GPU-CPU Memory Transfer

Authors

• Alonzo, Jose Anton
• Avelino, Joris Gabriel
• Cruz, Airon John
• Hernandez, Pierre Vincent

About the project

CPU and GPU are two separate processors; each has its memory. There are several ways in which data are transferred between processors, as follows:

• Unified memory introduced in CUDA 6
• Prefetching of data with memory advice
• Data transfer or initialization as a CUDA kernel
• Old method of transferring data between CPU and memory (memCUDA copy)

Discuss:

• Explain the concepts of each method and how it is implemented
• Compare (a) the execution time of the kernel as well as (b) the memory transfer time (taking into consideration page faults, overhead time if it’s a separate transfer time, or part of the kernel transfer time already).

Files

• CEPARCO S11_G2_GPU-CPU Memory Transfer_Data and Analysis.pdf
  • PDF file containing the data results and analysis for the GPU-CPU Memory transfer
• [CEPARCO]Group2_GPUCPUMemTransfer.ipynb
  • Notebook containing the source code implementation for each memory transfer method
• CUDA_Unified_Memory_Research.pdf
  • Researches on Unified Memory memory transfers

Important Links

• YouTube video presentation: https://youtu.be/pRprKLqauH4
• Google Colab link [View access only]: https://colab.research.google.com/drive/1fouudeoADH6v3W_3lIBvbTXAjzG6wS0_?usp=sharing
  • NOTE: Must be a part of the DLSU organzation when accessing the Google Colab link