Page 8 of 11
Re: So what will you be doing with your Parallella?
![Post Post](./styles/we_clearblue/imageset/icon_post_target.gif)
Posted:
Thu Aug 01, 2013 12:46 am
by JohnHoleman
I have a couple of goals.
First, back in another life, when I was in college, I did some HPC/Parallel computing (CFD, and some other computational physics/math applications). Sadly, as a youngster, I wasn't a very good student, and there is a certain amount of "use it or lose it" when it comes to these sort of skills (I'm currently taking remedial calculus at community college, because I found that even my maths skills had degraded). So, I intend to reacquaint myself with parallel software development with HPC considerations.
Next, a current interest of mine is in mobile robotics, with the attendant interest in low-power micro-controllers. It would be interesting to see if something like the Parallella could be integrated into such a mobile platform to carry out some of the more complex calculations that might be needed (image processing for stereoscopic vision, complex path-finding calculations, things of that nature). Currently, I'm working on a small, tracked robot, but if I'm successful with it, I may move on to an aerial platform... possibly a small to modest quadrotor.
Re: So what will you be doing with your Parallella?
![Post Post](./styles/we_clearblue/imageset/icon_post_target.gif)
Posted:
Thu Aug 01, 2013 4:40 am
by Jeremy Morgan
I still have not found "the project" that I'll do with it yet, but I'll likely do a lot of hacking and experimenting, much like I do with my Raspberry Pi. I'm a developer so most of my intentions will revolve around software.
I feel the Parallella is a bit different in scope than the Pi, and I plan on really ramping up my skills in the parallel computing field, likely using C, C++ and Haskell. I'm hoping this device can provide a good platform for me to learn and experiment freely and push myself to that next level.
Some ideas I have include image processing, large database processing and maybe some Forex analysis. I'll likely find something great between now and October to devote my time to.
Re: So what will you be doing with your Parallella?
![Post Post](./styles/we_clearblue/imageset/icon_post_target.gif)
Posted:
Thu Sep 05, 2013 10:54 pm
by exa
We're going to port our parallel artificial general intelligence and parallel data mining software to Parallela. (No it's not OpenCog, that's not parallel nor does what we want it to)
Re: So what will you be doing with your Parallella?
![Post Post](./styles/we_clearblue/imageset/icon_post_target.gif)
Posted:
Fri Sep 06, 2013 6:33 pm
by CIB
Re: So what will you be doing with your Parallella?
![Post Post](./styles/we_clearblue/imageset/icon_post_target.gif)
Posted:
Fri Sep 06, 2013 8:48 pm
by LamsonNguyen
Maybe it's, you know, classified.......
![Cool 8-)](./images/smilies/icon_cool.gif)
Re: So what will you be doing with your Parallella?
![Post Post](./styles/we_clearblue/imageset/icon_post_target.gif)
Posted:
Thu Sep 12, 2013 9:16 pm
by atoussaint
I've been using the Gentoo distribution since at least 10 years but now, I am trying to use aboriginal linux (
http://landley.net/aboriginal/) which is a very lightweight linux distribution for which the author (Rob Landley) uses qemu to build the distribution in several architecture natively and he also code for the toybox project (
http://landley.net/code/toybox/) which is a suite of command line utilities which are usually found in a regular linux system. Previously (before mid 2006 I think), he coded on busybox.
Aboriginal linux has been used to bootstrap Linux From Scratch and takes up very little memory and will be ported soon to run on the musl libc (
http://www.musl-libc.org/) and I'd like to develop on aboriginal linux to become a sufficient distribution to run on parallella board. In my case, I will buy the 4-node cluster next January and develop in it a bayesian meta-analysis software for neuroinformatics ( some recent work that I've done
http://www.ncbi.nlm.nih.gov/pubmed/21833294).
At the moment, I am getting comfortable with the code of Signed Differential Mapping software (
http://www.sdmproject.com/) with the goal of porting it to my mac but after that, I will probably extend it with some bayesian analysis code which could be run on the Parallella cluster.
Alain
Re: So what will you be doing with your Parallella?
![Post Post](./styles/we_clearblue/imageset/icon_post_target.gif)
Posted:
Mon Oct 14, 2013 10:37 am
by DeanoC
I'm going to bring up a bare-board system from scratch and write tutorials and film videos going through the various stages and talking to the chips and custom logic.
The project intends to make a working graphics accelerator (I won't say GPU as its not going to be anywhere near PC vid cards performance) that can either by used standalone or as a add-on board graphics co-processor (plan is to make it into a linux USB2 -> HDMI board for tacking on an accelerated extra monitor). Will be open source, so hopefully providing useful graphics bits that can be cherry picked by the parallella community as well.
Obviously a lot of work to do, and the hard to get at JTAG connector makes its a bit harder (at least to start with I'm going to try and leave u-boot in place and jump to the projects code rather than Linux). Whilst its a lot of work, I've been at the bleeding edge of game graphics for 20 odd years (Amiga and SNES graphics chips had some lovely tricks and epiphany are very PS3 SPU like) but this is the biggest RTL project I started (and hopefully will finish!) so should be an interesting experience.
The different processing systems should make it quite powerful and useful for the cost and 'plug and play'-ness, Arm's can handle the less parallel functions, PL providing specialised video and render HW with the Epiphanies bringing float horsepower to the show.
Initially it will be focused on a good 2D set but I'd like to see some simple 3D in future. As things like legacy interfaces and just copying traditional GPU designs isn't required (and a love of old-skool custom graphics chips tricks), we should be able to get some fairly powerful features.
I'd have loved to work with partial reconfiguration of the PL but unfortunately that is an expensive add-on to the Xilinx software, which is sad as the potential for reconfigurable computing is pretty awesome!
Can't wait (though latest news is that I have too *sob* though I understand) to push the ideas into HW.
Deano
mp^2: A multi-precision, multi-(platform/processor) maths li
![Post Post](./styles/we_clearblue/imageset/icon_post_target.gif)
Posted:
Mon Oct 14, 2013 8:29 pm
by over9000
I've lots of ideas for things I'd like to try on the Parallella, but this is the one that I'm currently most interested in.
I want some libraries to be able to do arbitrary precision arithmetic on integers, with a focus on RSA-type algorithms (finding large probable primes and doing power operations mod some large numbers).
All of the bignum libraries I've found are either too complicated (gmp) or too large to fit into the Epiphany's memory space without having some sort of dynamic code loading/overlay system so I'll write or port some simpler/smaller routines that should fit in 2Kbytes of program space or less without having to resort to overlays.
I'm also thinking of ways to exploit parallelism when doing calculations with bignums. Initially, I'll start by writing a simple grammar for a calculator and then figure out which calculations can be done in parallel. I'll output a form of 3-address code (or some other bytecode) that encapsulates the execution plan for the overall calculation. Instructions will include explicit synchronisation operations so that nodes wait for dependent operations to complete before operating on them.
The 3-address code will assume an abstract machine with multiple computation nodes, each of which is modelled as a bunch of registers (carved out of the local memory space) + ALU + synchronisation primitives. So in effect, each node is basically like a simple CPU for working with bignums, although it will also be able to address registers in other nodes as the target of an operation (eg, for a=b * c, both b and c must be local registers, while a, the result, can be remote).
It should be fairly simple to map the abstract machine and bytecode interpreter onto the Epiphany. I'll make things even simpler by doing all of the bytecode interpreting on the host side and implement simpler command queues on the Epiphany side, so it's more like the host is the (simulated) CPU and the Epiphany cores are ALUs.
I intend for this to be portable. Once I've decided on the details of the bytecode and synchronisation primitives, I want to be able to run it on a variety of platforms including multi-threaded, SMP systems (a quad-core x86_64 box), the PS3's Cell processor (which, like Parallella, has a host "PPU" and several "SPU" coprocessors), the Parallella platform itself and an OpenMPI-based virtual machine (with rank 0 host being the "CPU"). I might try to tweak the code generator so that it can take into account the different costs associated with each operation (eg, add, mul, div and copy) on each of the target platforms and try to generate an execution plan (ie, the bytecode) that is best for that platform.
If I manage to get all that working, I'll look at extending the grammar to include conditional statements and loops. I want the language to evolve to the point where it's able to express some key algorithms (finding probable primes, primality testing, gcd and extended gcd) within the language itself. I might also look into doing speculative execution, eg, following both sides of a conditional statement or executing several iterations of a loop in parallel (see "foreach" thread).
Besides the use I have for an RSA-type calculator (actually, I want it for one-way RSA accumulators), I'm mainly doing this as a learning exercise. Also, I have a bunch of machines of varying power (mainly Pis and other ARM-based machines, but also a PS3 and an old, but usable x86_64 machine as my main PC), so I'd like to come up with an application that can get them doing some useful work.
Re: So what will you be doing with your Parallella?
![Post Post](./styles/we_clearblue/imageset/icon_post_target.gif)
Posted:
Mon Oct 21, 2013 3:35 am
by Flyingfox
Got a board shipping in November, and I can't wait.
I hope to use the Parallella board as a wearable computer, complete with a transparent HUD and keyboard gloves.
My interest in this project is having a mobile hot-configurable powerhouse for augmented reality and human-machine interfaces.
I want to use the board for audio and video processing, and use the FPGA GPIO for direct camera and OLED microdisplay interfacing. I am also fiddling with the electronics for wearable EEG (brain activity) and EMG (muscle movement / heart rate) preamps and ADCs for other types of wearable computing interfaces, and the Parallella looks like a great way to do all the DSP needed in one tiny low power package.
![Geek :geek:](./images/smilies/icon_e_geek.gif)
Re: So what will you be doing with your Parallella?
![Post Post](./styles/we_clearblue/imageset/icon_post_target.gif)
Posted:
Wed Nov 13, 2013 2:49 pm
by vidarh
I've had a crazy idea today to port AROS to it:
http://aros.sourceforge.net/We'll see. But AROS is AmigaOS inspired, and AmigaOS already has a long tradition with co-processors, and even different architecture main CPU's (towards the end of commercial availability of the original Amiga's, there were PPC accelerator cards, where m68k code could hand off stuff to PPC code and vice versa. The beauty of AROS in this respect is that Amiga-inspired stuff is crazily small (fully multi-tasking GUI based system in 512K ROM + 512K RAM incl. user apps for systems like the A500), and the system was built around message passing (e.g. a shell/terminal on AmigaOS involves input.device handling the keyboard, console.device handling the output, console-handler handling the line editing, clipboard.device handling the clipboard, a thread for Intuition - the GUI - managing the windows, and more)..
Which makes me want to do fun and crazy stuff like experimenting with farming out various system processes to epiphany cores.