Shop OBEX P1 Docs P2 Docs Learn Events
Core envy ? — Parallax Forums

Core envy ?

The other day I got notification of what is a little bit old news now:

Parallella has actually managed to tape out a 1024 core processor.

https://www.parallella.org/2016/10/05/epiphany-v-a-1024-core-64-bit-risc-processor/

I'm not suggesting the Parallella is comparable to the Prop or P2, they are very different animals apart from using many cores.

The interesting thing is how cheaply they managed to do it. Also sometimes it's amazing what a Kickstarter campaign can lead to.

Comments

  • kwinnkwinn Posts: 8,697
    Heater. wrote: »
    The other day I got notification of what is a little bit old news now:

    Parallella has actually managed to tape out a 1024 core processor.

    https://www.parallella.org/2016/10/05/epiphany-v-a-1024-core-64-bit-risc-processor/

    I'm not suggesting the Parallella is comparable to the Prop or P2, they are very different animals apart from using many cores.

    The interesting thing is how cheaply they managed to do it. Also sometimes it's amazing what a Kickstarter campaign can lead to.

    It's amazing what a small dedicated group can do when they have a set goal and are not hampered by bureaucracy, marketing, and management. Also helps that this one was DARPA funded.
  • Heater what are the implications here, is it possible that what we thought was a supercomputer a few years ago, could sit on our desktop tomorrow ?

    I know having been into computers for a while, that it's what you accumulate. And that eventually makes the computer smarter, at least personally so, we hope.

    Is there any implications for the average computer user ? The world depends on a conection for information. How will data be accumulated for each individual, the same way I got it, "individual". If no one copies my database, it's just a loss.

    Point is: A computer is only as smart, as what you put into it.
  • Heater.Heater. Posts: 21,230
    @kwinn,
    ...Also helps that this one was DARPA funded.
    Yeah, that makes it sound like there was a bottomless pit of Yankee defense money available for this.

    But in this case it looks like the whole idea was to have some stringent limits on development cost in place.

    I have yet to determine what those limits were.

    @MikeDYur
    Heater what are the implications here, is it possible that what we thought was a supercomputer a few years ago, could sit on our desktop tomorrow
    I'm not sure what you mean by that. A couple of weeks ago I saw a Cray Super Computer, from 1980 something, in the science museum in Berlin. A machine somewhat slower than the phone in my pocket at the time.
    A computer is only as smart, as what you put into it
    Sure. The Parallella is all about maximizing the floating point math performance you can get per Watt. How "smart" that may end up being depends on how programmers can make use of it. Same as ever.




  • Heater. wrote: »
    @kwinn,
    ...Also helps that this one was DARPA funded.
    Yeah, that makes it sound like there was a bottomless pit of Yankee defense money available for this.

    But in this case it looks like the whole idea was to have some stringent limits on development cost in place.

    I have yet to determine what those limits were.

    @MikeDYur
    Heater what are the implications here, is it possible that what we thought was a supercomputer a few years ago, could sit on our desktop tomorrow
    I'm not sure what you mean by that. A couple of weeks ago I saw a Cray Super Computer, from 1980 something, in the science museum in Berlin. A machine somewhat slower than the phone in my pocket at the time.
    A computer is only as smart, as what you put into it
    Sure. The Parallella is all about maximizing the floating point math performance you can get per Watt. How "smart" that may end up being depends on how programmers can make use of it. Same as ever.



    Seems like it would be good at simulating neural networks.

  • jmgjmg Posts: 15,140
    Heater. wrote: »
    Parallella has actually managed to tape out a 1024 core processor.
    A couple of months to silicon...
    Claims 64MB RAM which sounds like a big number, until you divide by 1024, and get 64k/Chip.
    Still, you can just allocate multi-die 32 Cores is just 3% of the die, for ~2MB of shared RAM.
    Claims 64b Integer and 64b and 32b Floats.

    No price, but I wonder how this die split ratio compares with P2
    Function Value   (mmˆ2)   Share of Total Die Area
    SRAM              62.4    53.3%
    Register File     15.1    12.9%
    FPU               11.8    10.1%
    NOC               12.1    10.3%
    IO Logic           6.5     5.6%
    “Other” Core Stuff 5.1     4.4%
    IO Pads            3.9     3.3%
    Always on Logic    0.66    0.6%
    

    I'd guess P2 has more NOC/IO die share, if you include PAD Ring and Smart Pins & Cordic is less than FPU ?
  • Heater. wrote: »
    @kwinn,
    ...Also helps that this one was DARPA funded.
    Yeah, that makes it sound like there was a bottomless pit of Yankee defense money available for this.

    But in this case it looks like the whole idea was to have some stringent limits on development cost in place.

    I have yet to determine what those limits were.

    It's actually hard to get serious DARPA money unless you have something that is quite out of the box. The bigger question is what sort of stipulations came with the money, I've known a few interesting projects that went black the moment DARPA people saw something they liked.

    That said, Parallela is leaving quite a out of their PR. I don't think this beast will be aimed for public consumption for quite some time.
Sign In or Register to comment.