Personal View site logo
Make sure to join PV on Telegram or Facebook! Perfect to keep up with community on your smartphone.
Finally Launched: the Adapteva so called "supercomputer"
  • Update: See latest post or

    Semiconductor start-up Adapteva has launched a Kickstarter project with the aim of creating a 'supercomputer for everyone,' in the form of the Raspberry Pi-inspired Parallella board.

    Like the Raspberry Pi, which has been staggeringly successful since it launched despite a few hiccoughs along the way, the Parallella board packs an ARM processor onto a credit-card sized form-factor with a wodge of memory, general-purpose input-output capabilities, and various ports. Where the Parallella differs, however, is in its performance potential, with Adapteva claiming to achieve around 50GHz of CPU-equivalent general-purpose computing performance from its Epiphany co-processor.

    ...costing $99 for the 16-core version and a projected $199 for the 64-core version,...


  • 43 Replies sorted by
  • This sounds pretty bad, as it has very little real applications, so this guys went for Kisckstarter as they heard that they can get easy money.

    Problem with most ARM implementations are slow memory interface, quite small cache sizes. Some of existing ASICs already have GPU having 16 units.

    This thing, instead, ask you to invest significant money and time in developing one time software that can't be used on anything else and have questionable performance.

  • So I don't know the particular implications of this project in the world of nlves but I assume it could mix things up. Specs and details are summarized below:

    Parallella: A Supercomputer for Everyone
    Kickstarter Summary: Goal is to make parallel computing available to everyone via mass produced high performance computers inspired by the Rasberry Pi and Arduino design. If you happen to be unfamiliar with these two companies, they've been releasing very basic computers that are about the size of a portable HDD for 50 to 200 dollars. Adapteva (Parallella's developer) diverges only slightly from this m.o. but the result is inspired. They plan to release systems capable of (consumer level) parallel computing for a meagre 100 bucks. I am a bit skeptical but I'll include all the specs I comprehend or care about, and links to more comprehensive sources for your perusal.
    * Dual-core ARM A9 CPU
    * Epiphany Multicore Accelerator (16 or 64 cores)
    * Just 1 gig of ram (?)
    * Two "general purpose" expansion connectors
    * Micro SD/ HDMI/ Ethernet
    * Ships with Ubuntu
    * Ships with Adapteva's developer and programmer tools °

    °Serious note: From what I gather, a lot of development will have to be done to make parallella, and similar systems, into a viable option for video/graphics editing. I may be wrong, but I currently am under the impression that graphical NVLE's (or similar) applications have not been developed to exploit the capabilities offered by these systems.

    I will add more sources about this if people are interested, particularly about the implications to non-linear video editing systems and such, but I really don't know much about it. Just thought that it's a very interesting project and wanted to offer it up to the community for appraisal.

  • Looking at Adaptive history and CEO it looks like quite big question.

    They always overpromised and underdelivered.

    800 GHz Operating Frequency

    What is this shit doing on their site?

    This guys got that major players could improve fast GPUs in ARM based LSIs.

  • Makes me believe they're talking about a frequency band and accidentally hit 'm' instead of 'g.' I am clueless about this.

  • There's been a bunch of these type of diy supercomputers. I read a few years ago of somebody putting a bunch of PS3s together and making a nicely powered linux based supercomputer.

    Here's the new Rasberry Pi version, open source so you can do it now if you want.

    I'm still waiting to hear if mac minis will be able to connect through thunderbolt to form their own mini supercomputers.

  • @CRFilms

    Yep, just they all have very little practical use. Good for headlines, yes.

  • Hahah, that lego MPI system is so kool, yes rediculous but cool

  • HPCwire says Adapteva's long term plan is for volume-production of their Epiphany accelerator chip.

    The company is initially aiming Epiphany at embedded and mobile device applications with a need for lots of flops -- codes such as image recognition, signal processing, game physics, and the like. But since these same attributes of high flops and low power are also critical to supercomputing, the company is looking to break into the HPC business as well.

  • @goanna

    Fo to their site and read about architecture. It is wastly simplified approach made for some specific tasks and requirind very big work.

  • Sounds cool. I just contributed.

  • @Vitaliy_Kiselev
    I dunno - I'm just researching all this here - you seem to be hinting that this might all be an elaborate hoax.

    Bloomberg says:

    To date, Adapteva has collaborated with researchers at MIT, Boston University, Northeastern University and Halmstad University to help bring parallel computing technology to the next generation of engineers.

    Once the Parallella project is fully funded, Adapteva will also take the bold step of releasing its existing software development tools, drivers and libraries under a true open source license and will publish the Epiphany chip architecture reference manuals and datasheets publicly. Board design files and board support packages for the Parallella computer will be delivered to the public in open source format and available for everyone to use free of charge and without restrictions.

  • @brianluce .. Looking good so far almost $100K in less than one day

  • I have to agree with Vitaliy that investors have to be cautious.

    However, to see both sides:

    • Times are tough and credit is hard to get, even for Adapteva;

    • We'd like to see independent testing, of course - although commercial secrecy can prohibit that;

    • The list of testimonials would be enough if only they were qualitative - specifically referring to the technological aspects;

    • The Epiphany technology is obviously in Beta stage. A lot can still go wrong;

    • Finally, you can have a good product these days and still not find a big company to back it!

    Adapteva's own Epiphany test:

  • Does not have a good aproach since A9 arm will never be super computer Lol. The accelerators canot be super computer. Would be beter to have CP-GPU computing, like AMD in Trinity. Also nanometer wise, these ARM arquitecture are not good enough for Power to heatr disipation ratio.

    GPU-CPU computing is future, no ARM cortex cores alone. Those accelerators also 64 cores of what? ARM, Paralell computing like GPU? For me thech like this can only come from

    Intel, AMD/Ati / Nvidia, TSMC, and maybe ARM the european golden boy. Even china has developed a chip recently in 2012 (copy intel arquitecture)

    Ive seen so many techs fail in time for not having proper implementation of foundrys and resourses to develop new materials and stuff.

    So dont expect anything in lithography from this "say so" manufacturers.

  • Well, the reference to "supercomputer" is certainly a cheeky piece of sales-talk hyperbole.

    The Epiphany chip, "just like the bigger PCs will have, only you got it first," makes the Parallella a super computer about as much as wearing a "Just like the Astronauts wear" Seiko will make you travel to the moon.

    But I still expect it to accelerate an Dual-core ARM A9 CPU via parallel processing.

  • @endotoxic

    Check their site.

    It is not ARM. It is highly custom RISC cores with very specific RAM architecture making it unsuitable for most tasks. It fully lacks any cache in this parallel parts, btw.

  • Looking good so far almost $100K in less than one day

    It is not a problem with todays dumb copy&paste media spreading hype.

  • I dunno - I'm just researching all this here - you seem to be hinting that this might all be an elaborate hoax.

    It is not hoax, it just just product with company that do ot have any good record, seems to be in desperate need of money and company that is offering very doubtful product to masses who really do not understand a heck.

  • @CRFilms

    The "Supercomputers" you are referring to are render-farms. Very cheap and useful. Some companies in Sydney do outsourced mass-rendering for the film industry, particularly animation.

    You might have noticed the render-farm export option in Vegas and other NLE's.

  • @Vitaliy_Kiselev

    They state that it uses an A9, the epiphany provides the Parallel Processing .. do you think that is a mistake or typo?

  • @kavadni

    In no place they state that this is ARM cores (and it is impossiblle, btw).

    This is custom developed RISC cores suitable only for very thin number of tasks requiring floating point operations with very low memory requirements and very low power consumption requirements.

  • @Vitaliy_Kiselev I am getting confused then.

    I read this on the kickstarter page:

    The following list shows the major components planned for the Parallella computer:
    Dual-core ARM A9 CPU
    Epiphany Multicore Accelerator (16 or 64 cores)

    My simplistic understanding, the Epiphany is the fp chip for the arm

  • @kavadni

    Yep, you just mixing round and cold :-)

    Look at pictures at the links I provided. ARM A9 here is external CPU, and this "accelerator" is just connected to it.

  • @Vitality_Kiselev

    ARM : Advanced RISC Machine. They are RISC based, arm is a "complete" solution via hardware. Mainly they use their OS inside ROM from Special RAM as you said. They usually dont run complex OS like Mac or windows. Mainly uses Linux small distros for...lets say..Superkmarket money cashiers...Money expender in banks etc.

    So..lets put things simple.

    RISC is a prosessor architecture. Example..Was used in ALL machinotsh machines before intel era. The advantages of RISC arquitecture was it reduced instruction set. For example INTEL pentium prossesor used to had a pipeline were all prosesses where made of 34 steps. AMD original Athlon had on its counter part only 16 steps in the same x86 core, making it more efficient PER CLOCK. Thats why INTEL on that era was more prone to errors, since 36 steps por prosessing instruction and poor litography made too many mistakes, so reprosesing was shit, going back to cycle again.

    RISC processors are good, even today all ARM is based on that short instructions, BUT for todays mega complex software, they are more like not too good to use them, even if they have good litography, they also have more electric leaks when Nanometer prosses is smaller and smaller, since atoms "scape" from layer of copper and rare metals like indiun antimonite and shit like that.

    The only advantage of ARM as vitaly mentiones are its low power, and low requirements to function properly, but PLEASE dont belive super computers from this arquitecture since its not anymore the better one.

    I like RISC MORE than x86 since they are more efficient,but for today standars there should be a redesign in it, making it more programable, and less instruction fixed.