Utilizing the Raspberry Pi GPU Horsepower

Doing a little research on special purpose computation, I found it is possible to reprogram the Raspberry Pi GPU to do some powerful calculations.



The Raspberry Pi 12 cores GPU (more properly called Quad Processing Units (QPUs) ) it has theoretical max performance of 24 GPFLOS

   Theres an advanced book titled:
   Raspberry Pi GPU Audio Video Programming by Jan Newmarch
   in which he states: "The GPU is capable of 1 Gpixel/s, 1.5 Gtexel/s, or 24 GFLOPs of general-purpose compute and features a bunch of texture filtering and DMA infrastructure"
  

Here's a list of related articles and code libraries:

Broadcom released documentation on their proprietary GPU some time ago, the file is no longer available on their site, some of the material below may be based on that dated material.

   GPU Accelerated Object Recognition on Raspberry Pi 3 & Raspberry Pi Zero (not source so far)

  
   Hacking The GPU For Fun And Profit (Pt. 1)  2014
  
 
  Python library for GPGPU on Raspberry Pi - work in progress (includes GPU technical insights)
 
 
  OpenCL implementation running on the VideoCore IV GPU of the Raspberry Pi models

 
  GPU_FFT
 
 
  QPULib - A language and compiler for the Raspberry Pi GPU
 
 
  VC4ASM - macro assembler for Broadcom VideoCore IV aka Raspberry Pi GPU
  Docs Link

 
  Raspberry Pi VideoCore APIs
 
 
  Here's a 40 minute presentation to a bunch of engineers
  Hacking the Raspberry Pi's VideoCore IV GPU - Louis Howe
 

 
 
 
  - - - - -
 
  TensorFlow 1.9 Officially Supports the Raspberry Pi  Aug 2, 2018

 
  Awesome Deep Vision - A curated list of deep learning resources 2017

 
 

Comments

Popular posts from this blog

Hooking up the 6 + 1 Mic Array to the MaiX BiT

Holographic Fans Getting Real Time HDMI Input

LED Driver Roundup