Page 6 of 6

Re: 100 RPi supercomputer w/ 20 GPU boards, appreciate for h

Posted: Fri Aug 08, 2014 6:57 pm
by W. H. Heydt
drgeoff wrote:
W. H. Heydt wrote:
Ravenous wrote: The sooner you build something, even if it's not quite the right thing, the sooner you'll actually achieve the project. Believe me I know!
That rather depends on whether you work like Thomas Edison or Nikola Tesla.

(E.g. Edison tried--what?--1000 different ways to make a light bulb filament before he found one that worked...and that one was superceded by something else entirely when the bulbs went into everyday use. Tesla conceived of his AC motor and gave the machinists the specs for each of the parts. When he got the parts for the very first time, he assembled the motor (all the parts fit together properly on the first try) and the motor worked.)
Yes, but Tesla had a reasonable understanding of what he was doing, whereas ....
I wasn't going to go there, but I understand your point.

Re: 100 RPi supercomputer w/ 20 GPU boards, appreciate for h

Posted: Fri Aug 08, 2014 8:11 pm
by Supercomputing-RPi
I am also getting to a stage where I know a "reasonable amount of knowledge" to embark on this journey.

I my parents are both software and electrical engineers...

Using their help atleast, I can make this happen...

I am going to use PXE boot on all of the ITX nodes once they do they would all go thru PCPIP (Pre-Compute Procedures and Initialization Protocol).

Steps
1. Boot up into Archlinux or Ubuntu
2. Initialize Intel MKL library
3. Run FFT analysis
4. If any GPU error, send message about incident to main node...
5. If FFT succeeds, join DHCP in FCOE... (In select nodes), join DHCP in FCOE and/or GbE.
6. Set up MPI. (In select nodes), set up MPI and send "Hello World" to master node.
7. Wait for instructions from FCOE. (In select nodes), wait for instructions from FCOE and/or GbE.

Power would be managed by remotely switching on clusters and other devices using Arduino.

I do agree on starting in low scale.

1 week and 3 days and 127 posts in 6 pages, whoo!!

Re: 100 RPi supercomputer w/ 20 GPU boards, appreciate for h

Posted: Sun Aug 17, 2014 3:50 pm
by jamesh
This is clearly now nothing to do with the Raspberry PI.

And no-one has actually written that the Performance/Dollar ratio of the Pi is simply too small for supercomputers systems with any power. Fine for teaching about it, not fine for actually using it for any decent workload. Also, no-one has mentioned that actually using the GPU speed is fantastically difficult, bordering on impossible, and the time to develop the software to run the proposed system is multiple 10's of man years.

My suggestion to the OP - you need to read up on massively parallel systems - the computer architecture, the software required. Most supercomputers use Intel (or similar chips) on custom boards that provide very fast interconnects, using boards like the Pi, or even ITX style devices simply don't provide the speed on interconnect required.

Re: 100 RPi supercomputer w/ 20 GPU boards, appreciate for h

Posted: Mon Aug 18, 2014 1:04 pm
by wallarug
jamesh wrote:This is clearly now nothing to do with the Raspberry PI.

And no-one has actually written that the Performance/Dollar ratio of the Pi is simply too small for supercomputers systems with any power. Fine for teaching about it, not fine for actually using it for any decent workload. Also, no-one has mentioned that actually using the GPU speed is fantastically difficult, bordering on impossible, and the time to develop the software to run the proposed system is multiple 10's of man years.

My suggestion to the OP - you need to read up on massively parallel systems - the computer architecture, the software required. Most supercomputers use Intel (or similar chips) on custom boards that provide very fast interconnects, using boards like the Pi, or even ITX style devices simply don't provide the speed on interconnect required.
Good call @JamesH ;)