{
  "id": 22013,
  "title": "GPU-accelerated Theano & Keras on Windows 10 native",
  "url": "/competitions/state-farm-distracted-driver-detection/discussion/22013",
  "author_name": "",
  "post_date": "2016-07-02T21:49:38.170Z",
  "votes": 1,
  "comment_count": 1,
  "views": 669,
  "content": "<p>If you still need help setting up Theano &amp; Keras with GPU acceleration on Windows 10 in <strong>native mode</strong> (no VMs, no Docker) -- the following may help:</p>\n\n<p><a href=\"https://github.com/philferriere/dlwin\">https://github.com/philferriere/dlwin</a></p>\n\n<p>For best results, you'll want <strong>both CUDA and cuDNN</strong> (hugely improved convnet perf!)</p>\n\n<p>We tested it on the following hardware:</p>\n\n<ul>\n<li>Dell Precision T7500, 96BG RAM [Intel Xeon E5605 @ 2.13 GHz (2\nprocessors, 8 cores total)]</li>\n<li>NVIDIA GeForce Titan X, 12GB RAM [Driver version: 10.18.13.5390 Beta (ForceWare 353.90) / Win 10 64]</li>\n</ul>\n\n<p>We used the following libraries:</p>\n\n<ul>\n<li>Visual Studio 2013 Community Edition Update 4 [Used for its C/C++ compiler (not its IDE)]</li>\n<li>CUDA 7.5.18 (64-bit) [Used for its GPU math libraries, card driver, and CUDA compiler]</li>\n<li>MinGW-w64 (5.3.0) [Used for its Unix-like compiler and build tools (g++/gcc, make...) for Windows]</li>\n<li>Anaconda (64-bit) w. Python 2.7 (Anaconda2-4.1.0) [A Python distro that gives us NumPy, SciPy, and other scientific libraries]</li>\n<li>Theano 0.8.2 [Used to evaluate mathematical expressions on multi-dimensional arrays]</li>\n<li>Keras 1.0.5 [Used for deep learning on top of Theano]</li>\n<li>OpenBLAS 0.2.14 (Optional) [Used for its CPU-optimized implementation of many linear algebra operations]</li>\n<li>cuDNN v5 (Recommended) [Used to run vastly faster convolution neural networks]</li>\n</ul>\n\n<p>Hope this helps!</p>",
  "messages": [
    {
      "id": "125805",
      "postDate": "07/02/2016 21:49:38",
      "content": "<p>If you still need help setting up Theano &amp; Keras with GPU acceleration on Windows 10 in <strong>native mode</strong> (no VMs, no Docker) -- the following may help:</p>\n\n<p><a href=\"https://github.com/philferriere/dlwin\">https://github.com/philferriere/dlwin</a></p>\n\n<p>For best results, you'll want <strong>both CUDA and cuDNN</strong> (hugely improved convnet perf!)</p>\n\n<p>We tested it on the following hardware:</p>\n\n<ul>\n<li>Dell Precision T7500, 96BG RAM [Intel Xeon E5605 @ 2.13 GHz (2\nprocessors, 8 cores total)]</li>\n<li>NVIDIA GeForce Titan X, 12GB RAM [Driver version: 10.18.13.5390 Beta (ForceWare 353.90) / Win 10 64]</li>\n</ul>\n\n<p>We used the following libraries:</p>\n\n<ul>\n<li>Visual Studio 2013 Community Edition Update 4 [Used for its C/C++ compiler (not its IDE)]</li>\n<li>CUDA 7.5.18 (64-bit) [Used for its GPU math libraries, card driver, and CUDA compiler]</li>\n<li>MinGW-w64 (5.3.0) [Used for its Unix-like compiler and build tools (g++/gcc, make...) for Windows]</li>\n<li>Anaconda (64-bit) w. Python 2.7 (Anaconda2-4.1.0) [A Python distro that gives us NumPy, SciPy, and other scientific libraries]</li>\n<li>Theano 0.8.2 [Used to evaluate mathematical expressions on multi-dimensional arrays]</li>\n<li>Keras 1.0.5 [Used for deep learning on top of Theano]</li>\n<li>OpenBLAS 0.2.14 (Optional) [Used for its CPU-optimized implementation of many linear algebra operations]</li>\n<li>cuDNN v5 (Recommended) [Used to run vastly faster convolution neural networks]</li>\n</ul>\n\n<p>Hope this helps!</p>",
      "rawMarkdown": "If you still need help setting up Theano & Keras with GPU acceleration on Windows 10 in **native mode** (no VMs, no Docker) -- the following may help:\r\n\r\nhttps://github.com/philferriere/dlwin\r\n\r\nFor best results, you'll want **both CUDA and cuDNN** (hugely improved convnet perf!)\r\n\r\nWe tested it on the following hardware:\r\n\r\n - Dell Precision T7500, 96BG RAM [Intel Xeon E5605 @ 2.13 GHz (2\r\n   processors, 8 cores total)]\r\n - NVIDIA GeForce Titan X, 12GB RAM [Driver version: 10.18.13.5390 Beta (ForceWare 353.90) / Win 10 64]\r\n\r\nWe used the following libraries:\r\n\r\n- Visual Studio 2013 Community Edition Update 4 [Used for its C/C++ compiler (not its IDE)]\r\n- CUDA 7.5.18 (64-bit) [Used for its GPU math libraries, card driver, and CUDA compiler]\r\n- MinGW-w64 (5.3.0) [Used for its Unix-like compiler and build tools (g++/gcc, make...) for Windows]\r\n- Anaconda (64-bit) w. Python 2.7 (Anaconda2-4.1.0) [A Python distro that gives us NumPy, SciPy, and other scientific libraries]\r\n- Theano 0.8.2 [Used to evaluate mathematical expressions on multi-dimensional arrays]\r\n- Keras 1.0.5 [Used for deep learning on top of Theano]\r\n- OpenBLAS 0.2.14 (Optional) [Used for its CPU-optimized implementation of many linear algebra operations]\r\n- cuDNN v5 (Recommended) [Used to run vastly faster convolution neural networks]\r\n\r\nHope this helps!",
      "votes": null
    },
    {
      "id": "125872",
      "postDate": "07/03/2016 21:56:15",
      "content": "<p>I have naive question, If it is home computer, why not to install Linux as second operation system and follow one of many available tutorials about how to set up everything?  </p>",
      "rawMarkdown": "I have naive question, If it is home computer, why not to install Linux as second operation system and follow one of many available tutorials about how to set up everything?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 125872,
      "author_name": "iglovikov",
      "author_url": "",
      "post_date": "07/03/2016 21:56:15",
      "content": "<p>I have naive question, If it is home computer, why not to install Linux as second operation system and follow one of many available tutorials about how to set up everything?  </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "125805": "If you still need help setting up Theano & Keras with GPU acceleration on Windows 10 in **native mode** (no VMs, no Docker) -- the following may help:\r\n\r\nhttps://github.com/philferriere/dlwin\r\n\r\nFor best results, you'll want **both CUDA and cuDNN** (hugely improved convnet perf!)\r\n\r\nWe tested it on the following hardware:\r\n\r\n - Dell Precision T7500, 96BG RAM [Intel Xeon E5605 @ 2.13 GHz (2\r\n   processors, 8 cores total)]\r\n - NVIDIA GeForce Titan X, 12GB RAM [Driver version: 10.18.13.5390 Beta (ForceWare 353.90) / Win 10 64]\r\n\r\nWe used the following libraries:\r\n\r\n- Visual Studio 2013 Community Edition Update 4 [Used for its C/C++ compiler (not its IDE)]\r\n- CUDA 7.5.18 (64-bit) [Used for its GPU math libraries, card driver, and CUDA compiler]\r\n- MinGW-w64 (5.3.0) [Used for its Unix-like compiler and build tools (g++/gcc, make...) for Windows]\r\n- Anaconda (64-bit) w. Python 2.7 (Anaconda2-4.1.0) [A Python distro that gives us NumPy, SciPy, and other scientific libraries]\r\n- Theano 0.8.2 [Used to evaluate mathematical expressions on multi-dimensional arrays]\r\n- Keras 1.0.5 [Used for deep learning on top of Theano]\r\n- OpenBLAS 0.2.14 (Optional) [Used for its CPU-optimized implementation of many linear algebra operations]\r\n- cuDNN v5 (Recommended) [Used to run vastly faster convolution neural networks]\r\n\r\nHope this helps!",
    "125872": "I have naive question, If it is home computer, why not to install Linux as second operation system and follow one of many available tutorials about how to set up everything?"
  },
  "source": "meta"
}