Hello\nI have a general question here for the ones who own their GPU.\nMy experience is only with competing and training models here in Kaggle, so I am using a cloud platform like Kaggle and Google collab.\nI may do my first project as a consulter, and I may be required to install and use locally a GPU.\nUsually, do you install the CUDA driver for the GPU on a Linux environment or Windows environment?\nHow complicated is it to install the environment and then use Jupyter notebooks or Pycharm with GPU/CUDA?
", "messages": [ { "id": "728768", "postDate": "01/25/2020 08:23:07", "content": "Hello\nI have a general question here for the ones who own their GPU.\nMy experience is only with competing and training models here in Kaggle, so I am using a cloud platform like Kaggle and Google collab.\nI may do my first project as a consulter, and I may be required to install and use locally a GPU.\nUsually, do you install the CUDA driver for the GPU on a Linux environment or Windows environment?\nHow complicated is it to install the environment and then use Jupyter notebooks or Pycharm with GPU/CUDA?
", "rawMarkdown": "Hello\nI have a general question here for the ones who own their GPU.\nMy experience is only with competing and training models here in Kaggle, so I am using a cloud platform like Kaggle and Google collab.\nI may do my first project as a consulter, and I may be required to install and use locally a GPU.\nUsually, do you install the CUDA driver for the GPU on a Linux environment or Windows environment?\nHow complicated is it to install the environment and then use Jupyter notebooks or Pycharm with GPU/CUDA?", "votes": null }, { "id": "728774", "postDate": "01/25/2020 08:32:36", "content": "I have done it in windows, and its not that complex. You would to install CUDA, CudaNN, and a relevant GPU supported library such as Tensorflow to use GPU's. You can develop locally and submit results to Kaggle through their API.
", "rawMarkdown": "I have done it in windows, and its not that complex. You would to install CUDA, CudaNN, and a relevant GPU supported library such as Tensorflow to use GPU's. You can develop locally and submit results to Kaggle through their API.", "votes": null }, { "id": "728778", "postDate": "01/25/2020 08:38:36", "content": "I've had a bit of both experiences: cloud and local, so here's my two cent (Note this is for TensorFlow):\nMost of the time, Linux makes it just that much easier to create environments and add packages.\nCUDA is a bit of a pain, but you can also make a docker image. This is much more simple in setting up, because you don't have to do all the CUDA related installs. It also works great just to make a container locally, I've done it a few times (and then ran it from PyCharm).
\n\nI'm not very familiar with using Docker and jupyter notebooks together, but I would tend to say that a local install is probably easier if you want to use notebooks. Also a local install is probably easier if you want to add other packages / libraries, where the docker image would need to be re-build each time, from what I understand.
\n\nYou've probably already seen this, but just in case, here's the GPU install link for TensorFlow. I've also just tinkered a little bit with PyTorch and didn't have as much issues as Tensorflow.
\n\nHope that helps!
", "rawMarkdown": "I've had a bit of both experiences: cloud and local, so here's my two cent (Note this is for TensorFlow):\nMost of the time, Linux makes it just that much easier to create environments and add packages.\nCUDA is a bit of a pain, but you can also make a [docker image](https://www.tensorflow.org/install/docker?hl=it). This is much more simple in setting up, because you don't have to do all the CUDA related installs. It also works great just to make a container locally, I've done it a few times (and then ran it from PyCharm).\n\nI'm not very familiar with using Docker and jupyter notebooks together, but I would tend to say that a local install is probably easier if you want to use notebooks. Also a local install is probably easier if you want to add other packages / libraries, where the docker image would need to be re-build each time, from what I understand.\n\nYou've probably already seen this, but just in case, here's the [GPU install link for TensorFlow](https://www.tensorflow.org/install/gpu?hl=it). I've also just tinkered a little bit with PyTorch and didn't have as much issues as Tensorflow.\n\nHope that helps!", "votes": null }, { "id": "728798", "postDate": "01/25/2020 09:39:36", "content": "One lifesaving trick is to use conda to install tensorflow and pytorch with GPU support. This way you only have to install drivers for your GPU. The cuda, cudnn, cuda-toolkit will be installed by conda package manager automatically. All this with the benefit of environment separation is cherry on top.
", "rawMarkdown": "One lifesaving trick is to use conda to install tensorflow and pytorch with GPU support. This way you only have to install drivers for your GPU. The cuda, cudnn, cuda-toolkit will be installed by conda package manager automatically. All this with the benefit of environment separation is cherry on top.", "votes": null }, { "id": "728839", "postDate": "01/25/2020 11:09:08", "content": "This might help you.
\n\nhttps://blog.slavv.com/the-1700-great-deep-learning-box-assembly-setup-and-benchmarks-148c5ebe6415 (Refer software setup)
", "rawMarkdown": "This might help you. \n\nhttps://blog.slavv.com/the-1700-great-deep-learning-box-assembly-setup-and-benchmarks-148c5ebe6415 (Refer software setup)", "votes": null } ], "comments": [ { "id": 728774, "author_name": "hassanamin", "author_url": "", "post_date": "01/25/2020 08:32:36", "content": "I have done it in windows, and its not that complex. You would to install CUDA, CudaNN, and a relevant GPU supported library such as Tensorflow to use GPU's. You can develop locally and submit results to Kaggle through their API.
", "votes": null, "replies": [] }, { "id": 728778, "author_name": "maxlenormand", "author_url": "", "post_date": "01/25/2020 08:38:36", "content": "I've had a bit of both experiences: cloud and local, so here's my two cent (Note this is for TensorFlow):\nMost of the time, Linux makes it just that much easier to create environments and add packages.\nCUDA is a bit of a pain, but you can also make a docker image. This is much more simple in setting up, because you don't have to do all the CUDA related installs. It also works great just to make a container locally, I've done it a few times (and then ran it from PyCharm).
\n\nI'm not very familiar with using Docker and jupyter notebooks together, but I would tend to say that a local install is probably easier if you want to use notebooks. Also a local install is probably easier if you want to add other packages / libraries, where the docker image would need to be re-build each time, from what I understand.
\n\nYou've probably already seen this, but just in case, here's the GPU install link for TensorFlow. I've also just tinkered a little bit with PyTorch and didn't have as much issues as Tensorflow.
\n\nHope that helps!
", "votes": null, "replies": [] }, { "id": 728798, "author_name": "dhananjay3", "author_url": "", "post_date": "01/25/2020 09:39:36", "content": "One lifesaving trick is to use conda to install tensorflow and pytorch with GPU support. This way you only have to install drivers for your GPU. The cuda, cudnn, cuda-toolkit will be installed by conda package manager automatically. All this with the benefit of environment separation is cherry on top.
", "votes": null, "replies": [] }, { "id": 728839, "author_name": "gyuv4r4j", "author_url": "", "post_date": "01/25/2020 11:09:08", "content": "This might help you.
\n\nhttps://blog.slavv.com/the-1700-great-deep-learning-box-assembly-setup-and-benchmarks-148c5ebe6415 (Refer software setup)
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "728768": "Hello\nI have a general question here for the ones who own their GPU.\nMy experience is only with competing and training models here in Kaggle, so I am using a cloud platform like Kaggle and Google collab.\nI may do my first project as a consulter, and I may be required to install and use locally a GPU.\nUsually, do you install the CUDA driver for the GPU on a Linux environment or Windows environment?\nHow complicated is it to install the environment and then use Jupyter notebooks or Pycharm with GPU/CUDA?", "728774": "I have done it in windows, and its not that complex. You would to install CUDA, CudaNN, and a relevant GPU supported library such as Tensorflow to use GPU's. You can develop locally and submit results to Kaggle through their API.", "728778": "I've had a bit of both experiences: cloud and local, so here's my two cent (Note this is for TensorFlow):\nMost of the time, Linux makes it just that much easier to create environments and add packages.\nCUDA is a bit of a pain, but you can also make a [docker image](https://www.tensorflow.org/install/docker?hl=it). This is much more simple in setting up, because you don't have to do all the CUDA related installs. It also works great just to make a container locally, I've done it a few times (and then ran it from PyCharm).\n\nI'm not very familiar with using Docker and jupyter notebooks together, but I would tend to say that a local install is probably easier if you want to use notebooks. Also a local install is probably easier if you want to add other packages / libraries, where the docker image would need to be re-build each time, from what I understand.\n\nYou've probably already seen this, but just in case, here's the [GPU install link for TensorFlow](https://www.tensorflow.org/install/gpu?hl=it). I've also just tinkered a little bit with PyTorch and didn't have as much issues as Tensorflow.\n\nHope that helps!", "728798": "One lifesaving trick is to use conda to install tensorflow and pytorch with GPU support. This way you only have to install drivers for your GPU. The cuda, cudnn, cuda-toolkit will be installed by conda package manager automatically. All this with the benefit of environment separation is cherry on top.", "728839": "This might help you. \n\nhttps://blog.slavv.com/the-1700-great-deep-learning-box-assembly-setup-and-benchmarks-148c5ebe6415 (Refer software setup)" }, "source": "meta" }