Deep Learning on Local or Cloud

Deep learning requires high computation power to process data. Once your Data models are ready then you are ready to host the model either in local system or in cloud solution

Deep Learning local Desktop Configuration

In case you want to setup your local Desktop as Deep learning system. Please remember do not invest must on CPU but other parts are important. Specially RAM and GPU both are important.

No special hardware is required for the deep learning Machine setup. But some component do require higher configuration.

CPU: CPU is less important in deep learning. Core i3 or Core i7 will be enough for the purpose. 

Motherboard: Motherboard can be considered important factor. But while purchasing motherboard you need to check some parameters

  • It should support huge amount of RAM, let’s say at about 24Gb+ RAM
  • Graphics card slot it should support latest GPU (Graphics card) PCIe 3.0
  • It should support DDR4 RAM
  • Support for SSD good

RAM: DDR4 is must. At minimum level it should be DDR3. 1600 to 3000 FSB should be good enough. At minimum 16GB is required for deep learning.

GPU: For Deep Learning most important hardware is GPU. Here you have 2 options one is Nvidia and other one is AMD

  • Titan XP (High Budget)
  • Nvidia GTX 1060 (Mid Budget)
  • Nvidia GTX 1050 TI (Tight Budget)

Hard Drive: 256GB SSD or 2TB+ HDD in same budget will be fine.

Installing OS and Other Software Packages

  • CoreOS (Because Docker has out of Box in CoreOS)
  • Ubuntu (My 2nd Preference)
  • Install SSH Server
    sudo apt-get install openssh-server
  • CUDA (For GPU usage)
  • cuDNN (By Nvidia)
  • Anaconda (Python for Data Science)
  • OpenBLAS (Linear Algebra)
  • TensorFlow ( Google’s Machine Learning Algo)
  • Theano (Alternative to TensorFlow)
  • Keras (Helper tool for TensorFlow and Theano)
  • Torch
  • Jupyter
  • facebook CF2

 

Deep Learning Cloud Configuration

Nasuni_cover_pic

There are 4 top runner in this area

Deep learning on cloud is a cheaper option as compare to dedicated Machine. With $10 to $40 per month, you can find a good configuration machine. In my suggestion, if you want to setup application for production, you should opt one Cloud, but initially you can go for local Machine opt.

  • Amazon AWS
  • Google Cloud
  • Microsoft Azure
  • Floyd (This is based on AWS) – In case you are new or you are in early stage. this will be good
  • Nervanasys

Just in-case you want to know bit more about AWS, Google Cloud and Azure