Shop OBEX P1 Docs P2 Docs Learn Events
FYI - a 64x processor — Parallax Forums

FYI - a 64x processor

HarleyHarley Posts: 997
edited 2007-08-21 16:13 in Propeller 1
Just for info, Tilera, www.tilera.com has a multiprocessor in an 8 x 8 matrix. NOT A DIP PACKAGE, for sure.

I'm not suggesting putting aside your Propellers in any way. Just FYI....

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Harley Shanko
h.a.s. designn

Comments

  • BeanBean Posts: 8,129
    edited 2007-08-20 15:25
    Interesting... Any word on prices ?
    Bean.

    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    Teacher: What is the difference between ignorance and apathy ?
    Student: I don't know and I don't care
    Teacher: Correct !
    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    www.hittconsulting.com
    ·
  • scottascotta Posts: 168
    edited 2007-08-20 15:32
    I hope that Parallax is using top-notch IC layout software.

    Anyone know of the latest on the second generation chip (prop-2) ?

    Scott
  • Beau SchwabeBeau Schwabe Posts: 6,568
    edited 2007-08-20 15:56
    scotta,

    The layout software that we are using is called "Laker" from Silicon Canvas.· It has the look feel, and smell of Cadence's Virtuoso Layout Editor·,·... the software·used·at National Semiconductor.


    Currently I am putting together the I/O PADS

    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    Beau Schwabe

    IC Layout Engineer
    Parallax, Inc.
  • AleAle Posts: 2,363
    edited 2007-08-20 16:20
    That sounds really nice, both the soft and the job, I mean designing and laying out ics. Sadly I am a chemist. so, not that much place _yet_ at semiconductor companies.

    We are struggling to keep pace with 8 cores at 20 MIPS, those 64 run at +600 !... that needs serous programming...
  • hippyhippy Posts: 1,981
    edited 2007-08-20 16:54
    Bean (Hitt Consulting) said...
    Interesting... Any word on prices ?
    Bean.
    435 USD each in 10K quantities, from reports I've seen.
  • SteelSteel Posts: 313
    edited 2007-08-20 17:03
    I will definately not put my propeller away at that price.
  • Beau SchwabeBeau Schwabe Posts: 6,568
    edited 2007-08-20 17:15
    Ale,
    ·
    Just wait until·Moore's Law·catches up to the physical size of the atom and we will need your chemistry.
    We have a little ways to go, but in the current process (180nm) we are using·uses·aluminum wires.· The
    diameter of an Aluminum nucleus is about 7.2fm (femto meters)
    ·
    micro is 1x10-6
    nano is· 1x10-9
    pico is 1x10-12
    femto is 1x10-15
    ·
    So for example If I'm running a wire that is 180 nm (nano meters) wide, then that's·25 million Aluminum Atoms wide.
    ·
    Other processes that use Copper for example, Copper·has a diameter of about·9.6fm·, so for a 180nm wire that's about
    18.75·million Copper Atoms wide.
    ·
    25 million or 18 million may seem like a lot, but consider less than 7 years ago where 360nm was popular.· Respectively
    your wires then would be 50 million and 36 million atoms wide.· Now it's even lower... 90nm, 65nm, even 30nm processes
    are available.

    Other nasty quantum quirks will·creep up way before we get to single atom wire widths though.


    Reference: 
Nucleus radius calculator   [url=http://hyperphysics.phy-astr.gsu.edu/hbase/hframe.html]http://hyperphysics.phy-astr.gsu.edu/hbase/hframe.html[/url][url=http://hyperphysics.phy-astr.gsu.edu/hbase/hframe.html][/url]  - selcet Nuclear Physics, then select Nuclear Size
Periodic Chart              [url=http://www.webelements.com/]http://www.webelements.com/[/url] 
SI prefixes                 [url=http://en.wikipedia.org/wiki/Femto]http://en.wikipedia.org/wiki/Femto[/url]

    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    Beau Schwabe

    IC Layout Engineer
    Parallax, Inc.

    Post Edited (Beau Schwabe (Parallax)) : 8/20/2007 5:25:43 PM GMT
  • HarleyHarley Posts: 997
    edited 2007-08-20 17:29
    Beau said...
    Currently I am putting together the I/O PADS
    Thank you for the 'atomic' information. Interesting the things an IC designer must consider. Quite different from 'simple pcb layout' task.

    Would it be too far a leap, if working on the I/O pads portion, that the major parts of the Prop2 design is well along (complete) in the design/layout process? Or am I just wishing?

    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    Harley Shanko
    h.a.s. designn
  • Beau SchwabeBeau Schwabe Posts: 6,568
    edited 2007-08-20 17:41
    Harley,
    ·
    Certain "core" elements are about 90% or more complete.··The physical size of the IO pad will determine how much total real estate is available within the chip, and thus determine how some of the upper level guts of the·Propeller·are floor planned.· Aside from that I can't comment.
    ·


    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    Beau Schwabe

    IC Layout Engineer
    Parallax, Inc.
  • HarleyHarley Posts: 997
    edited 2007-08-20 17:56
    Beau,

    Thank you for the comments. Having done many pcb layouts, one can 'feel' for some of the IC layout details, but only some.

    Understandable why you cannot comment further on progress.

    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    Harley Shanko
    h.a.s. designn
  • deSilvadeSilva Posts: 2,967
    edited 2007-08-20 18:21
    hippy said...
    435 USD each in 10K quantities, from reports I've seen.
    This is as expected quite low cost compared to the Prop!

    Each processor = 30x faster(?) than a COG, 8 x more processors = 240 x Prop

    A similarly priced INTEL XEON 4x 2,6 GHz sports 10 GHz with 16 MB fast Cache!
    - Clock = 10 GHz/640 Mhz = 20 x Prop, but consider FP and SIMD units!
    - RAM = 16 MB / 48 k = 333 x Prop
  • rjo_rjo_ Posts: 1,825
    edited 2007-08-20 18:40
    Beau,

    There is no hurry... the wait simply flattens the playing field.

    About your comments to Ale.

    You are assuming that future computers will use wires to connect logic elements.

    Moores Law is in no danger. Before we fully arrive at the atomic scale, we will pass through the molecular scale. There are biologic molecules capable of all known modes of transduction... (and many that we don't know yet[noparse]:)[/noparse] Imagine molecular FETs that require no physical connection... only defined topologic relationships, which can be both determinant and self-assembling.

    We won't use electricity to operate molecular computers... we will use sugar and light.

    And we won't have to build the first molecular computers. They already exist. We simply need to "extract" them from nature.

    For every circuit there is a biologic equivalent.

    Rich

    If you can "think it," "it" exists somewhere in your brain. Extracting "it" is the hard part[noparse]:)[/noparse]
  • MightorMightor Posts: 338
    edited 2007-08-20 18:55
    rjo_ said...

    We won't use electricity to operate molecular computers... we will use sugar and light.

    And we won't have to build the first molecular computers. They already exist. We simply need to "extract" them from nature.

    For every circuit there is a biologic equivalent.
    I don't want my prop to die because I forgot to feed it its daily sugar cube.

    Gr,
    Mightor

    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    | To know recursion, you must first know recursion.
  • rjo_rjo_ Posts: 1,825
    edited 2007-08-20 19:00
    When the Prop goes biologic... you will need to hide your sugar[noparse]:)[/noparse]
  • AleAle Posts: 2,363
    edited 2007-08-20 19:06
    Dear Beau,

    A nucleus can may be ~10 fm, but the actual distance from atom to atom is much larger at around 236 pm (Al) (222 Si), so the actual amount of atoms across ... is a bit smaller... 180000 / 236 = ~762 smile.gif

    Anyways... very en lighting !, Thanks !!
  • scottascotta Posts: 168
    edited 2007-08-20 20:38
    Beau,

    Make each IO point as complicated as possible!

    Analog IO,Digital IO,...
  • deSilvadeSilva Posts: 2,967
    edited 2007-08-20 20:49
    ... switchable Pull-Up...
  • LeonLeon Posts: 7,620
    edited 2007-08-20 22:18
    hippy said...
    Bean (Hitt Consulting) said...
    Interesting... Any word on prices ?
    Bean.
    435 USD each in 10K quantities, from reports I've seen.
    Not too bad! I paid £750 (about $2,000) each 20 years ago for T800 Inmos transputers.

    Leon


    ▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
    Amateur radio callsign: G1HSM
    Suzuki SV1000S motorcycle
  • Chuck DavisChuck Davis Posts: 23
    edited 2007-08-21 00:07
    Considering their relationship with Ubicom on the SX, I'm suprised that Parallax didn't do something with their other processors.

    The 8-thread 3000 series processor with 256K of program space and 64k of data ram, running at 250 mhz would have made a nice system comparable to the Propeller, and with more built-in hardware and memory.· Not sure what it costs, though.
  • deSilvadeSilva Posts: 2,967
    edited 2007-08-21 06:02
    Leon said...
    ..I paid about $2,000 each 20 years ago for T800 Inmos transputers.
    The "Price Decay Timescale" for VLSI chips is about 1.6 years, i.e. 12 periods ago = 1 / 4096. They should cost 50 cents now, plus package and handling smile.gif
  • ericballericball Posts: 774
    edited 2007-08-21 16:03
    ARS has a short article about the TILE64 http://arstechnica.com/articles/paedia/cpu/MIT-startup-raises-multicore-bar-with-new-64-core-CPU.ars

    IMHO this chip is a beast, both from a processing power perspective and a programming perspective.· I'm sure it will excel at tasks which can be parallized to fit it's mesh network.· But I doubt programmers will be able to effectively use each core independently like you would on a Prop.· So you'll have stuff where >75% of the cores are executing identical code with the rest of the cores shuffling control data to & from the outside world.

    The price point is also very high.· Yeah, compared to current PC CPUs and GPUs it's competitive, especially on a processing power perspective.· But consider the cost versus embeded and consumer electronics platforms where $200-$500 is the price point for the entire device.

    ·
  • deSilvadeSilva Posts: 2,967
    edited 2007-08-21 16:13
    Prices scale exteremely well in terms of MIPS and memory, e.g.

    10 MIPS (8-bit) $1 (AVR/SX/PIC)
    160 MIPS (32-bit) $16 (Prop)
    1.000 MIPS (32-bit) $100 (AMD/Intel PC processor)

    You can be sure to get an equivalent raw processing power for whatever price.
Sign In or Register to comment.