Shop OBEX P1 Docs P2 Docs Learn Events
Parallela project on Kickstarter — Parallax Forums

Parallela project on Kickstarter

AntoineDoinelAntoineDoinel Posts: 312
edited 2014-03-04 16:21 in General Discussion
Hello fellow forumists!
I'm participating in this Kickstart project:

http://www.kickstarter.com/projects/adapteva/parallella-a-supercomputer-for-everyone

At the moment funding projections are not looking very bright, unless more people participate. I thought somebody here on the Parallax forums might be interested as well.
It's $99 for a dual core ARM and a 16-core Epiphany III chip, programmable with GCC (both the ARM front end and the parallel array), and I don't see it as directly competing with the Propeller II: their 16-core chip would be specialized in floating point operations.

Alessandro


P.S.: I see that the topic was already brought to your attention:
http://forums.parallax.com/showthread.php?142734-64-cores-Kickstarter-page
I tried to search a few words before posting this, but didn't get any result.
Sorry for the inconvenience, moderators feel free to delete this thread if you think it's redundant.
«1

Comments

  • prof_brainoprof_braino Posts: 4,313
    edited 2012-10-20 07:03
    Also check the thread on the GA144

    http://forums.parallax.com/showthread.php?142915-Any-forth-users-interested....&p=1134618&viewfull=1#post1134618

    This is $10 for 144 cores, the schmart board kit is $34

    It appears we use the simple FORTH language to directly write "microcode for each core or group of cores. We're still studying what the heck one would do with such a toy, but I imagine we will be applying it to any vision processing bottle neck in a RPi / prop arrangement, etc. We're not in design phase yet, more like research phase at this point.
  • AntoineDoinelAntoineDoinel Posts: 312
    edited 2012-10-20 09:02
    prof_braino wrote: »
    Also check the thread on the GA144

    http://forums.parallax.com/showthread.php?142915-Any-forth-users-interested....&p=1134618&viewfull=1#post1134618

    This is $10 for 144 cores, the schmart board kit is $34

    It appears we use the simple FORTH language to directly write "microcode for each core or group of cores. We're still studying what the heck one would do with such a toy, but I imagine we will be applying it to any vision processing bottle neck in a RPi / prop arrangement, etc. We're not in design phase yet, more like research phase at this point.

    There are white papers about soft peripherals like 10Mbit ethernet, and external RAM with service nodes to redistribute it on demand of other nodes. But the core consuption is a bit higher than expected, IIRC 7 cores for ethernet and 5 for external memory.

    If I understood routing correctly, it implies participation of all the crossed nodes, I imagined it would be better if a regular area in the center could be reserved for pure computational work, so message passing does not affect external interfaces and vice versa.

    My reservation with the GA144 is that I have no longer any ability to work with FORTH, my only experience with it was 30 years ago, even before my first home computer... paper was the program storage, and I was the stack machine! :lol:

    But I seem to remember you FORTH guys had process migration from prop to prop already in place, and that was the first thing that came to my mind reading the GA144 docs: it would be cool if some common subset could run transparently in both cogs and cores from that array.
  • prof_brainoprof_braino Posts: 4,313
    edited 2012-10-20 18:25
    AntoineDoinel wrote: »
    There are white papers about soft peripherals like 10Mbit ethernet, and external RAM with service nodes to redistribute it on demand of other nodes. But the core consuption is a bit higher than expected, IIRC 7 cores for ethernet and 5 for external memory.

    If I understood routing correctly, it implies participation of all the crossed nodes, I imagined it would be better if a regular area in the center could be reserved for pure computational work, so message passing does not affect external interfaces and vice versa.

    My reservation with the GA144 is that I have no longer any ability to work with FORTH, my only experience with it was 30 years ago, even before my first home computer... paper was the program storage, and I was the stack machine! :lol:

    But I seem to remember you FORTH guys had process migration from prop to prop already in place, and that was the first thing that came to my mind reading the GA144 docs: it would be cool if some common subset could run transparently in both cogs and cores from that array.

    I didn't digest all the papers yet, but Sal is looking into it. But baisically we figure with 144 cores we have plenty to mess with. This isn't intended to replace a prop or RPi, rather to act as a tailored "coprocessor" to do some very specific set of functions to address our bottleneck (whatever that may be at the time.

    I also last worked with forth 30 years ago, now I only play with it. :)

    The prop to prop and the prop to RPi is that CSP channels-ish stuff that is also like GO-language channels. That stuff is single wire synchronous serial (2 wires for bi-directional). The GA144 has a bazillion lines, so something will likely be different for the GA144 interfaces (if this even happens). Since we appear to control the microcode (maybe) in the GA144 we might be able to find something that can to programmed that is helpful. Finding out who is the "we" that can program this is the next hurdle.
  • AntoineDoinelAntoineDoinel Posts: 312
    edited 2012-10-26 15:14
    Something incredible is happening with this project:
    http://www.kicktraq.com/projects/adapteva/parallella-a-supercomputer-for-everyone/#chart-daily

    Just two days ago it seemed hopeless :-O
  • prof_brainoprof_braino Posts: 4,313
    edited 2012-10-26 16:07
    I asked a co-worker to support this a few hours ago, it was $75k short, now it looks like it might make it!
  • David BetzDavid Betz Posts: 14,516
    edited 2012-10-26 18:07
    prof_braino wrote: »
    I asked a co-worker to support this a few hours ago, it was $75k short, now it looks like it might make it!
    Looks like it met its goal. These boards should be fun to play with!
  • Heater.Heater. Posts: 21,230
    edited 2012-10-26 23:50
    A surprising upturn towards the end there.

    I just spent the morning skimming though the architecture document. Amazing. Well the amazing thing is that it pretty much all staight forward and understandable. The processor nodes architecture and instruction set is simple to get your head around and the description of the mesh network is comprehendable.
    One interesting aspect is that every processing node has access to all of a 4GB address space. Accessing off core memory gets shunted through the network without any programmer intervention. Of course you want to keep your working data locally but having shared memory makes it much easier to distribute tasks and data.

    If they can make the proposed board for 99 dollars that's incrdible, it's worth it just for the dual ARM cores an other normal stuff on there:)
  • LeonLeon Posts: 7,620
    edited 2012-10-27 03:19
    I signed up for it just after it was announced and couldn't see them making their target a couple of weeks ago. It does look like all the publicity they have got in the last couple of days has done the trick.
  • LoopyBytelooseLoopyByteloose Posts: 12,537
    edited 2012-10-27 10:29
    They are way over their goal of $750,000, - a very interesting device. VERY NEAR TO CLOSING DEADLINE. HOURS TO GO...
  • rod1963rod1963 Posts: 752
    edited 2012-10-27 11:42
    It looks quite interesting. I can see uses in Neural Nets and Fuzzy Logic type applications right off the bat. I'll probably get one if it's under $200.00
  • LoopyBytelooseLoopyByteloose Posts: 12,537
    edited 2012-10-27 12:24
    It is $99 for the board and you still have a few hours to go if you really want one. It will be delivered in May of 2013 if all goes well.
  • Heater.Heater. Posts: 21,230
    edited 2012-10-27 12:41
    rod1963,

    Sign up now and it's only 99 dollars.

    Well, after reading the available architecture documents (which are very well done), watching the videos and taking a couple of beers I just put in a pledge. Looks like they reached their target already and perhaps don't need me. It's a little gamble and now I have something else to look forward to, like the Raspberry Pi before it, and the Prop II even now.
  • RickBRickB Posts: 395
    edited 2012-10-27 13:34
    For another take on this "supercomputer" hype, read this thread on AVR freaks.
    http://www.avrfreaks.net/index.php?name=PNphpBB2&file=viewtopic&t=126143
  • tingotingo Posts: 87
    edited 2012-10-27 15:11
    I pitched in. $99 for a completely open project was too tempting. :-) It willl be fun to play with.
  • Heater.Heater. Posts: 21,230
    edited 2012-10-27 16:04
    So, everyone who has pledged for this wonder device or just waiting in case it turns out to come true, what are you thinking of doing with it if you get one in your hands? And, as we are here, how does the Prop or Prop II fit into this picture?
    There must be a killer app for a ton of I/O matched to a ton of floating point processing power.
  • AntoineDoinelAntoineDoinel Posts: 312
    edited 2012-10-27 16:57
    Heater. wrote: »
    So, everyone who has pledged for this wonder device or just waiting in case it turns out to come true, what are you thinking of doing with it if you get one in your hands? And, as we are here, how does the Prop or Prop II fit into this picture?
    There must be a killer app for a ton of I/O matched to a ton of floating point processing power.

    Some ideas, more dreams that what I specifically intend to do (too many things are WAYS over my head):

    * Orbit propagation for Orbiter Space Flight simulator, working when the simulation is offline
    * Beamforming of audio signals (I know more or less how to implement spatial filtering, but not how to calculate "Direction Of Arrival")
    * "Retro" ray tracing? Retro because yes, I know a GPU would be faster but... similarly last time I did a rendering overnight with Sculpt Animate 4D on a vanilla 68000, and saving the animation on the second floppy disk, the 486DX2@66 was already out... so what? :lol:
    * Don't know anything about VST plugin programming, but I guess that that could be an interesting option, expecially considering the price of specialized hardware like stuff from TC-Helicon, Sonic-Core, etc.

    The board will have HDMI, so the Propeller II might not fit in for video. But in case of input from an array of microphones, it could be an option as the interface in place of separate ADCs+programmable logic, giving its analog capabilities per pin.

    The Propeller II would fit perfectly IMHO if somebody will design a board just with an Epiphany III (or IV) chip and a CPLD to speed up the link (i.e. the prop accessing it as a large parallel bus to get more bandwidth).
    In that case it would replace the ARM host processors, and if not on par on computing capabilities, sure it would make a simpler (and more deterministic) board for many dataflow audio and video applications.

    BTW, audio beamforming would be an interesting project on a bare Propeller 1 also, limited to an array of 4 microphones.
  • prof_brainoprof_braino Posts: 4,313
    edited 2012-10-27 22:13
    RickB wrote: »
    For another take on this "supercomputer" hype, read this thread on AVR freaks.
    http://www.avrfreaks.net/index.php?name=PNphpBB2&file=viewtopic&t=126143

    Some of those contributors seems very negative with little imagination. I'm glad they don't post here. Obviously a $99 prototype is not going to whomp the snot out a CRAY CX1000.

    But it will be cool to see how whether or not folks manage to get all the parallela cores saturated and working on the same problem. If I could get a grasp of that, I'd do so on the prop as well.
  • RickBRickB Posts: 395
    edited 2012-10-27 23:34
    No! It is the over the top over hyping that is being criticized.
  • LeonLeon Posts: 7,620
    edited 2012-10-28 00:31
    Just got this email:

    Congratulations!

    Thanks to you and 4960 other backers, Parallella: A Supercomputer For Everyone has been successfully funded. Amazon will now charge your credit card.

    Pledge Summary
    Amount pledged: $120.00
    Reward: MAKER: You get everything in the SUPPORTER reward and an Epiphany-III based Parallella board loaded with the development software needed to create your own innovative projects.(International order should add $20 to the pledge amount). We can't wait to see what you come up with.
    Estimated delivery: May 2013

    When your reward is ready, Adapteva will send you a survey via email to request any info needed to deliver your reward (mailing address, T-shirt size, etc).

    If you'd like to visit the project page, click here:
    http://www.kickstarter.com/projects/adapteva/parallella-a-supercomputer-for-everyone?ref=email

    KICKSTARTER

    They ended up with nearly $900k!
  • LoopyBytelooseLoopyByteloose Posts: 12,537
    edited 2012-10-28 02:03
    Well, I wanted one, but Amazon hung up on providing me with my password in order to contribute. But at least I have a confirmation of a CubieBoard. And I won't have to wait until May.

    I am not sure I could really handle a Parallela, but by name is it and Italian sister company of Parallax -- good luck and a bit of envy to all that got in line for one.

    Still waits for May 2013 I suppose. My Cubieboard has been in use for nearly a month now.
  • prof_brainoprof_braino Posts: 4,313
    edited 2012-10-28 06:54
    RickB wrote: »
    No! It is the over the top over hyping that is being criticized.

    Just like that Raspberry Pi hoax! What a crock! Neither of these things could ever remotely have a chance to really happen! :)
  • David BetzDavid Betz Posts: 14,516
    edited 2012-10-28 07:09
    prof_braino wrote: »
    Just like that Raspberry Pi hoax! What a crock! Neither of these things could ever remotely have a chance to really happen! :)
    They already have the Epiphany III 16 core chips. They just need to make the board. That doesn't seem so risky.
  • Heater.Heater. Posts: 21,230
    edited 2012-10-29 07:10
    I've just been looking over what I was rash enough to sign up for over the weekend.

    Seems Andreas Olofsson has taken something of a Chip Gracey approach to his processor design. Basically KISS. Start with a clean slate, forget what everyone else does and focus on what you want to do. In this case floating point crunching at low power. So, no CPU caches, no instruction reordering pipelines etc etc just take a floating point unit, wrap a simple RISC processor around it with enough smarts to support C programming. Clone that many times on your chip. Create a communication network between nodes, no channel protocol or message routing complications. Neighbor to Neighbor only connections. Just have the processor node ID's in the address space and then processor read/writes just propagate left/right and up/down until they find the right nodes memory space. Sounds simple, perhaps was a bit harder to do. Finally in Chip style, do all this design and verilog code on your own.

    All sounds like a fun new toy to play with. But there is more.

    The promised board has dual ARM cores on it and all the SD card, ethernet, serial etc interfaces you expect on an ARM board.

    Then there is the FPGA style programmable logic blocks, which means hours of fun toying with Verilog/VHDL to create your own custom dodads.

    All of this make the board, if it materializes, worth the price even if you never use the Parallella chip:)

    Enter the Prop II. This board has 21 free GPIO pins (42 if you for go HDMI). Unlike normal ARM GPIO pins though they are driven from the FPGA section which means you can put whatever fast hardware protocol on there you like.

    So now I see a Prop II daughter board for this, like the Propeller "Plates" for the Raspberry Pi. That would be the "Propeller Parallella" board, a bit of a mouth full:)
  • David BetzDavid Betz Posts: 14,516
    edited 2012-10-29 12:15
    Heater. wrote: »
    I've just been looking over what I was rash enough to sign up for over the weekend.

    Seems Andreas Olofsson has taken something of a Chip Gracey approach to his processor design. Basically KISS. Start with a clean slate, forget what everyone else does and focus on what you want to do. In this case floating point crunching at low power. So, no CPU caches, no instruction reordering pipelines etc etc just take a floating point unit, wrap a simple RISC processor around it with enough smarts to support C programming. Clone that many times on your chip. Create a communication network between nodes, no channel protocol or message routing complications. Neighbor to Neighbor only connections. Just have the processor node ID's in the address space and then processor read/writes just propagate left/right and up/down until they find the right nodes memory space. Sounds simple, perhaps was a bit harder to do. Finally in Chip style, do all this design and verilog code on your own.

    All sounds like a fun new toy to play with. But there is more.

    The promised board has dual ARM cores on it and all the SD card, ethernet, serial etc interfaces you expect on an ARM board.

    Then there is the FPGA style programmable logic blocks, which means hours of fun toying with Verilog/VHDL to create your own custom dodads.

    All of this make the board, if it materializes, worth the price even if you never use the Parallella chip:)

    Enter the Prop II. This board has 21 free GPIO pins (42 if you for go HDMI). Unlike normal ARM GPIO pins though they are driven from the FPGA section which means you can put whatever fast hardware protocol on there you like.

    So now I see a Prop II daughter board for this, like the Propeller "Plates" for the Raspberry Pi. That would be the "Propeller Parallella" board, a bit of a mouth full:)
    A Ephiphany/Propeller combo sounds interesting but doesn't the Epiphany have access to its I/O pins itself? I'll have to read the docs more in the next six months while I wait for my board to arrive.
  • Heater.Heater. Posts: 21,230
    edited 2012-10-29 12:41
    Oh yeah, we have time to ponder on this:)

    It seems the ARM cores get to talk to the Ephiphany via whatever is configured in the FPGA logics blocks. Which happen to be the same logic blocks that talk to the GPIO and hence Propeller.

    So potentially one can short circuit all of that and configure the logic blocks to connect the     Ephiphany directly to the Propeller. It's just a sea of logic after all, to do what you want with.

    However if you want to use the     Ephiphany dev tools of the ARM, I hope they run there and we don't need a PC, it's best to keep that path open.
  • David BetzDavid Betz Posts: 14,516
    edited 2012-10-29 12:54
    Heater. wrote: »
    Oh yeah, we have time to ponder on this:)

    It seems the ARM cores get to talk to the Ephiphany via whatever is configured in the FPGA logics blocks. Which happen to be the same logic blocks that talk to the GPIO and hence Propeller.

    So potentially one can short circuit all of that and configure the logic blocks to connect the     Ephiphany directly to the Propeller. It's just a sea of logic after all, to do what you want with.

    However if you want to use the     Ephiphany dev tools of the ARM, I hope they run there and we don't need a PC, it's best to keep that path open.





    Any idea how much "space" there is for configurable logic? Is this equivilent to what you'd find on a CPLD or a small FPGA? Does one use Verilog or VHDL to configure the programmable logic?
  • Heater.Heater. Posts: 21,230
    edited 2012-10-29 13:22
    No idea how many logic blocks you get there, must check the data sheets, but I did read somewhere that it is configurable using the same ISE tool used for other Xilinx FPGA's. I guess that does not run on the ARM though. However the idea seems to be that the ARM boot's and downloads the logic block config file to the logic blocks and away you go.
  • prof_brainoprof_braino Posts: 4,313
    edited 2012-10-29 13:24
    so, what do you see as an "ideal application" for this hardware?

    i haven't looked past the coolness factor, i'm hoping the smart people show me what to do with it.
  • David BetzDavid Betz Posts: 14,516
    edited 2012-10-29 13:29
    prof_braino wrote: »
    so, what do you see as an "ideal application" for this hardware?

    i haven't looked past the coolness factor, i'm hoping the smart people show me what to do with it.
    Okay, I'll admit it. This looks like a cool toy to me. I don't have any application in mind other than playing with it. On the other hand, that was my only application for the Propeller too. Both seem like interesting architectures, much more interesting than a bare ARM core. :-)

    I'd also like to play with the GreenArrays chip but I'm waiting for a ready-made development board that doesn't require SMT soldering and is affordable. I may be waiting a long time for that!
  • Heater.Heater. Posts: 21,230
    edited 2012-10-29 13:51
    prof_braino,

    "..ideal application..?" Who knows? I'm sure that getting into the world of parallel algorithms is not so simple. But consider, here on the forums at least two projects have been a success using the FFT on the Propeller. Something I would never imagine possible. Then we have the likes of Phil Pilgrim, if I remember correctly, building software defined radio with the Prop. Things like this would love a lot of fast calculating possibility tightly coupled to the I/O. So I see a situation that the Prop II does the high speed I/O tightly coupled to the Epiphany for the number crunching.
Sign In or Register to comment.