Cuda and Acer Chromebook 13
November 22, 2014Here's how to use Tegra K1 on Acer Chromebook 13.
Tegra finally arrived to the chromebook world! The TK1 chip gives really cool possibilites with 192 Cuda and 4+1 ARM cores. This post will show how to install chrubuntu on the Acer Chromebook 13.
Steps:
- Make sure chromebook is in the developer mode, good instructions are on the crouton's. I recommend setting up chrubuntu on the external SD card. Acer's formfactor hides SD card pretty well as opposed to Samsung Chromebook 3
Download chrubuntu-tegra.sh from this repository and run from Chrome shell as root. Original author is arm000 from reddit
sudo bash chrubuntu-tegra.sh default 14.04 /dev/mmcblk1Login with user and password user
Download cuda
wget http://developer.download.nvidia.com/compute/cuda/6_5/rel/installers/cuda-repo-l4t-r21.1-6-5-prod_6.5-14_armhf.deb sudo dpkg -i cuda-repo-l4t-r21.1-6-5-prod_6.5-14_armhf.deb cd /var/cuda-repo-6-5-prod sudo dpkg -i cuda*Let's test cuda
cd ~ cuda-install-samples-6.5.sh cd NVIDIA_CUDA-6.5_Samples/1_Utilities/deviceQuery make ./deviceQuery ./deviceQuery Starting CUDA Device Query (Runtime API) version (CUDART static linking) Detected 1 CUDA Capable device(s) Device 0: "GK20A" CUDA Driver Version / Runtime Version 6.5 / 6.5 CUDA Capability Major/Minor version number: 3.2 Total amount of global memory: 4017 MBytes (4212305920 bytes) (1) Multiprocessors, (192) CUDA Cores/MP: 192 CUDA Cores GPU Clock rate: 72 MHz (0.07 GHz) Memory Clock rate: 13 Mhz Memory Bus Width: 64-bit L2 Cache Size: 131072 bytes Maximum Texture Dimension Size (x,y,z) 1D=(65536), 2D=(65536, 65536), 3D=(4096, 4096, 4096) Maximum Layered 1D Texture Size, (num) layers 1D=(16384), 2048 layers Maximum Layered 2D Texture Size, (num) layers 2D=(16384, 16384), 2048 layers Total amount of constant memory: 65536 bytes Total amount of shared memory per block: 49152 bytes Total number of registers available per block: 32768 Warp size: 32 Maximum number of threads per multiprocessor: 2048 Maximum number of threads per block: 1024 Max dimension size of a thread block (x,y,z): (1024, 1024, 64) Max dimension size of a grid size (x,y,z): (2147483647, 65535, 65535) Maximum memory pitch: 2147483647 bytes Texture alignment: 512 bytes Concurrent copy and kernel execution: Yes with 1 copy engine(s) Run time limit on kernels: No Integrated GPU sharing Host Memory: Yes Support host page-locked memory mapping: Yes Alignment requirement for Surfaces: Yes Device has ECC support: Disabled Device supports Unified Addressing (UVA): Yes Device PCI Bus ID / PCI location ID: 0 / 0 Compute Mode: < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) > deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 6.5, CUDA Runtime Version = 6.5, NumDevs = 1, Device0 = GK20A Result = PASSLet's run matrix multiplication using BLAS
cd ~/NVIDIA_CUDA-6.5_Samples/0_Simple/matrixMulCUBLAS make ./matrixMulCUBLAS Matrix Multiply CUBLAS] - Starting... GPU Device 0: "GK20A" with compute capability 3.2 MatrixA(320,640), MatrixB(320,640), MatrixC(320,640) Computing result using CUBLAS...done. Performance= 99.09 GFlop/s, Time= 1.323 msec, Size= 131072000 Ops Computing result using host CPU...done. Comparing CUBLAS Matrix Multiply with CPU results: PASSHere's performance test of opencv using cpu and gpu
opencv-2.4.9/samples/gpu$ ./performance/test Device 0: "GK20A" 4017Mb, sm_32, Driver/Runtime ver.6.50/6.50 Note: the timings for GPU don't include data transfer CPU, ms GPU, ms SPEEDUP DESCRIPTION matchTemplate: 858.114 72.0609 x11.9 3000x3000, 32FC1, templ 5x5, CCORR 985.549 255.025 x3.86 3000x3000, 32FC1, templ 25x25, CCORR 1748.13 789.76 x2.21 3000x3000, 32FC1, templ 125x125, CCORR minMaxLoc: 19.315 20.6993 x0.933 2000x2000, 32F 77.0511 15.5263 x4.96 4000x4000, 32F 307.729 72.4855 x4.25 8000x8000, 32F remap: 90.363 7.50256 x12 1000x1000, 8UC4, INTER_LINEAR, BORDER_REPLICATE 366.899 17.3507 x21.1 2000x2000, 8UC4, INTER_LINEAR, BORDER_REPLICATE 1473.39 53.4025 x27.6 4000x4000, 8UC4, INTER_LINEAR, BORDER_REPLICATE dft: 108.717 78.3432 x1.39 1000x1000, 32FC2, complex-to-complex 484.56 400.052 x1.21 2000x2000, 32FC2, complex-to-complex 2134.01 1620.15 x1.32 4000x4000, 32FC2, complex-to-complex cornerHarris: 108.305 103.327 x1.05 1000x1000, 32FC1, BORDER_REFLECT101 450.851 496.52 x0.908 2000x2000, 32FC1, BORDER_REFLECT101 1881.31 1641.21 x1.15 4000x4000, 32FC1, BORDER_REFLECT101 integral: 2.96307 5.38107 x0.551 1000x1000, 8UC1 9.89759 16.2014 x0.611 2000x2000, 8UC1 38.8796 747.072 x0.052 4000x4000, 8UC1 norm: 66.5294 10.6661 x6.24 2000x2000, 32FC4, NORM_INF 148.849 143.465 x1.04 3000x3000, 32FC4, NORM_INF 265.1 83.408 x3.18 4000x4000, 32FC4, NORM_INF meanShift: 898.304 68.3235 x13.1 400x400, 8UC3 vs 8UC4 3665.37 296.622 x12.4 800x800, 8UC3 vs 8UC4 SURF: 11462.1 1404.94 x8.16 FAST: 44.5021 26.4897 x1.68 ORB: 395.996 133.94 x2.96 BruteForceMatcher: 1499.82 80.2282 x18.7 match 1498.06 79.4344 x18.9 knnMatch 1562.1 117.872 x13.3 radiusMatch magnitude: 31.7564 10.762 x2.95 2000x2000, 32FC1 71.0263 35.5927 x2 3000x3000, 32FC1 125.714 128.134 x0.981 4000x4000, 32FC1 add: 18.707 7.99203 x2.34 2000x2000, 32FC1 42.7588 11.0964 x3.85 3000x3000, 32FC1 74.9048 133.128 x0.563 4000x4000, 32FC1 log: 74.6291 4.42015 x16.9 2000x2000, 32F 167.232 8.72713 x19.2 3000x3000, 32F 297.186 14.2314 x20.9 4000x4000, 32F mulSpectrums: 40.4223 9.7163 x4.16 2000x2000 90.7949 21.1238 x4.3 3000x3000 169.98 36.9861 x4.6 4000x4000 resize: 106.637 11.4671 x9.3 1000x1000, 8UC4, up 425.265 32.8241 x13 2000x2000, 8UC4, up 916.219 71.8048 x12.8 3000x3000, 8UC4, up 3.19822 1.97371 x1.62 1000x1000, 8UC4, down 11.768 6.89636 x1.71 2000x2000, 8UC4, down 26.8637 8.29564 x3.24 3000x3000, 8UC4, down cvtColor: 41.3608 9.47918 x4.36 4000x4000, 8UC1, CV_GRAY2BGRA 256.231 16.2589 x15.8 4000x4000, 8UC3 vs 8UC4, CV_BGR2YCrCb 305.405 15.5065 x19.7 4000x4000, 8UC4, CV_YCrCb2BGR 243.121 17.6773 x13.8 4000x4000, 8UC3 vs 8UC4, CV_BGR2XYZ 262.035 15.625 x16.8 4000x4000, 8UC4, CV_XYZ2BGR 316.473 39.6782 x7.98 4000x4000, 8UC3 vs 8UC4, CV_BGR2HSV 1482.77 22.8313 x64.9 4000x4000, 8UC4, CV_HSV2BGR erode: 113.027 30.9059 x3.66 2000x2000 258.124 60.2367 x4.29 3000x3000 466.415 236.753 x1.97 4000x4000 threshold: 4.94594 2.41176 x2.05 2000x2000, 8UC1, THRESH_BINARY 11.0253 6.11446 x1.8 3000x3000, 8UC1, THRESH_BINARY 19.598 8.82527 x2.22 4000x4000, 8UC1, THRESH_BINARY 10.016 3.97195 x2.52 2000x2000, 32FC1, THRESH_TRUNC [NPP] 22.4392 31.2924 x0.717 3000x3000, 32FC1, THRESH_TRUNC [NPP] 40.0042 26.4149 x1.51 4000x4000, 32FC1, THRESH_TRUNC [NPP] pow: 11.9931 2.18587 x5.49 1000x1000, 32F 47.2433 3.71193 x12.7 2000x2000, 32F 105.984 7.8099 x13.6 3000x3000, 32F 189.018 13.3482 x14.2 4000x4000, 32F projectPoints: 148.576 15.1322 x9.82 1000000 108.388 10.4727 x10.3 714285 76.3413 10.1459 x7.52 510203 55.0224 7.87239 x6.99 364430 38.8461 6.65774 x5.83 260307 solvePnPRansac: 554.475 343.844 x1.61 5000 1292.95 399.338 x3.24 18800 5010.72 627.099 x7.99 70688 17729.6 1454.83 x12.2 265786 GaussianBlur: 36.0077 16.7083 x2.16 1000x1000, 8UC4 146.934 29.5295 x4.98 2000x2000, 8UC4 333.92 58.5911 x5.7 3000x3000, 8UC4 601.25 197.405 x3.05 4000x4000, 8UC4 filter2D: 41.9738 13.9314 x3.01 ksize = 3, 512x512, 8UC4 72.3195 17.1541 x4.22 ksize = 5, 512x512, 8UC4 115.287 20.4971 x5.62 ksize = 7, 512x512, 8UC4 181.174 25.6706 x7.06 ksize = 9, 512x512, 8UC4 181.107 34.2712 x5.28 ksize = 11, 512x512, 8UC4 181.266 46.1384 x3.93 ksize = 13, 512x512, 8UC4 181.523 58.7195 x3.09 ksize = 15, 512x512, 8UC4 168.459 19.6847 x8.56 ksize = 3, 1024x1024, 8UC4 290.151 32.4336 x8.95 ksize = 5, 1024x1024, 8UC4 461.43 52.3748 x8.81 ksize = 7, 1024x1024, 8UC4 566.945 81.0616 x6.99 ksize = 9, 1024x1024, 8UC4 567.505 115.59 x4.91 ksize = 11, 1024x1024, 8UC4 568.561 183.934 x3.09 ksize = 13, 1024x1024, 8UC4 570.085 237.825 x2.4 ksize = 15, 1024x1024, 8UC4 675.046 49.528 x13.6 ksize = 3, 2048x2048, 8UC4 1170.26 110.083 x10.6 ksize = 5, 2048x2048, 8UC4 1859.11 196.855 x9.44 ksize = 7, 2048x2048, 8UC4 1999.81 395.12 x5.06 ksize = 9, 2048x2048, 8UC4 1994.11 518.987 x3.84 ksize = 11, 2048x2048, 8UC4 2032.05 729.437 x2.79 ksize = 13, 2048x2048, 8UC4 2005.78 977.641 x2.05 ksize = 15, 2048x2048, 8UC4 pyrDown: 235.279 39.2804 x5.99 4000x4000, 8UC4 130.707 40.4058 x3.23 3000x3000, 8UC4 57.1578 13.7517 x4.16 2000x2000, 8UC4 13.6359 8.41556 x1.62 1000x1000, 8UC4 pyrUp: 263.64 89.4818 x2.95 2000x2000, 8UC4 65.418 16.3824 x3.99 1000x1000, 8UC4 equalizeHist: 3.23604 8.55969 x0.378 1000x1000 11.3514 20.3959 x0.557 2000x2000 25.293 36.5682 x0.692 3000x3000 Canny: 61.9438 47.8773 x1.29 reduce: 3.04195 1.15759 x2.63 1000x1000, dim = 0 2.69111 2.31457 x1.16 1000x1000, dim = 1 9.98761 3.39094 x2.95 2000x2000, dim = 0 9.66781 3.91534 x2.47 2000x2000, dim = 1 21.2211 8.76042 x2.42 3000x3000, dim = 0 20.7015 5.4969 x3.77 3000x3000, dim = 1 gemm: [error: The library was build without CUBLAS] 512x512 GoodFeaturesToTrack: 188.499 40.3744 x4.67 PyrLKOpticalFlow: 312.542 151.67 x2.06 1000 614.213 420.687 x1.46 2000 1177.81 697.847 x1.69 4000 1216.7 611.392 x1.99 8000 FarnebackOpticalFlow: 641.565 611.623 x1.05 dataset=rubberwhale, fastPyramids=0, useGaussianBlur=0 1182.49 363.342 x3.25 dataset=rubberwhale, fastPyramids=0, useGaussianBlur=1 640.733 170.038 x3.77 dataset=rubberwhale, fastPyramids=1, useGaussianBlur=0 1180.53 173.912 x6.79 dataset=rubberwhale, fastPyramids=1, useGaussianBlur=1 893.146 965.563 x0.925 dataset=basketball, fastPyramids=0, useGaussianBlur=0 1621.17 623.748 x2.6 dataset=basketball, fastPyramids=0, useGaussianBlur=1 896.168 747.02 x1.2 dataset=basketball, fastPyramids=1, useGaussianBlur=0 1655.28 312.789 x5.29 dataset=basketball, fastPyramids=1, useGaussianBlur=1 FGDStatModel: 513.208 257.048 x2 MOG: 96.1007 10.7464 x8.94 MOG2: 84.3416 13.3616 x6.31 average GPU speedup: x6.240
View Comments