The Parallella Computer 38


Parallella_board_3d_1_1000x475

The Parallella platform is based on the Epiphany multicore chips developed by Adapteva over the last 5 years and field tested since May 2011. The Epiphany chips consists of a scalable array of simple RISC processors programmable in C/C++ connected together with a fast on chip network within a single shared memory architecture. Here is a link to the Epiphany Architecture Reference Manual

Overview:

  • Zynq-7000 Series Dual-core ARM A9 CPU (Z-7010 or Z-7020)
  • 16 or 64-core Epiphany Multicore Accelerator
  • 1GB RAM
  • MicroSD Card
  • 2x USB 2.0
  • 4 general purpose expansion connectors
  • 10/100/1000 Ethernet
  • HDMI port
  • Linux Operating System
  • 3.4″ x 2.15″ form factor

Once completed, the 66-core version of the Parallella computer would deliver over 90 GFLOPS on a board the size of a credit card while consuming only 5 Watts under typical work loads. For certain applications, this would provide more raw performance than a high end server costing thousands of dollars and consuming 400W.

To get an idea just how powerful this little board will be, check out benchmark scores for the Epiphany-IV and Epiphany-III processors at coremark.org or read the Adapteva blog post.

For detailed information see:

Pricing and Availability

Reserve a board using sign up form on the right. We will notify you as soon as order entry begins.


38 thoughts on “The Parallella Computer

  • Reply
    Sohail Alam

    Dear Sir/Madam,

    This looks like an interesting opportunity for us to include your product in our project.
    Please do let us know when the starter kit / development kit is ready for market.
    Also kindly let us know where to get the SDK software for the same.

    Thank you,
    Sohail Alam
    Rancore Technologies Pvt Ltd

  • Reply
    TOG

    Hi,

    just wanted to let you know that udacity.com has now an intro classroom to parallel computing programming (using CUDA). Maybe they are interested in an alternative platform, as CUDA is not open source (as far as I know).

    Kind regards,
    Steffen

    • Reply
      Frederik

      Prolly, if you make a haskell compiler for it. The point is that there already exists a C++ (and a C) compiler which will allow you to compile directly for the epiphany instruction set. This ofc means that there is an assembler which is what you will want your haskell compiler to output instructions for, making it much easier than having to write a whole compiler.

      • Reply
        GreyGeek

        Out of curiosity I checked the Kubuntu repository for Haskell apps and found the “ghc”, a Haskell compiler.
        “The Glorious Glasgow Haskell Compilation system (GHC) is a compiler for Haskell.” and dozens of more Haskell apps, docs, tools, etc,, including
        “This package provides a library for the Haskell programming language. See http://www.haskell.org/ for more information on Haskell.
        Provides a library for parallel programming in Haskell.”

    • Reply
      Jimmymmy

      I’m not exactly sure how you would make a PCIe edition of a Single-Board Computer.

      If you’re talking about the epiphany chips being used as a form of on-board computing accelerator, then I could see that happening.

  • Reply
    Julian Guarin

    I was a backer in KS, my question is I would like to buy a Epiphany IV (besides the Epiphay I’ll get for backing you guys up) When it would be possible, and how much is it going to cost?

    Great Job Guys
    Julian

  • Reply
    Frederik

    You keep saying that the board will consume only 5 watt under typical workload — what’s the maximum? And what is “typical workload”? My desktop at home has very varied workload, from barely idling to full pedal to the metal, while my server has pretty consistent (rather low) workload. Both could be considered “typical workloads”.

    Also, what’s the minimum? i.e. what kind of power does it draw when just running a kernel, doing nothing worth anything?

  • Reply
    kenclassmakerritchie

    Any chance USB port can be upgraded to USB 3.0? USB 3.0 consumes far less power, and my go up to 5 Gbps. What about Thunderbolt, to 10 Gbp? Cheers!

  • Reply
    Anonymous

    Is Java supported on the Epiphany cores ??

    I see that there is a java implementation for ARM based systems (in fact raspberry has one) so i assume that it should be only a matter of some “apt-get install” to have them working on the main cpu…

    but how do i offload some of my processing to the epiphany cores ??
    Are there special classes or interfaces to take advantage of the specialized cores?

  • Reply
    dc

    Is Java supported on the Epiphany cores ??

    I see that there is a java implementation for ARM based systems (in fact raspberry has one) so i assume that it should be only a matter of some “apt-get install” to have them working on the main cpu…

    but how do i offload some of my processing to the epiphany cores ??
    Are there special classes or interfaces to take advantage of the specialized cores?

  • Reply
    Amithlon

    I specifically want the unit for folding at home and TheSkynet , and some video conversion (which is way to slow on my i7-2600K @4.8GHz – with GTX670) . could be just the ticket….. nice for bragging rights too…

  • Reply
    Ivan G

    It looks like the Zynq-7020 boards are not yet available on the online shop. Is it a openCL backend for Epiphany ready or planned? If not, It could be a very interesting project … :-)

  • Reply
    Dan

    Would it be possible to turn a handful of parallel boards into a render cluster for 3D applications like Blender, or would there be a problem because of the ARM architecture?

  • Reply
    na4

    Hi,
    Would it be possible to program Parallela in FORTRAN? I would need it for a tough simulation problem.

  • Reply
    Anonymous

    I’m very interested about the board. Can we plug multiple webcams using usb? Which webcam models are supported?

  • Reply
    Crusader ky

    I don’t understand how global memory works. The Epiphany itself has only local memory, so whenever you want to access global memory you must do Epiphany -> FPGA -> CPU -> RAM -> CPU -> FPGA -> Epiphany?!? Am I missing something?

  • Reply
    Germmare

    really ? –> Once completed, the 66-core version of the Parallella computer would deliver over 90 GFLOPS on a ……

  • Reply
    Nelson

    If you are intelligent and wish to rapidly build a massive support and buyer base, you should contact the authors of blender (blender.org), and of the blender and mikumikudance external rendering engines, and get their support to produce compatible versions for your board.

    The video rendering performance increase would be truly spectacular, and both MMD and Blender have immense user bases. The MMD user base is much larger, especially in Japan, but would require you to produce a PCI card or USB device, and MMD (Windows 32 it app) would have to be modified to detect the device and use it if it exists. Keep up the great work.

    Hollywood will surely love banks of these for CGI rendering too..

Leave a Reply Using Facebook or Twitter Account