Deep Learning hardware

Requirement1

  • Memory: $\gt$ 8G;
  • Graphics card: $\gt$ 1 unit CUDA
  • OS: Ubuntu 16.04??
  • PC23: 最好是台式机

Computing power

Solution 1: GPU4

  • High Budget GPU: Titan XP, Amazon, Price: $1,649.99
  • Medium Budget GPU: GeForce GTX 1080 Ti 8G, Amazon

8GB有点小,但对很多任务都足够了,比如足够应付Kaggle比赛里大多数图像数据集合自然语言理解(NLP)的任务。7

  • Medium Budget GPU: GeForce GTX 1060 6G, Amazon, Price: about $318.04
  • Small Budget GPU: GeForce GTX 1050 TI, Amazon, Price: $157.99

Solution 2: Cloud45

Using a cloud service is a good choice for getting started. But note that building a local deep learning rig does become cost effective if you need to train models for 1500+ hours. See Andrej Karpathy’s setup if you want to give it a try.5

  • AWS: AWS上的GPU比较慢(GTX 1080速度是AWS GPU的四倍)7

    • On demand
    • Reserved
    • Spot
  • Google Cloud

  • Floyd

Comparison: AWS vs. Google Cloud

  • AWS: 2 CPU, 8GB RAM, $69/Month, Pay per hour
  • Google Cloud: 2 CPU, 8GB RAM, $52/Month, Pay per minute

写在最后

购买前最好在pcpartpicker.com上列好清单,网站可以自动检查兼容性。8
初期,我打算先在AWS上测试模型9,计算一下时间复杂度,如果有必要再考虑是否自己搭建一台Local DL Machine吧。


1. 如何配置一台适用于深度学习的工作站?
2. 学习tensorflow,买什么笔记本好?
3. 15寸macbook pro如何使用CUDA对深度学习进行gpu加速?
4. YouTube|How to Train Your Models in the Cloud by Siraj Raval
5. What is best cloud solution for deep learning?
6. Local DL rig of Andrej Karpathy’s setup:List
7. 哪些GPU更适合深度学习和数据库?
8. 成本14,000元,如何自己动手搭建深度学习服务器?
9. 在AWS上配置深度学习主机
-------------End of postThanks for your time-------------
BDG飽蠹閣 wechat
Enjoy it? Subscribe to my blog by scanning my public wechat account