Share this project

Done

Share this project

Done
The Parallella project will make parallel computing accessible to everyone.
The Parallella project will make parallel computing accessible to everyone.
The Parallella project will make parallel computing accessible to everyone.
Created by
4,965 backers pledged $898,921 to help bring this project to life.

The Parallella Uncovered and Other Updates

In our careers,we have found that some engineers get very quiet when they don't work others get quiet when the do. We would like to think that we belong to the second group...

Over the last week we have put all of our energy (and then some) into putting together a compelling use case for non-programmers.  We finished this work Monday and spent all of today putting together a new pitch for the front page of the project. It will be posted tomorrow afternoon!!

Once it's up there we really need you help to cover the globe with Parallella news for the remaining 3 days of the campaign!!

In the meantime, here are more details about the Parallella board feature set.

Parallella Host CPU Introduction

Throughout the Parallella Kickstarter project we have described the Epiphany accelerator in great detail but we haven’t talked much about the dual-core ARM-A9 host processor on the board. Today we are officially announcing that the host processor is the Zynq7010 from Xilinx, a wonderful little device that combines a very capable CPU sub system with programmable logic inside a single low cost chip. For us, it was love at first sight! Finally an FPGA device with the right balance of logic gates, CPU horse power, and price point for our needs. The figure below shows how we are combining the Xilinx Zynq processor with our Epiphany accelerator chip to create a very disruptive platform. The Parallella board is truly a heterogeneous parallel computing platform that encourages programmers to use the most efficient resource to get the job done.

(The complete post can be found here:)

http://www.adapteva.com/white-papers/the-parallella-board-uncovered/

  • Image 173715 original

Comments

Only backers can post comments. Log In
    1. AJ on November 1, 2012

      Any chance of an upgrade option in the future to another ZYNQ? That chip is pretty popular already, zedboard is doing quite well, and the z-7020 shares a footprint option with the 7010 you've announced. That would put you ahead in price quite a bit, and help with the GPIO worries some are having. Honestly, I hope you offer the 7045 in a future board, perhaps when the 64 core epiphany hits mainstream production. I'd shell out cash for a couple of those.

    2. Missing avatar

      Walter van Holst on October 25, 2012

      Ok, but what would be the minimum number available under any circumstance? Also, will the VHDL for the communications messh in the FPGA be open source too? Alternatively, would it be thinkable to sacrifice the connection between the Epiphany and the ARM cores for a eSATA controller but to retain the Epiphany's connectivity with other boards in a cluster?

    3. Missing avatar

      Yann Vernier on October 24, 2012

      I did check the implementation once, but it was a long time ago (targeting Spartan 3 and XP2). It would take less logic cells on the Zynq as its LUTs are wider anyhow. Anyhow it is a perfect example of a job where the Epiphany could compete; the FPGA is simply too fine-grained when the algorithm was designed for CPUs. As a rule of thumb, graphics cards are fastest at bitcoin (but waste energy on unused memories) while FPGAs are a little more energy efficient. Beat one while matching the other, and you're a winner; match either, and there's reason to be proud. Likely you'll find the code fetches cost a bit (but not significantly) compared to GPUs since the job does not branch at all. The OpenCL bitcoin page links to measurements in Mhash per second/joule/dollar.

    4. Martin Tingeling Lindgren on October 24, 2012

      @Lamson: Haha! Thank you for that beautiful analogy! I totally get it! Really looking forward to seeing what this beauty will blossom into!

    5. Missing avatar

      Yann Vernier on October 24, 2012

      @Martin Tingeling Lindgren : yes, that FPGA can run a soft processor in its programmable logic. It simply won't be as efficient as a hard processor, and in this case it includes two rather advanced ones. The idea is that you often want a processor of some description so they just included one that will suffice for a lot of designs. In other Adapteva boards (EMEK), I believe there is a soft processor running the Epiphany architecture in an FPGA, a common technique for prototyping. It connects to a hardware Epiphany chip with more cores as well. For this design, they traded the architecture match for hard processors of high performance and reasonable pricing capable of running Linux.

    6. Adapteva Creator on October 24, 2012

      @Lamson Thank you for pointing out the difference between Intel and us. Every day, we feel the pain of trying to compete with them on 0.1% of their R&D budget. Still, customers care about results, not excuses. We are certainly trying achieve "David vs Goliath" type success, but it hasn't been easy...

      @Yann Thanks for sending the link to everyone. Have you checked the implementation? How many Xilinx logic gates does it take?

      @Martin Thanks for the support!

    7. Missing avatar

      Yann Vernier on October 24, 2012

      You might find it easier to port an OpenCL bitcoin miner: https://en.bitcoin.it/wiki/OpenCL_miner ... and set the epiphany to integer mode only, as all operations are 32-bit integer or bitwise logic.

    8. Missing avatar

      Lamson Nguyen on October 24, 2012

      @Martin Think of Intel and other major chip-makers' cpu's as the bosses you would encounter in any game: they're big, complex, and have many abilities at the cost of high energy costs and only doing an ok job at doing all types of data.

      The Ephiphany, on the other hand, can be likened to an army of minions, with each minion only being able to process one type of data, but can do so optimally. They also need a host cpu to tell them what to do. As mentioned before, just this one type of data is useful for many, many applications.

      Keep in mind that Adapteva only started with $2.5M if I remember correctly, which is pocket change compared to Intel. Of course, in time, these 'minions' will level up and things will only keep on getting better from here.

    9. Missing avatar

      Yann Vernier on October 24, 2012

      @Adapteva regarding bitcoin: https://github.com/progranism/Open-Source-FPGA-Bitcoin-Miner - the core of it is a double SHA-256 hashing (IIRC), which is then checked for a partial collission. Simple pipeline work with a Python controlling script.

    10. Martin Tingeling Lindgren on October 24, 2012

      I was planning to buy the new retina iMac sometime next year, but seeing as this nextgen tech is about to cannon loose I would much rather be a part of this development.

      I am not sure how to help here (other than pledging which I've done).. I know next to little about computer programming but I understand the logic behind language somehow. This may be rudimentary stuff but I have the impression that the challenge has been to kind of serial-wire processors to share their load?

      Also I'm curious as to why the little nugget can't run all the programmable logic and itself do the job the ARM processor would? Or would the epiphany accelerator would act like something of a motherboard chameleon, taking on all the tasks needed, right?

      Also! Will you get this program running without the 750,000$ pledged funding? Because this really, really needs to hit the market. Fast. Keep it up guys, I really have my fingers crossed for you. :)

    11. Adapteva Creator on October 24, 2012

      @Yann Do you have RTL code for bitmining that you can share with everyone?

    12. Adapteva Creator on October 24, 2012

      @Walter The logic fabric has 28k logic cells. We are still working on optimizing the epiphany interface logic and would rather not quote a number right now. There will be gates available for GPIO logic.

    13. Missing avatar

      Walter van Holst on October 24, 2012

      How many of the FPGA cells are occupied by logic needed for the interface with the Epiphany cores? Would it be thinkable to implement an eSATA controller in the remaining cells and use the GPIOs for that purpose?

    14. Adapteva Creator on October 24, 2012

      @All Of course it would have been great to disclose as much information as possible before the start of the project. Once this project is funded, we will tell the whole story of how it unfolded.

    15. Missing avatar

      Yann Vernier on October 24, 2012

      The Xilinx ISE WebPACK tools (available gratis for download) for logic development do cover this chip (Zynq 7010, 7020 and 7030 supported). So assuming a decent path for JTAG or similar is on the board, we'll be able to reprogram the logic as well. Considering a Zynq development board is $895, the Parallella certainly doesn't look expensive.

    16. Missing avatar

      Yann Vernier on October 24, 2012

      It is able to do bitmining; it would have even been a simple application to port. And if you'd gone out with RTL source (after all, it's simply interface code now, unlike your earlier board with the softcore epiphany) and the Zynq presence, that alone would easily have sold the board at this price point. But with three days, I think it will be hard to nearly double the supporter count. :(
      @Chris Naeger: there is no GPU as such; but the FPGA is quite capable of doing framebuffer readout and outputting an HDMI signal. A (current day) GPU is a massively multithreaded processor with its own memory controller and video output capability, so we essentially see the parts laid bare in this case (epiphany multiprocessor, Zynq hard memory controller, Zynq FPGA-based video output).

    17. Missing avatar

      Walter van Holst on October 24, 2012

      I concur with those who think this should have been available at day 1 of the kickstarter. And if you would be able to do bitmining you would have the bitcoin community fighting to chip in and the stretch goals would have been met already. I hope there is a plan B in which some VC steps in for the remainder.

    18. Missing avatar

      alxxG on October 24, 2012

      This is what you guys should have had before starting the kickstarter.

      If it doesn't make it (there's still time to get over the line but looking unlikely) , maybe having a few application kits or base software builds like sdr, audio effects may help get another kickstarter over the line. Have a small addon hardware board for each kit.

      e.g for audio effects have a guitar in , line in, line out and headphones out. maybe a mic in as well.
      or just a stereo in and out

      sponsor a version of the fun cube dongle to use Epiphany

      @Beau if you stick with standard vhdl or verilog , it'll work on both a and x (and the others). Just need to separate out the dcm/clock generator code and any a or x specific hardware wrappers.

    19. Beau Webber on October 24, 2012

      Hmm, It looks as though your hardware will talk to my hardware ! At fast enough rates.
      I am a happy bunny.
      I will have to learn how to program Xilinx firmware as well as Altera, but worth it.
      I wish you the best, I want this project to fly. I will go re-count my pennies.

    20. Chris Naeger on October 24, 2012

      Please excuse my lack of knowledge, but shouldn't there be a Graphics chip somewhere between the CPU and the HDMI output? Or is the GPU built into the FPGA?

    21. Missing avatar

      notzed on October 24, 2012

      An update at last - good-o - i was starting to fear you'd had a melt-down over the lack of preparation. Regardless, I hope you manage to make it somehow or other - this is a very important device, particularly the `open' nature - i'm sick of buying a machine with some cool chip only to be effectively locked out of half the silicon due to 'eye pee' issues. Android or such is a big step up from appliances as you can customise the application software, but it still isn't enough.

      Although it looks tricky to code for at the lowest level, libraries can abstract that away - and it looks devilishly fun in any event.

    22. Frank Buss on October 24, 2012

      @Adapteva I guess it would help, if you could confirm in the next update, that all development tools are free for it. Would be bad for many backers, if they have to buy the $4k Xilinx ISE version.

      But in general the decision for this chip sounds great (maybe a bit late to attract more people). The tightly coupled combination of a general purpose ARM core, a FPGA and a high speed parallel computing platform allows a lot of applications. Maybe you could write a bit more about possible applications? I can think of a powerful gaming platform, with realtime physics emulation and 3D graphics, or realtime image processing for robots, or with some ADCs and DACs connected, a sophisticated multi channel signal generator and analyzer with FFT etc. for a spectrum analyzer, or software defined radio, or maybe even something wild, like implementing a phased array radar :-)

      Potential new users (and the media) might not imagine what is possible with it, so would be good to describe some cool applications, not just one use-case for non-programmers, but which is of course great, too.

    23. Brooks Moses
      Superbacker
      on October 24, 2012

      @jh298: I believe that Xilinx's zero-cost "WebPack edition" of their software should work fine for programming this FPGA -- that's what I'm planning to use for the Zynq 7020 on the ZedBoard I got recently. (http://www.xilinx.com/products/design-tools/ise-design-suite/ise-webpack.htm for details.)

    24. Adapteva Creator on October 23, 2012

      @jh298

      1.) Access to FPGA programming tools will have to be done through Xilinx or one of their distribution channels.

      2.) Yes, that makes sense! Most of the smaller IPs at openCores could probably be embedded within the Zynq programmable logic on the Parallella board.

    25. Missing avatar

      jh298 on October 23, 2012

      @Adapteva Two questions:

      1) Will the board have licensing for the tools we need to use the FPGA?

      2) Does it make sense collaborate with the http://opencores.org/ community to position this Kickstarter project as a "future supported" development platform?

    26. Steven Rogers on October 23, 2012

      This is looking better all the time. I'm not matching Lansom, but I've increased my pledge.

    27. Missing avatar

      Lamson Nguyen on October 23, 2012

      I am a poor college student who managed to scrap enough money to pledge $975 to this world-changing project. However, because of my financial status, I have been unable to secure a loan to back at the $10000 level though I would love to. So, to those not on a shoestring budget, this project, and ultimately, the world needs you, desperately! And if that's not enough, do you really want to be out-pledged by poor, little me? ;)

    28. Missing avatar

      Paul T. Shackett on October 23, 2012

      This is very excellent! I was hoping to use an FPGA with the Epiphany for this exact thing! Develop some VHDL hardware algorithyms and see them fly through those numbers like it was a direct feed...
      Excellent choice and very well thought out product! I am in AWE...
      Now PLEASE get more backers!!!

    29. Missing avatar

      Tincman on October 23, 2012

      This certainly answered some questions I had, thanks for the update! Definitely very excited!

    30. Paul D on October 23, 2012

      Thank you for the update and I look forward to seeing the information for non-programmers tomorrow. Fingers crossed it will be enough to get this wonderful project across the line.