Parallella: A Supercomputer For Everyone - Kickstarter Project

vv111y

Member
May 6, 2011
76
4
8
Niagara Falls, Canada
Hi,

[Admin let me know if this okay]

I'd like to promote this kickstarter project in the hopes it can actually reach its goals(s).

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

I won't regurgitate all the info on the project page - I think they cover everything well enough. Where this really shines is in power consumption, and to me, in the promotion of an open-source, grass roots based community around the chip/platform.

Price/GFLOP is worse than if you got a current GPGPU setup, but if you ran Your system 24/7 my guess is you would get your money back in a year. Also, the value is in promoting a community based platform which is an investment that will keep on giving returns long into the future.

This comment says it best I think:

Kai Staats 2 days ago

I spent ten years in supercomputing, building large scale systems built upon the POWER architecture. I was engaged in dozens of discussions with leading scientists at NASA, the DoE and DoD, at Universities and in the meetings with Freescale, IBM, and Sony -- and this is what everyone wanted to do, but no one could make it happen for the burden of the corporate structure.

The big guys are too blinded by short-term profit, answering to their shareholders in a roller coaster market to be able to take a step back and consider the long-term, the value in establishing a new paradigm for open source supercomputing.

Yes. You are doing this right. This is exactly how supercomputing should be developed. It is the only way for advanced parallel systems to take the next, big leap forward. No one company has the resources to make the philosophical and realworld breakthroughs in-house. Only through engaging the bigger landscape of closet genius, hacker kids, retired software engineers, and everyone in between can the true potential of an architecture be fully explored.

This is Arduino on steroids. Make if happen!
I've pledged $199 for the 64-core option if they reach $3 million,reverts to 2x 16 core systems if only $750K. Please spread the word, thanks!
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,248
5,348
113
Hi,

[Admin let me know if this okay]

I'd like to promote this kickstarter project in the hopes it can actually reach its goals(s).

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

I won't regurgitate all the info on the project page - I think they cover everything well enough. Where this really shines is in power consumption, and to me, in the promotion of an open-source, grass roots based community around the chip/platform.

Price/GFLOP is worse than if you got a current GPGPU setup, but if you ran Your system 24/7 my guess is you would get your money back in a year. Also, the value is in promoting a community based platform which is an investment that will keep on giving returns long into the future.

This comment says it best I think:



I've pledged $199 for the 64-core option if they reach $3 million,reverts to 2x 16 core systems if only $750K. Please spread the word, thanks!
I really like the general concept. If nothing else, it is a low cost platform for students to learn on. Doing this when I was in college was tens of thousands of dollars and more space than my dorm room could handle.

Better thought... want to do a main page writeup? Could slot in next Friday. Let's get this more exposure.

Edit: Just sent them a note.
 

vv111y

Member
May 6, 2011
76
4
8
Niagara Falls, Canada
I really like the general concept. If nothing else, it is a low cost platform for students to learn on. Doing this when I was in college was tens of thousands of dollars and more space than my dorm room could handle.

Better thought... want to do a main page writeup? Could slot in next Friday. Let's get this more exposure.

Edit: Just sent them a note.
Thanks for the positive feedback Patrick! And yes, I'll volunteer for a write up if it works out that way. Just let me know what works for you.
 

cactus

Moderator
Jan 25, 2011
830
77
28
CA
Very cool little chip. I would love to get a datasheet to look over. I want to know what they are using to get the quoted 8GB/sec off chip communication and how the cores are arranged.


Reading through the Hackaday from September 28, I think the last paragraph in Andreas' post is revealing to why this chip is good. It would be very nice for them to release they 64 core chip, but looks like they will have a hard time getting to the $750k mark.
Thanks for the thoughtful comments! I hope you believe me when I tell you that I have been struggling with this question of parallel programming for 4 years now and I don’t have a great answer.(but I don’t think anyone else does either) The idea (maybe its naive?) is that if we put this platform in a lot of different universities for close to nothing,then at least it could be used as a tool for quickly teaching all the current methods. We don’t see this happening without access to cheap and orthogonal hardware.

I am familiar with the examples you mentioned, and I do think there are some differences:

GreenArrays GA144–>people didn’t want forth

XMOS–>great effort but not well known,
no floating point, not high enough performance.
(please correct me if I am wrong)

Parallax–>not modern enough.

GPUs–>not general purpose enough, not really
ANSI-C programmable. Constrains
programming model too much.

FPGAs–>not really software programmable

We do feel that the Epiphany would serve as a better experimentation platform and teaching platform for parallel programming. We already support C/C++/OpenCL and we have people interested in porting openMP and MPI(lite). Halmstad U in Sweden is even playing around with Occam.

The future is parallel, and nobody has really figured out the parallel programming model. I must have heard the question “how are we going to program this thing†over a 100 times in the last 4 years, and my answer was never good enough.(we did well in places where nobody asked the question like in HPC and military). Without broad parallel programming adoption our architecture will never survive, so we obviously have some self serving interest in trying to provide a platform for people to do parallel programming on.

Andreas
 
Last edited:

john4200

New Member
Jan 1, 2011
152
0
0
Judging from the slow progress, it looks like it is going to have a tough time reaching the $750K goal. It is less than half way there with only 11 days to go.
 

vv111y

Member
May 6, 2011
76
4
8
Niagara Falls, Canada
Thanks mobilenvidia, I'll do that now. I'll paste first post to make it quick.

John, I know, this one is having a rough time. I noticed that at the tech community sites there were criticisms and not so much enthusiasm. Some of those criticisms are not justified as they have good answers for that. But it's all spread around in different posts and it adds up to quite a bit of info. I think they needed to talk to folks like that before they started so they could hone their initial message, or make the kickstarter longer to give time to disseminate all the info and arguments for their case. Or something.
 

vv111y

Member
May 6, 2011
76
4
8
Niagara Falls, Canada
done, with this edit that would have been good to do with the initial post

Price/GFLOP is worse than if you got a current GPGPU setup. But:
1) these are the hobby setups to get You started on their system and getting up to speed with programming in OpenCL,
2) these are meant for smaller hacks a'la arduino & rasberry pi. There are 2 interfaces for connecting various hardware
3) any project that needs low-power but high performance
4) their boards that will directly compete with GPU's are coming out later with 1000+ cores. They'll still take 1/10th the power, saving electrical costs for both the board itself, cooling, cheaper PSU, etc, etc.
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,248
5,348
113
I think they have a plan to answer a lot of those questions by Friday. One other thing they could do is show it as a higher compute Raspberry Pi alternative that ships about as fast. My Pi just shipped and it has taken months.
 

Mike

Member
May 29, 2012
482
16
18
EU
If you only F5'ed at launch you would have it collecting dust by now. ;) Atleast mine is till the gpu functions are 'unlocked' with their camera module to get some decent amount of fps.
 

vv111y

Member
May 6, 2011
76
4
8
Niagara Falls, Canada
Yeah I'm happy about that.
Now that they've changed their message and put out a lot more info they are really getting in stride. I thought it was a gonner but I think they will make it - it'll be close. I noticed kicktraq lags.
Now I've got a better idea what we're getting - as another backer posted the ARM Soc with the FPGA sub unit is actually a steel - getting a hobby board with that on it costs $300-400 vs $100.
With the FPGA you can customize the interface with their general hardware plug, so you can fit the logic to match whatever you've plugged in and then feed it to either the ARM cores or the Epiphany for some serious number crunching over the inputs.
I was way off comparing this to GPGPU.
I wonder about using either the GbE+IPMI or the general IO plug to monitor/manage gear and then do analytics over it. An admin box that uses only 5 watts. Just a thought.
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,248
5,348
113
Yeah I'm happy about that.
Now that they've changed their message and put out a lot more info they are really getting in stride. I thought it was a gonner but I think they will make it - it'll be close. I noticed kicktraq lags.
Now I've got a better idea what we're getting - as another backer posted the ARM Soc with the FPGA sub unit is actually a steel - getting a hobby board with that on it costs $300-400 vs $100.
With the FPGA you can customize the interface with their general hardware plug, so you can fit the logic to match whatever you've plugged in and then feed it to either the ARM cores or the Epiphany for some serious number crunching over the inputs.
I was way off comparing this to GPGPU.
I wonder about using either the GbE+IPMI or the general IO plug to monitor/manage gear and then do analytics over it. An admin box that uses only 5 watts. Just a thought.
I think this is actually more of a mix between something like a CPU/ GPU combo and Kinght's Corner. Except Intel's many core revolution (more on this as I can discuss) is a lot more robust. These guys may have an advantage in terms of the fact that they can scale out at a low power.
 

vv111y

Member
May 6, 2011
76
4
8
Niagara Falls, Canada
We did it!

Thanks for all the help and the support! Truly appreciated.

One backer said this has happened before where there's a last minute sprint with a kickstarter that looks doomed.
 

vv111y

Member
May 6, 2011
76
4
8
Niagara Falls, Canada
I think this is actually more of a mix between something like a CPU/ GPU combo and Kinght's Corner. Except Intel's many core revolution (more on this as I can discuss) is a lot more robust. These guys may have an advantage in terms of the fact that they can scale out at a low power.
I'll watch out for your posts, and I'll look into the Phi some more myself. They have the on-chip address space for 4096 cores, so they apparently are ready to scale. Also someone mentioned that Intel's 'ring of rings' is better than it sounds while there are downsides to a 2D mesh. Don't know about that, but we'll see.

I'm happy that there is a open source alternative out there - small but there. It sets a precedent and can change some minds