.96 | 2655.92 |
| 268,435,456 | 8,795,824,586,752 | 3433.08 | 2562.08 |
(N gigaflops = Nx109 operations per second)
|
|
|
Results for Backslash (single)
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 Tesla K20m - Backslash (single)
|
Array size
(elements) | Num
Operations | Time
(ms) | GigaFLOPS |
|---|
| 1,024 | 23,381 | 0.94 | 0.02 |
| 4,096 | 180,907 | 0.60 | 0.30 |
| 16,384 | 1,422,677 | 1.06 | 1.34 |
| 65,536 | 11,283,115 | 1.19 | 9.45 |
| 262,144 | 89,871,701 | 3.56 | 25.24 |
| 1,048,576 | 717,400,747 | 9.55 | 75.15 |
| 4,194,304 | 5,732,914,517 | 23.36 | 245.41 |
| 16,777,216 | 45,838,150,315 | 81.90 | 559.68 |
| 67,108,864 | 366,604,539,221 | 343.62 | 1066.88 |
(N gigaflops = Nx109 operations per second)
|
|
|
Results for FFT (single)
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 Tesla K20m - FFT (single)
|
Array size
(elements) | Num
Operations | Time
(ms) | GigaFLOPS |
|---|
| 1,024 | 51,200 | 0.38 | 0.13 |
| 4,096 | 245,760 | 0.31 | 0.79 |
| 16,384 | 1,146,880 | 0.14 | 8.00 |
| 65,536 | 5,242,880 | 0.14 | 36.45 |
| 262,144 | 23,592,960 | 0.20 | 120.69 |
| 1,048,576 | 104,857,600 | 0.82 | 128.44 |
| 4,194,304 | 461,373,440 | 2.38 | 194.17 |
| 16,777,216 | 2,013,265,920 | 8.67 | 232.15 |
| 67,108,864 | 8,724,152,320 | 40.17 | 217.16 |
(N gigaflops = Nx109 operations per second)
|
|
|
|