Shop OBEX P1 Docs P2 Docs Learn Events
$99 Supercomputer? — Parallax Forums

$99 Supercomputer?

HumanoidoHumanoido Posts: 5,770
edited 2013-07-08 00:41 in General Discussion
http://www.eejournal.com/archives/articles/20121003-adapteva/

http://www.adapteva.com/products/epiphany-ip/epiphany-architecture-ip/

http://www.adapteva.com/products/eval-kits/parallella/

"The Epiphany multicore coprocessor is a scalable shared memory architecture, featuring up to 4,096 processors on a single chip connected through a high-bandwidth on-chip network."

Comments

  • Heater.Heater. Posts: 21,230
    edited 2013-07-07 03:42
    Yep. I put 100 dollars into the kickstarter for this. I figured the ARM processor plus FPGA on that board was worth the investment even if the parallel floating point chip did not work out for me.
    I hope it's arriving soon.
  • LeonLeon Posts: 7,620
    edited 2013-07-07 03:54
    So did I.

    They will be shipping by the end of August, according to the latest update.
  • Heater.Heater. Posts: 21,230
    edited 2013-07-07 04:02
    Yeah, Leon, we're going to blow the socks off the Big Brain with these:)
  • mindrobotsmindrobots Posts: 6,506
    edited 2013-07-07 07:08
    Heater. wrote: »
    Yeah, Leon, we're going to blow the socks off the Big Brain with these:)

    Dang it! I missed the BLOG update that the Big Brain was now wearing socks!! Do you have a link to it? :lol:
  • Heater.Heater. Posts: 21,230
    edited 2013-07-07 07:46
    The Big Brain has been able to dress itself and tie its own shoe laces for a long time now:)
  • prof_brainoprof_braino Posts: 4,313
    edited 2013-07-07 10:32
    I backed it too.
  • localrogerlocalroger Posts: 3,451
    edited 2013-07-07 13:28
    There are very few problems for which this kind of floating point performance is justified. In most cases if you can't solve your problem with integer math, you don't understand your problem.
  • prof_brainoprof_braino Posts: 4,313
    edited 2013-07-07 16:38
    localroger wrote: »
    There are very few problems for which this kind of floating point performance is justified. In most cases if you can't solve your problem with integer math, you don't understand your problem.

    Both these statements arer true. But neither is a reason NOT to have insane float ability if its cheap and fast! :)
  • ercoerco Posts: 20,256
    edited 2013-07-07 18:23
    Very cool. It's funny though. Historically, it's neither the fastest or best product that succeeds, but the one with the best business plan. This drives me mad, but it's been proved often enough. Might be hard to get millions to changes horses midstream in this case.
  • Martin_HMartin_H Posts: 4,051
    edited 2013-07-07 19:19
    localroger wrote: »
    There are very few problems for which this kind of floating point performance is justified. In most cases if you can't solve your problem with integer math, you don't understand your problem.

    My Propeller pet peeve is that there's no good fixed point trig library as F32's performance is better.

    On a computer with floating point in hardware there's no reason to avoid using it. Your trig libraries will expect floating point arguments and return floating point results. So why fight city hall?
  • HumanoidoHumanoido Posts: 5,770
    edited 2013-07-07 21:08
    The Big Brain wanted to order multiples of the Epiphany "4,096 processors" chip because it can add 40 million enhanced processors to the collective. I held back the Brain (no easy task), telling it to keep the faith and hold out another 2 months for the prop 2.
  • Heater.Heater. Posts: 21,230
    edited 2013-07-08 00:41
    Humanoid,

    The Big Brain has time. Current Parallella chips are only 16 or 64 parallel floating point units. The one we will be getting for the kickstarter backing are only 16.
    4096 FPU's is the maximum for a single chip that the architecture supports I believe.
Sign In or Register to comment.