5 ms·

Parallella: Raspberry Pi-like open parallel computing hardware

110 points by JonasH 14y ago

AceJohnny2 14y ago
What's wrong with the Parallax Propeller? You can get a starter kit for $25 [1]. Granted, you used to have to program it in its own language, but C compilers are popping up [2] [1] http://adafruit.com/products/791 http://adafruit.com/products/791 [2] http://propeller.wikispaces.com/Programming+in+C+-+Catalina http://propeller.wikispaces.com/Programming+in+C+-+Catalina
- wtracy 14y ago
  The Propeller doesn't have sixteen cores running at 2+ GHz with a gig of ram. The Propeller can't even run a real, modern OS.
  - ErrantX 14y ago
    The Propeller can't even run a real, modern OS Well... technically neither can this :) Although they do appear to have got it working really neatly with an ARM-based host board. Which is a step forward.
wtracy 14y ago
Has anyone here worked with Epiphany chips before? Are there any special APIs required to use them effectively? How hard would it be to get Haskell code running on one?
- JonasH 14y ago
  There is a C/C++ SDK available.
  - wtracy 14y ago
    Is the SDK publicly available right now?
    - cookiemonster44 14y ago
      http://www.businesswire.com/news/home/20120822005093/en/Adapteva-Announces-Availability-OpenCL-SDK-Epiphany-Multicore http://www.businesswire.com/news/home/20120822005093/en/Adap...
ricardobeat 14y ago
The "Making parallel computing easy" and "The goal of the Parallella project is to democratize access to parallel computing" lines made me immediately think of the Parallax Propeller. Turns out their product isn't remotely similar: up to 64 cores at 800mhz, plus a dual-core ARM CPU. I wonder why they are doing a kickstarter instead of harvesting their dollars from server appliances.
- blacksmythe 14y ago
  Because they are not ready to harvest dollars from server appliances, and would like someone to fund their engineering.
backprojection 14y ago
This sounds great, but why is it a kick starter project? If the claims they're making are reasonable, it seems to me that, for instance, Google would be more than a little interested, and could easily cough up $750K. What am I missing?
- ChuckMcM 14y ago
  This is a really good question. Not necessarily Google, they are famous for not investing in other people to do engineering, but it seems like a modest amount of money to get something that should be fairly widely applicable. Of course if it is widely applicable and this investment gets to company what it needs to take off, well the folks who gave them the money aren't really going to benefit in a leveraged way. (No equity) Perhaps they were hoping for a 'raspberry pi' like response (which would be hundreds of thousands of units) and be able to do a sort of stealth funding round kinda thing. No idea of course, but it would be a sweet result if it worked out for them right?
- wmf 14y ago
  Perhaps the claims are not reasonable. Their site has a pretty high ratio of [citation needed] per page.
- SamuelKillin 14y ago
  I'd say they're looking for exposure/excitement in the hobbiest/startup datamining space. Partly marketing play, partly validation. I backed $99. I'm very excited.
- mng2 14y ago
  I also find it strange that a chip company is out on Kickstarter. From their website, though, apparently they started on an initial $2 million, which is peanuts for this industry. On some level I think it's an opportunistic marketing move, but it's possible that it was driven by necessity. It's an interesting concept. If it's cheap enough, it's definitely useful as a DSP. But the custom instruction set is going to keep it away from general usage. They seem to be hoping to get their IP into a cellphone, but that's a seriously uphill battle.
iyulaev 14y ago
Hasn't the whole massive multicore thing been floating around the server space for a few years now? I remember HP Project Moonshot and Sea Micro and a few other projects to built multi-chip ARM servers. However, I don't recall seeing ANY benchmarks that demonstrated that they were any more efficient, per watt, in a REAL application, than the x86 competition. I would really like to see such a data point. I guess now it's being sold as a novelty to let people play with such a technology, which is fine I guess. The CISC vs. RISC days are long over and the battle between the two architectures is a bit silly at this point since the gap between the instruction set and the underlying implementation has gotten quite dramatic. Claims that RISC chips are inherently more efficient may have been true in 1995, but I don't see this holding water today.
- wmf 14y ago
  For Linpack it has been demonstrated that GPUs and MIC are more energy-efficient than server x86, although Blue Gene beats them all. http://www.green500.org/lists/green201206&green500from=1&green500to=100 http://www.green500.org/lists/green201206&green500from=1...
  - iyulaev 14y ago
    I'm certainly not going to argue against GPUs (or FPGAs) being used in scientific applications where performance is limited by floating point or vector performance. In certain scientific applications they can hand general-purpose CPUs their butts. I'm talking about ARM/RISC vs x86 CPUs for server applications.
    - wmf 14y ago
      Sorry, I went on a tangent there. The benefits of wimpy cores have certainly been elusive. Some good links http://static.googleusercontent.com/external_content/untrusted_dlcp/research.google.com/fr//pubs/archive/36448.pdf http://static.googleusercontent.com/external_content/untrust... http://research.microsoft.com/en-us/um/people/trishulc/papers/low_power_cores.pdf http://research.microsoft.com/en-us/um/people/trishulc/paper... http://www.seamicro.com/sites/default/files/MozillaCaseStudy.pdf http://www.seamicro.com/sites/default/files/MozillaCaseStudy...
malandrew 14y ago
Sounds awesome, but I have a few comments I'm going to braindump: $750k seems really ambitious for a kickstarter project with such a niche audience. Is that realistic for 29 days, considering you need to sell 7500 units of the $99 pledge amount? At the end of day 7, Leapmotion had 15,000 applications for a free leap motion device and SDK, and that is a device with a much larger audience. Do you have any investor lined up that would be willing to maybe match a Kickstarter total pledge amount of something realistic like $375k? Have you considered approaching a fund like In-Q-Tel? This seems like the kind of project they would fund, since I imagine a lot of the best parallel computing work is being done in government-funded agencies and labs. I also imagine the government is probably the biggest employer of people working on parallel processing devices. With that in mind, getting a device like this into the hands of many, allows more people to get hands on exposure to parallel computing. Overall, it feels like the funding strategy needs to be diversified, because I imagine it will be difficult to get $750k all from one source, with the exception of a VC fund whose thesis aligns with your goals. Lastly, it feels like a project like this would be a bit too soon. Many developers who are playing with hardware have been playing around with the arduino for a few years, some are now graduating to the Raspberry Pi, which offers clear benefits over the Arduino because you can run tons of stuff simply not possible on the Arduino. However as the Raspberry Pi just came out, I imagine that most developers are still trying to get their hands on something like it and still don't feel the pain of trying to solve problems with it, that could only be solved with something like the parallella. As a hobbyist, besides exploring parallel computing for its own sake, what other kinds of problems can I explore/solve with the parallella which simply wouldn't be possible on the raspberry pi? Sell a dream and possibilities here. I'm personally not familiar with what would only be possible on a parallella and I might feel more interested in this project if I know why I'd want it (besides learning pp for its own sake).
- SoapSeller 14y ago
  According to Ars-Technica[0] they already raised $2.5 million from VCs. [0] http://arstechnica.com/information-technology/2012/09/99-raspberry-pi-sized-supercomputer-touted-in-kickstarter-project/ http://arstechnica.com/information-technology/2012/09/99-ras...
aristidb 14y ago
I'm hardly an expert in this, but doesn't that architecture look similar to Intel's MIC architecture (http://en.wikipedia.org/wiki/Intel_MIC http://en.wikipedia.org/wiki/Intel_MIC) with the upcoming Xeon Phi product? See also http://semiaccurate.com/2012/08/28/intel-details-knights-corner-architecture-at-long-last/ http://semiaccurate.com/2012/08/28/intel-details-knights-cor...
- yvdriess 14y ago
  It's more similar to the Tilera64 chips and ye olde Transputer. Key difference is the reliance of the MIC on cache coherency across all cores.
gbraad 14y ago
It is a nice try. Looks quite similar to the Creative Zii (Egg platform) and the Zii Labs' ZMS-05 processor which they marketed as "Stemcell Computing"... and a lot of other attempts. They use a Zenboard (Xilinx Zync, ARM+FPGA) as a base platform. My first reaction; they claim this being 'open source'. Nothing about the ARM processor or even the core inside the FPGA is open source. What they will deliver is the toolchain and the documentation, but no IP or RTL code for the cores. Another Fauxpen source project... using it merely as a buzzword to get people involved; comparable to the Beagleboard to get usecases and branding out. When hardware is called "open source", they need to look at how Milkymist does it. PCB design files are offered, but also the RTL verilog is available for the CPU (in fact the whole SoC).
- jws 14y ago
  They are specific about their open source intentions. Open Source: The Parallella platform will be based on free open source development tools and libraries. All board design files will be provided as open source once the Parallella boards are released. So we won't be able to fabricate our own derivative silicon. But we will have all open source drivers and tools. We won't have chips full of DSPs that we can't use, or GPUs that work a little bit through some driver that the silicon vendor had to get to MVP for a single version of Linux and can abandon in a year. Sounds good to me.
mbq 14y ago
Sorry, but you can't do any serious computation with 1GB of RAM; do it 149$ but with 4GB minimum.
- cdwhite 14y ago
  Not true. I spent a summer not too long ago doing quantum Monte Carlo calculations for condensed-matter physics, and IIRC Valgrind showed that the whole thing was under 5 MB. Not only did it not require 1GB of RAM, but code + data fit in the L2 cache of each node. (We were seriously CPU bound, not memory bound.) I can't speak to how much of a market there is for something with 1GB RAM, but you certainly can do "serious computation" with far less than that.
  - mbq 14y ago
    Well yeah, this is an exaggeration; things like brute force hash reversing can take kilobytes. Yet you must admit that such problems are in minority even in material science. Also note that some Linux distros in GUI mode can take like half of that alone (this is this household-experimentation-with-parallelism use-case they advocate in the proposal).
xradionut 14y ago
It's an interesting project, the Epiphany chips. For my interests, the alternatives are FPGAs and GPUs. I'm thinking their sweet spot might be in portable, lower power devices that need performance. I have a few applications (mainly DSP and SDR) that require a certain amount of MIPs that is hard to achive in a microcontroller, but is possible in a robust laptop. (But who wants to drag a portable workstation in the field?)
Destroyer661 14y ago
The Parallella project will make parallel computing accessible to everyone. I think their marketing is a bit off on this one. Does everyone actually need parallel computing on this scale? I don't think so. Those who do need it likely already have CUDA running on GPGPUs. Not that I'm implying stifle innovation but the crowd they're marketing to seems to be way off base.
- hn_is_vile 14y ago
  Yes, everyone needs parallel computing on this scale, it's just that they don't know it (yet). In the same way that everyone needed 3Ghz single core CPU when they already had 2GHz single core CPUs. GPUs are different to CPUs in many ways, not least of which is that they are very difficult to debug, do not support recursion (this might have changed?), need special,non-standard data structures (streams). Multi-core CPUs suffer none of these limitations and developers can use non-proprietary standard build tools to develop software for them.
colomon 14y ago
Was all ready to kick in $99 on the off chance I might get an interesting little machine out of it, but when I clicked on their "Epiphany Multicore Accelerator (16 or 64 cores)" link, it came up with a 404 not found page error. That's not exactly reassuring.
- cottonseed 14y ago
  Seems to be a typo. Looks like this is the correct link: http://www.adapteva.com/products/silicon-devices/ http://www.adapteva.com/products/silicon-devices/.
cottonseed 14y ago
What's with the 800GHz operating frequency nonsense? In the 64 core chip, each core runs at 1GHz. I thought a $199 64 core option would have been more interesting.

110 points by JonasH 14y ago

What's wrong with the Parallax Propeller? You can get a starter kit for $25 [1]. Granted, you used to have to program it in its own language, but C compilers are popping up [2] [1] http://adafruit.com/products/791 http://adafruit.com/products/791 [2] http://propeller.wikispaces.com/Programming+in+C+-+Catalina http://propeller.wikispaces.com/Programming+in+C+-+Catalina

wtracy 14y ago
The Propeller doesn't have sixteen cores running at 2+ GHz with a gig of ram. The Propeller can't even run a real, modern OS.
- ErrantX 14y ago
  The Propeller can't even run a real, modern OS Well... technically neither can this :) Although they do appear to have got it working really neatly with an ARM-based host board. Which is a step forward.

wtracy 14y ago

Has anyone here worked with Epiphany chips before? Are there any special APIs required to use them effectively? How hard would it be to get Haskell code running on one?

JonasH 14y ago
There is a C/C++ SDK available.
- wtracy 14y ago
  Is the SDK publicly available right now?
  - cookiemonster44 14y ago
    http://www.businesswire.com/news/home/20120822005093/en/Adapteva-Announces-Availability-OpenCL-SDK-Epiphany-Multicore http://www.businesswire.com/news/home/20120822005093/en/Adap...

ricardobeat 14y ago

The "Making parallel computing easy" and "The goal of the Parallella project is to democratize access to parallel computing" lines made me immediately think of the Parallax Propeller. Turns out their product isn't remotely similar: up to 64 cores at 800mhz, plus a dual-core ARM CPU. I wonder why they are doing a kickstarter instead of harvesting their dollars from server appliances.

blacksmythe 14y ago
Because they are not ready to harvest dollars from server appliances, and would like someone to fund their engineering.

backprojection 14y ago

This sounds great, but why is it a kick starter project? If the claims they're making are reasonable, it seems to me that, for instance, Google would be more than a little interested, and could easily cough up $750K. What am I missing?

ChuckMcM 14y ago
This is a really good question. Not necessarily Google, they are famous for not investing in other people to do engineering, but it seems like a modest amount of money to get something that should be fairly widely applicable. Of course if it is widely applicable and this investment gets to company what it needs to take off, well the folks who gave them the money aren't really going to benefit in a leveraged way. (No equity) Perhaps they were hoping for a 'raspberry pi' like response (which would be hundreds of thousands of units) and be able to do a sort of stealth funding round kinda thing. No idea of course, but it would be a sweet result if it worked out for them right?
wmf 14y ago
Perhaps the claims are not reasonable. Their site has a pretty high ratio of [citation needed] per page.
SamuelKillin 14y ago
I'd say they're looking for exposure/excitement in the hobbiest/startup datamining space. Partly marketing play, partly validation. I backed $99. I'm very excited.
mng2 14y ago
I also find it strange that a chip company is out on Kickstarter. From their website, though, apparently they started on an initial $2 million, which is peanuts for this industry. On some level I think it's an opportunistic marketing move, but it's possible that it was driven by necessity. It's an interesting concept. If it's cheap enough, it's definitely useful as a DSP. But the custom instruction set is going to keep it away from general usage. They seem to be hoping to get their IP into a cellphone, but that's a seriously uphill battle.

iyulaev 14y ago

Hasn't the whole massive multicore thing been floating around the server space for a few years now? I remember HP Project Moonshot and Sea Micro and a few other projects to built multi-chip ARM servers. However, I don't recall seeing ANY benchmarks that demonstrated that they were any more efficient, per watt, in a REAL application, than the x86 competition. I would really like to see such a data point. I guess now it's being sold as a novelty to let people play with such a technology, which is fine I guess. The CISC vs. RISC days are long over and the battle between the two architectures is a bit silly at this point since the gap between the instruction set and the underlying implementation has gotten quite dramatic. Claims that RISC chips are inherently more efficient may have been true in 1995, but I don't see this holding water today.

wmf 14y ago
For Linpack it has been demonstrated that GPUs and MIC are more energy-efficient than server x86, although Blue Gene beats them all. http://www.green500.org/lists/green201206&green500from=1&green500to=100 http://www.green500.org/lists/green201206&green500from=1...
- iyulaev 14y ago
  I'm certainly not going to argue against GPUs (or FPGAs) being used in scientific applications where performance is limited by floating point or vector performance. In certain scientific applications they can hand general-purpose CPUs their butts. I'm talking about ARM/RISC vs x86 CPUs for server applications.
  - wmf 14y ago
    Sorry, I went on a tangent there. The benefits of wimpy cores have certainly been elusive. Some good links http://static.googleusercontent.com/external_content/untrusted_dlcp/research.google.com/fr//pubs/archive/36448.pdf http://static.googleusercontent.com/external_content/untrust... http://research.microsoft.com/en-us/um/people/trishulc/papers/low_power_cores.pdf http://research.microsoft.com/en-us/um/people/trishulc/paper... http://www.seamicro.com/sites/default/files/MozillaCaseStudy.pdf http://www.seamicro.com/sites/default/files/MozillaCaseStudy...

malandrew 14y ago

Sounds awesome, but I have a few comments I'm going to braindump: $750k seems really ambitious for a kickstarter project with such a niche audience. Is that realistic for 29 days, considering you need to sell 7500 units of the $99 pledge amount? At the end of day 7, Leapmotion had 15,000 applications for a free leap motion device and SDK, and that is a device with a much larger audience. Do you have any investor lined up that would be willing to maybe match a Kickstarter total pledge amount of something realistic like $375k? Have you considered approaching a fund like In-Q-Tel? This seems like the kind of project they would fund, since I imagine a lot of the best parallel computing work is being done in government-funded agencies and labs. I also imagine the government is probably the biggest employer of people working on parallel processing devices. With that in mind, getting a device like this into the hands of many, allows more people to get hands on exposure to parallel computing. Overall, it feels like the funding strategy needs to be diversified, because I imagine it will be difficult to get $750k all from one source, with the exception of a VC fund whose thesis aligns with your goals. Lastly, it feels like a project like this would be a bit too soon. Many developers who are playing with hardware have been playing around with the arduino for a few years, some are now graduating to the Raspberry Pi, which offers clear benefits over the Arduino because you can run tons of stuff simply not possible on the Arduino. However as the Raspberry Pi just came out, I imagine that most developers are still trying to get their hands on something like it and still don't feel the pain of trying to solve problems with it, that could only be solved with something like the parallella. As a hobbyist, besides exploring parallel computing for its own sake, what other kinds of problems can I explore/solve with the parallella which simply wouldn't be possible on the raspberry pi? Sell a dream and possibilities here. I'm personally not familiar with what would only be possible on a parallella and I might feel more interested in this project if I know why I'd want it (besides learning pp for its own sake).

SoapSeller 14y ago
According to Ars-Technica[0] they already raised $2.5 million from VCs. [0] http://arstechnica.com/information-technology/2012/09/99-raspberry-pi-sized-supercomputer-touted-in-kickstarter-project/ http://arstechnica.com/information-technology/2012/09/99-ras...

aristidb 14y ago

I'm hardly an expert in this, but doesn't that architecture look similar to Intel's MIC architecture (http://en.wikipedia.org/wiki/Intel_MIC http://en.wikipedia.org/wiki/Intel_MIC) with the upcoming Xeon Phi product? See also http://semiaccurate.com/2012/08/28/intel-details-knights-corner-architecture-at-long-last/ http://semiaccurate.com/2012/08/28/intel-details-knights-cor...

yvdriess 14y ago
It's more similar to the Tilera64 chips and ye olde Transputer. Key difference is the reliance of the MIC on cache coherency across all cores.

gbraad 14y ago

It is a nice try. Looks quite similar to the Creative Zii (Egg platform) and the Zii Labs' ZMS-05 processor which they marketed as "Stemcell Computing"... and a lot of other attempts. They use a Zenboard (Xilinx Zync, ARM+FPGA) as a base platform. My first reaction; they claim this being 'open source'. Nothing about the ARM processor or even the core inside the FPGA is open source. What they will deliver is the toolchain and the documentation, but no IP or RTL code for the cores. Another Fauxpen source project... using it merely as a buzzword to get people involved; comparable to the Beagleboard to get usecases and branding out. When hardware is called "open source", they need to look at how Milkymist does it. PCB design files are offered, but also the RTL verilog is available for the CPU (in fact the whole SoC).

jws 14y ago
They are specific about their open source intentions. Open Source: The Parallella platform will be based on free open source development tools and libraries. All board design files will be provided as open source once the Parallella boards are released. So we won't be able to fabricate our own derivative silicon. But we will have all open source drivers and tools. We won't have chips full of DSPs that we can't use, or GPUs that work a little bit through some driver that the silicon vendor had to get to MVP for a single version of Linux and can abandon in a year. Sounds good to me.

mbq 14y ago

Sorry, but you can't do any serious computation with 1GB of RAM; do it 149$ but with 4GB minimum.

cdwhite 14y ago
Not true. I spent a summer not too long ago doing quantum Monte Carlo calculations for condensed-matter physics, and IIRC Valgrind showed that the whole thing was under 5 MB. Not only did it not require 1GB of RAM, but code + data fit in the L2 cache of each node. (We were seriously CPU bound, not memory bound.) I can't speak to how much of a market there is for something with 1GB RAM, but you certainly can do "serious computation" with far less than that.
- mbq 14y ago
  Well yeah, this is an exaggeration; things like brute force hash reversing can take kilobytes. Yet you must admit that such problems are in minority even in material science. Also note that some Linux distros in GUI mode can take like half of that alone (this is this household-experimentation-with-parallelism use-case they advocate in the proposal).

xradionut 14y ago

It's an interesting project, the Epiphany chips. For my interests, the alternatives are FPGAs and GPUs. I'm thinking their sweet spot might be in portable, lower power devices that need performance. I have a few applications (mainly DSP and SDR) that require a certain amount of MIPs that is hard to achive in a microcontroller, but is possible in a robust laptop. (But who wants to drag a portable workstation in the field?)

Destroyer661 14y ago

The Parallella project will make parallel computing accessible to everyone. I think their marketing is a bit off on this one. Does everyone actually need parallel computing on this scale? I don't think so. Those who do need it likely already have CUDA running on GPGPUs. Not that I'm implying stifle innovation but the crowd they're marketing to seems to be way off base.

hn_is_vile 14y ago
Yes, everyone needs parallel computing on this scale, it's just that they don't know it (yet). In the same way that everyone needed 3Ghz single core CPU when they already had 2GHz single core CPUs. GPUs are different to CPUs in many ways, not least of which is that they are very difficult to debug, do not support recursion (this might have changed?), need special,non-standard data structures (streams). Multi-core CPUs suffer none of these limitations and developers can use non-proprietary standard build tools to develop software for them.

colomon 14y ago

Was all ready to kick in $99 on the off chance I might get an interesting little machine out of it, but when I clicked on their "Epiphany Multicore Accelerator (16 or 64 cores)" link, it came up with a 404 not found page error. That's not exactly reassuring.

cottonseed 14y ago
Seems to be a typo. Looks like this is the correct link: http://www.adapteva.com/products/silicon-devices/ http://www.adapteva.com/products/silicon-devices/.

cottonseed 14y ago

What's with the 800GHz operating frequency nonsense? In the 64 core chip, each core runs at 1GHz. I thought a $199 64 core option would have been more interesting.