Chapter 15 Heterogeneous Computing

1. 15.1 What is heterogeneous computing? 361

2. 15.2 What is a GPU? 361

3. 15.3 Introduction to GPU Architecture 361

4. 15.3.1 Why use a GPU?

5. 15.3.2 What is the core of CUDA?

6. 15.3.3 What is the role of the tensor core in the new Turing architecture for deep learning?

7. 15.3.4 What is the connection between GPU memory architecture and application performance?

8. 15.4 CUDA framework

9. 15.4.1 Is it difficult to do CUDA programming?

10. 15.4.2 cuDNN

11. 15.5 GPU hardware environment configuration recommendation

12. 15.5.1 GPU Main Performance Indicators

13. 15.5.2 Purchase Proposal

14. 15.6 Software Environment Construction

15. 15.6.1 Operating System Selection

16. 15.6.2 Is the native installation still using docker?

17. 15.6.3 GPU Driver Issues

18. 15.7 Frame Selection

19. 15.7.1 Comparison of mainstream frameworks

20. 15.7.2 Framework details

21. 15.7.3 Which frameworks are friendly to the deployment environment?

22. 15.7.4 How to choose the framework of the mobile platform?

23. 15.8 Other

24. 15.8.1 Configuration of a Multi-GPU Environment

25. 15.8.2 Is it possible to distribute training?

26. 15.8.3 Can I train or deploy a model in a SPARK environment?

27. 15.8.4 How to further optimize performance?

28. 15.8.5 What is the difference between TPU and GPU?

29. 15.8.6 What is the impact of future quantum computing on AI technology such as deep learning?

P