MacTuxLin

OctaProp - A 64-Core Project (8 x Prop1)

Rate this Entry
...
The first time I came across this idea was when I attended a user group in Osaka, Japan about this time last year (2012). It was an invitation by an active forum member here. I must say, each of the member are highly talented & knowledgeable, so much so I was rather hesitant to show my projects

Anyway, they were talking about a board with 8 Propellers that worked out 64 cogs. I thought it was a fantastic idea but was wondering who would need a board with more than 2 Prop1. I created schematics with 8 Props but that much was it as I didn't know what it ought to achieve. So, placed it in the back-burner until end of last year. I started working on problems that requires to implement machine learning algorithms & this project crossed my mind. So I continued drafting out its design.... I intend to release the final design (Gerber file) once I've completed for anyone who would be interested to fabricate it.

Schematics: Attached Updated as of 19th March 2013

Objective:
In training large datasets on neural network algorithms, the dataset could be sliced into smaller segments & dish out to each Prop (7 possible cogs) to process using the same algorithms & parsing them back to the Master Prop. The board would worked out to consume only 6 watts when all 8 Props are active.


Design:
Master Prop
:- 6 x 32KB SRAM
:- microSD socket with DAT1 connected to support future features (Tubular designed peripherals on microSD that utilises the 5th pin)
:- 2 x CD4067 multiplexers to boot Client Props & performs high-speed communicates (Beau's code)
:- Controls powering ON/OFF individual Client Props
:- FT232RL USB Comm

Each Client Prop (7 x Props)
:- 1 x 32KB SRAM

Operations:
The Master Prop will communicate with host either in real-time mode via FT232RL or TTL to fetch/return algorithms, datasets & results or in batch-mode by reading/writing them from/to microSD. Master Prop will read each Client Prop on its microSD, boot them & load them data before moving on to the next Client. Since Master Prop can power ON/OFF each Client, you could basically controls how many Client Props you would need.

Questions:
1. Board form-factor: Since there are so many available pins on the Clients, what type of form-factor should this adopt? How about castellations with all components placed on top while having the board the flexibility to either solder with headers or mount on a larger board?

2. ...More to come...



Attached Thumbnails Attached Files

Submit "OctaProp - A 64-Core Project (8 x Prop1)" to Digg Submit "OctaProp - A 64-Core Project (8 x Prop1)" to del.icio.us Submit "OctaProp - A 64-Core Project (8 x Prop1)" to StumbleUpon Submit "OctaProp - A 64-Core Project (8 x Prop1)" to Google

Updated 03-19-2013 at 03:01 PM by MacTuxLin

Categories
Propeller , ‎ Programming , ‎ Engineering , ‎ DIY and Hacks , ‎ Art and Display

Comments

  1. Mike Green's Avatar
    I would suggest a total of 9 Props, 8 clients and 1 master. The master Prop would have several cogs supporting the various I/O devices. By having 8 clients, you'd have 64 cogs available and you could expand the functionality of the master Prop including various debugging tools. You'd have 8 x 2 = 16 I/O pins dedicated to the 8 clients' TX / RX lines. You don't need to power the client Props on and off because, if a client is held in reset or simply allowed to finish its boot sequence without a program being executed, its power requirements drop massively. That leaves 4 I/O pins for the SD card, 2 I/O pins for the usual I2C bus, 2 I/O pins for the USB adapter, and 8 I/O pins for other I/O like SPI SRAMs. You'd use some kind of I2C I/O Expander to drive the 8 /RST lines.

    Oh yeah ... use the new 128K byte SRAMs instead of 32K SRAMs. They're not that more expensive and give you 4x the space. You might also consider putting a 128K EEPROM on the master Prop instead of the usual 32K EEPROM
  2. MacTuxLin's Avatar
    Hi Mike,

    Thank you very much for your advice. Yeap, I'm gonna be updating the schematics over the weekends !!!
  3. MacTuxLin's Avatar
    OK, I've basically completed the schematics (Updated Attachment above) but I'll need to chew on it for a while before I starts the board layout.

    Feel free to drop a comments and/or suggestions.

    Thanks
  4. MacTuxLin's Avatar
    Continue with parts placement today. With so little personal time, I think I'll need another week to get it done ....