Parallella: A Supercomputer For Everyone

Started by Claggy, October 25, 2012, 07:59:59 PM

Previous topic - Next topic
Kickstarter has a Supercomputer project on it with just 36 Hours to go, maybe in a year or so we can run Seti@home on it (as long it meets one of it's goals):

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

Mike Hewson over at Einstein has a thread about it, and both him and i have gone for the 64 core Epiphany IV on the stretch goal:

http://einstein.phys.uwm.edu/forum_thread.php?id=9686

I hope some of you will also donate/buy.

Claggy

Already got someone looking into that that PM'd me  ;) looks interesting. I've suggested we see they get their support & look further into it.  Performance of th 64 core Risc unit is supposed to be on the order of one single 'theoretical' 45GHz CPU.  Not a lot by our GPU standards, though at the target of $100 per unit when the 64 core is made, and low power, racking them up could do something decent.
It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change.
Charles Darwin
---
Chaos: When the present determines the future, but the approximate present does not approximately determine the future.
Edward Lorenz

Well, I've pledged in (same level as Claggy and Mike H) - if the hardware comes through next year, you're welcome to program it for me, and then we'll find out 8)

If anyone missed the Kickstarter drive, you can now reserve a board, first come first served.

Reserving a Parallella board

Claggy

For anyone who is interested, and in the UK:

http://forums.parallella.org/viewtopic.php?f=9&t=397

QuotePreparing for Parallella, Sunday 21st July 2013 (UK)

Unread postby 9600 Ã,» Wed Jun 26, 2013 4:06 pm

On Sunday 21st July 2013 Embecosm will be hosting an event at The National Museum of Computing at Bletchley Park, which may be of interest to folks based in the UK. For details see:

http://www.embecosm.com/2013/06/26/preparing-for-parallella-sunday-21st-july-2013-bletchley-uk/

Cheers,

Andrew

Claggy

That sounds possibly do-able - though with a 10:00 Sunday start, I'd need to get a train down the day before and spend the night in a hotel. Anyone else thinking of doing that, and maybe meeting up for a beer somewhere?

Quote from: Richard Haselgrove on June 26, 2013, 03:29:02 AM
That sounds possibly do-able - though with a 10:00 Sunday start, I'd need to get a train down the day before and spend the night in a hotel. Anyone else thinking of doing that, and maybe meeting up for a beer somewhere?

good ness gracious, that's the most roundabout way of asking something I've seen in a while ::)
Have a pint on me.
A person who won't read has no advantage over one who can't read. (Mark Twain)

Mike Hewson over at Einstein has received his Parallellas, built and modded them, pictures at:

Interesting Project on Kickstarter

and

Show your Parallella!! with another pic on page 3,

Claggy

What would be even more interesting I think would be a chassis that could accept between 5 and 30 nVidia K1 chip add-in boards (all small credit card sized things), that you could then run a parallel multi cpu and gpu system. With all 30 add-in boards installed the maximum power draw would be a small 150W and I am thinking its processing capacity could be quite substantial. But hey, I could be wrong :P

Parallella manufacturing and distribution has reached backer number 3662 - moi.

After paying a hefty VAT and customs clearance bill, it was delivered late Saturday afternoon - on a holiday weekend. Now, anybody making any progress on apps? (Mike Hewson is having a real struggle to fit any meaningful FFT into the limited on-chip local memory, and paging to off-chip memory is so slow that it negates the speed of the parallel processing)

Mine haven't reached me yet, according to DHL they are only 40 miles away now.

Claggy

Quote from: Richard Haselgrove on May 03, 2014, 01:19:46 AM
Parallella manufacturing and distribution has reached backer number 3662 - moi.

After paying a hefty VAT and customs clearance bill, it was delivered late Saturday afternoon - on a holiday weekend. Now, anybody making any progress on apps? (Mike Hewson is having a real struggle to fit any meaningful FFT into the limited on-chip local memory, and paging to off-chip memory is so slow that it negates the speed of the parallel processing)

Basically I'm engineering the regression bench part to preceed x42 as cross platform.  Progress so far is a rough GUI, for single device tests) under wxwidgets, and initial probing at suitable device kernels.  Online submission and auto update facility will probably follow during alpha tests (various platforms).  Building the bench will be made to be easy.

The bench part is meant to start with basic chirp, memory, FFT and other functions, accelerating refinement.  That's because I'll be replacing the FFTs with a 'cascade freaky power spectrum' to strip redundant processing specifically around the FFTs.  That may well help matters, as only small simple strided FFTs will be needed here.

Probably as that gets a bit further, whether some basic large FFT optimisation strategies would help can be clearer, though memory bandwidth constraints tend to always be some barrier in plain FFT because it isn't very arithmetically dense.  That's why I'm looking toward compounding/Combining/tiling algorithms instead... to maximise locality and reduce the overall number of passes.  Make more computations per byte fetched. 

Thankfully FFT is rarely used in isolation, and breakign the black-box mentality will probably help a lot with this sortof device, as with others.

It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change.
Charles Darwin
---
Chaos: When the present determines the future, but the approximate present does not approximately determine the future.
Edward Lorenz


Quote from: Claggy on May 05, 2014, 09:34:13 PM
Mine have arrived.

Claggy
I was wondering if they might be taking the scenic route. ;D
A person who won't read has no advantage over one who can't read. (Mark Twain)

Quote from: Claggy on May 05, 2014, 09:34:13 PM
Mine have arrived.

Claggy

Yay toys :)  how long till you can crack those puppies open and hopefully get GCC running ? Command line only, or some desktop GUI?  Will it build base and core wxwigdets libraries ? 
It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change.
Charles Darwin
---
Chaos: When the present determines the future, but the approximate present does not approximately determine the future.
Edward Lorenz

Quote from: Jason G on May 06, 2014, 01:10:55 PM
Quote from: Claggy on May 05, 2014, 09:34:13 PM
Mine have arrived.

Claggy

Yay toys :)  how long till you can crack those puppies open and hopefully get GCC running ? Command line only, or some desktop GUI?  Will it build base and core wxwigdets libraries ?
Well, I've been building Boinc 7.2.4x on the Pi for a couple of months, so once I've got them built into their cases with sufficient cooling (the Zynq-7020 runs hot and must have a heatsink and fan) that won't be long,
there are Ubuntu and Debian images out there or the Parallella, so there is a GUI, I have been getting wxwidgets libs from the repository on the Pi, not been building them.
On the Pi side I've tried compiling a Stock Set app for it but it fails with an error, (already fixed in the Debian source, but not the seti source), can't compile AKv8 ethier, one, all the files that need to be executed
don't that the execution bit set, so they need to be manually set before doing autosetup, etc, two, the compilation, or the make fails with an error, once I have the parallella built i'll report back,
(I've already posted about it at Lunatics), I forget the exact errors, compiling the Stock Seti app shouldn't be a problem on the Parallella like it is on the Pi.

Been running two SDCards with Raspbian Wheezy and Sid on them, Wheezy didn't have a recent Seti app in it, it was a 5.xx app, but after a post on the forums it now has 6.97,
I've not run that app yet, have tried running the Daniel Carrion 7.00 port, that is very slow, times for Wu completion are on the 800 hour scale, and I haven't managed to complete a Wu yet,
On Raspbian Sid we have there is 6.19 svn5051-2 app, this has the more recent optimisations as is a lot faster, some thing like 150 hours a Wu, BUT, not all Wu's completed have come out valid,
Gianfranco Costamagna reports his first Wu on his AMD64 host also was invalid, probably not something to do with arm v6 with vfp, more to do with the repository way of running apps,
or compilation choices, I also have no way of benching repository apps, they can't seem to be run on their own in my home directory like I could with the Daniel Carrion apps (no Arm bench program)

On the Parallella, I have four microSD cards, One of them I've loaded with ubuntu-14.01-lxde but I'm unsure if it was done correctly and if it'll work,
I had file permission errors when I tried writing it, and had to reset the permissions, then force it to get it to write, (I'm still a Linux Noob, and might have got the device name wrong),
I'll try again with the supplied cards, to build the parallella's I need to assemble the third party cases I bought and do a bit of soldering, need to find a bit of wire for the shorting link first.
(I also have the Stock supplied cases from adapteva, five extra small blue heatsinks and loads of extra M3 Nylon bits and pieces, and tags that I bought separate in preparation for making the ABOpen case,
also a fan for the Pi in medium speed spec, (as opposed to the ABopen's kit's hi speed spec fan), need to drill it's case, and find 5volt points on the Pi for it yet)

Claggy

My Parellella's are now both built into their ABOpen Cases, so they have sufficient cooling, the supplied psu's don't work, someone disassembled theirs and found bulged capitors,
replacing those got them going, Bikeman from Einstein has foundout they were recalled by verizon because they didn't work with their routers:

http://forums.parallella.org/viewtopic.php?f=50&t=1145&start=20

so i've got myself two STONTRONICS - T2301ST,  just had to find a pair of power cables for them,
yesterday i got the parallella's to boot, learn't how to ssh into them,  updated them, build Boinc 7.2.47, only one missing dependency,
rebuilt boinc again, attached them to both the Einstein projects and both the Seti projects, got work for both of them the Einstein projects, one's got too much,
temps are running around 55 to 56 degrees C on one, and 59 to 60 degrees C on the other, both under load,
today i've got the seti_boinc and branches sources, bult Seti 7.28 on my T8100 Ubuntu Laptop, and tried to build it on one of the Parallella's,
but it fails with the same error that building Seti on the Pi fails with (Already fixed in the Debrian tree):

/usr/bin/ld: error: seti_boinc uses VFP register arguments, analyzeFuncs_neon.o does not
/usr/bin/ld: failed to merge target specific data of file analyzeFuncs_neon.o
/usr/bin/ld: error: seti_boinc uses VFP register arguments, analyzeFuncs_vfp.o does not
/usr/bin/ld: failed to merge target specific data of file analyzeFuncs_vfp.o

collect2: error: ld returned 1 exit status
make[2]: *** [seti_boinc] Error 1
make[2]: Leaving directory `/home/linaro/seti_boinc/client'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/linaro/seti_boinc'
make: *** [all] Error 2
linaro-nano:~/seti_boinc>

I'll try the AKv8 sources next, but don't have much confidence that i'll make any more progress.

Claggy

Powered by EzPortal