6 ms·

Hardware Hackers Create a Modular Motherboard

68 points by naish 17y ago

natemartin 17y ago
This is really interesting. I definitely want to read more, because I've got some questions about it. I wonder if there's a limit to how many can be plugged in before you start seeing power problems on the downstream processors, and how it routes information between processors. Definitely intriguing though! Edit: Unfortunately, there doesn't seem to be much documentation on their website. A lot of talk about open source hardware, but I don't see any schematics, datasheets, etc. Oh well. Hopefully they'll be adding that soon.
- Retric 17y ago
  Each X Machina module has a 72 MHz processor (currently an ARM chip), a solid state drive of 16KB and 128KB of storage in an EPROM (erasable programmable read-0nly memory) chip. There’s also an LED for display output. These are not going to suck down much power. They have around 1/1000th the processing power of a 3Ghz quad core chip and 1/250th the RAM. If you want to build a useful mesh like this then cooling is going to be the first concern, followed by latency. The best design would be a back plane with these cards inserted into the side. Add power at the top and cool down the sides with the back plane for communication. But that's what a standard motherboard with PCI-X slots look like so it's less interesting.
  - natemartin 17y ago
    Oh, I know these are very low power. But even if each board only takes 1 watt, if you've got 40 of them you're talking the same power consumption as a laptop CPU, but with a much longer distance for power delivery. But I'm still curious exactly what the power is, what the resistance of the power connections are, etc. As you said, it gets more interesting when they're not in a backplane, so I wanted to figure out what the constraints are. Someone on one of their blogs linked to http://www.xmos.com/ http://www.xmos.com/ which is also pretty interesting. It's a single chip with 4 cores, each of which can run 8 threads. And it takes care of all the switching between cores or even between chips. It doesn't have all the same features as this board does, such as the dynamically switched power/signal pins, but same sort of idea. Edit: In fact, the modularity of this is one of the more interesting aspects of it to me. I'd like to see how they're switching pins between signal and power dynamically. But, there's no schematics that I can find.
    - Retric 17y ago
      There is little problem in supplying power to a grid like this. It's related to the classic Given an infinite grid of 1 ohm risistors, what is the resistance across a single diagonal? Well known answer: 2/pi ohms. http://answers.google.com/answers/threadview?id=92564 http://answers.google.com/answers/threadview?id=92564 Basically, as the size of the grid increases so do the number of paths to an edge element. The practical limit relates to how much current you can drive through a single, element, but you can always use more than one power connector. Edit: USB connection it can support a 3 x 3 grid (500 milliamps * 5 volts / 9) ~= 0.3 watts. So 130 of them should equal a low power CPU.
- natemartin 17y ago
  One of the developers just replied on his blog. They're working on the documentation, and it will be up soon. As far as power delivery goes, off of a USB connection it can support a 3 x 3 grid, and they're working on a power supply board that will support 15 x 15. Pretty cool.
mattyfo 17y ago
Hmmm, modular, clusterable and able to reprogram each other. Is this the rise of the borg?
kiba 17y ago
Look like you can have your own cake and eat it too! http://www.liquidware.com/shop/show/IXM/Illuminato+X+Machina http://www.liquidware.com/shop/show/IXM/Illuminato+X+Machina
bkj123 17y ago
very interesting. I wonder if this architecture would make for a good mpp database appliance - something that is lean and very focused and somewhat power hungry!
jwilliams 17y ago
This is part of the theory behind PCIe - there is a spec for a PCIe "external cabling". The idea is that you'll have your lanes cabled, and you can build a computer out of modular parts (i.e. modular bricks). (although the main driver at the moment seems to be the ability to plug a decent video card into your laptop).
- nazgulnarsil 17y ago
  why hasn't this idea taken off? instead of packing a high end graphics card into a laptop where it is likely irreplaceable I'd rather have a dock for my laptop that includes a bay for a better video card. hardcore gaming at home (with an upgradeable part, low power consumption on the road.
  - jwilliams 17y ago
    Yeah. Dell used to support a dock with PCI (not PCIe) plus an extra hard drive, and the usual port replication -- was great.
srn 17y ago
This project is interesting, given that it's open source and you can dynamically add or remove units. As far as power goes, crank down the clock speeds and any processor is going to run cooler. arm is also well known for taking less power. But about the whole changing your basic architecture, consider Tilera: http://www.itjungle.com/tlb/tlb082107-story02.html http://www.itjungle.com/tlb/tlb082107-story02.html http://www.theregister.co.uk/2008/09/23/tilera_cpu_upgrade/ http://www.theregister.co.uk/2008/09/23/tilera_cpu_upgrade/ And it runs linux! ;)Granted it's closed and probably costs a bazillion, but the big guys haven't exactly missed out on the memory bus being a bottleneck.
smithjchris 17y ago
I've seen this before. It's definitely not new. INMOS did this in the 80's with their Transputer platform (Google it). That rocked as it had excellent software support with OCCAM, C and Pascal support. Oh and they had a concurrent parallel version of UNIX that ran on it. The implementation here is not likely to match it or be scalable as it's bolting functionality onto a rather ancient architecture, doing most of the work in software. If they did it in hardware with an embedded ARM core or FPGA with instruction set extensions for message passing, it might rock but this won't. I'd bet my life on it.

68 points by naish 17y ago

This is really interesting. I definitely want to read more, because I've got some questions about it. I wonder if there's a limit to how many can be plugged in before you start seeing power problems on the downstream processors, and how it routes information between processors. Definitely intriguing though! Edit: Unfortunately, there doesn't seem to be much documentation on their website. A lot of talk about open source hardware, but I don't see any schematics, datasheets, etc. Oh well. Hopefully they'll be adding that soon.

Retric 17y ago
Each X Machina module has a 72 MHz processor (currently an ARM chip), a solid state drive of 16KB and 128KB of storage in an EPROM (erasable programmable read-0nly memory) chip. There’s also an LED for display output. These are not going to suck down much power. They have around 1/1000th the processing power of a 3Ghz quad core chip and 1/250th the RAM. If you want to build a useful mesh like this then cooling is going to be the first concern, followed by latency. The best design would be a back plane with these cards inserted into the side. Add power at the top and cool down the sides with the back plane for communication. But that's what a standard motherboard with PCI-X slots look like so it's less interesting.
- natemartin 17y ago
  Oh, I know these are very low power. But even if each board only takes 1 watt, if you've got 40 of them you're talking the same power consumption as a laptop CPU, but with a much longer distance for power delivery. But I'm still curious exactly what the power is, what the resistance of the power connections are, etc. As you said, it gets more interesting when they're not in a backplane, so I wanted to figure out what the constraints are. Someone on one of their blogs linked to http://www.xmos.com/ http://www.xmos.com/ which is also pretty interesting. It's a single chip with 4 cores, each of which can run 8 threads. And it takes care of all the switching between cores or even between chips. It doesn't have all the same features as this board does, such as the dynamically switched power/signal pins, but same sort of idea. Edit: In fact, the modularity of this is one of the more interesting aspects of it to me. I'd like to see how they're switching pins between signal and power dynamically. But, there's no schematics that I can find.
  - Retric 17y ago
    There is little problem in supplying power to a grid like this. It's related to the classic Given an infinite grid of 1 ohm risistors, what is the resistance across a single diagonal? Well known answer: 2/pi ohms. http://answers.google.com/answers/threadview?id=92564 http://answers.google.com/answers/threadview?id=92564 Basically, as the size of the grid increases so do the number of paths to an edge element. The practical limit relates to how much current you can drive through a single, element, but you can always use more than one power connector. Edit: USB connection it can support a 3 x 3 grid (500 milliamps * 5 volts / 9) ~= 0.3 watts. So 130 of them should equal a low power CPU.
natemartin 17y ago
One of the developers just replied on his blog. They're working on the documentation, and it will be up soon. As far as power delivery goes, off of a USB connection it can support a 3 x 3 grid, and they're working on a power supply board that will support 15 x 15. Pretty cool.

mattyfo 17y ago

Hmmm, modular, clusterable and able to reprogram each other. Is this the rise of the borg?

kiba 17y ago

Look like you can have your own cake and eat it too! http://www.liquidware.com/shop/show/IXM/Illuminato+X+Machina http://www.liquidware.com/shop/show/IXM/Illuminato+X+Machina

bkj123 17y ago

very interesting. I wonder if this architecture would make for a good mpp database appliance - something that is lean and very focused and somewhat power hungry!

jwilliams 17y ago

This is part of the theory behind PCIe - there is a spec for a PCIe "external cabling". The idea is that you'll have your lanes cabled, and you can build a computer out of modular parts (i.e. modular bricks). (although the main driver at the moment seems to be the ability to plug a decent video card into your laptop).

nazgulnarsil 17y ago
why hasn't this idea taken off? instead of packing a high end graphics card into a laptop where it is likely irreplaceable I'd rather have a dock for my laptop that includes a bay for a better video card. hardcore gaming at home (with an upgradeable part, low power consumption on the road.
- jwilliams 17y ago
  Yeah. Dell used to support a dock with PCI (not PCIe) plus an extra hard drive, and the usual port replication -- was great.

srn 17y ago

This project is interesting, given that it's open source and you can dynamically add or remove units. As far as power goes, crank down the clock speeds and any processor is going to run cooler. arm is also well known for taking less power. But about the whole changing your basic architecture, consider Tilera: http://www.itjungle.com/tlb/tlb082107-story02.html http://www.itjungle.com/tlb/tlb082107-story02.html http://www.theregister.co.uk/2008/09/23/tilera_cpu_upgrade/ http://www.theregister.co.uk/2008/09/23/tilera_cpu_upgrade/ And it runs linux! ;)Granted it's closed and probably costs a bazillion, but the big guys haven't exactly missed out on the memory bus being a bottleneck.

smithjchris 17y ago

I've seen this before. It's definitely not new. INMOS did this in the 80's with their Transputer platform (Google it). That rocked as it had excellent software support with OCCAM, C and Pascal support. Oh and they had a concurrent parallel version of UNIX that ran on it. The implementation here is not likely to match it or be scalable as it's bolting functionality onto a rather ancient architecture, doing most of the work in software. If they did it in hardware with an embedded ARM core or FPGA with instruction set extensions for message passing, it might rock but this won't. I'd bet my life on it.