Who is running Adapteva now? New Parallella Kickstarter?

Forum for anything not suitable for the other forums.

Who is running Adapteva now? New Parallella Kickstarter?

Postby KNERDY » Mon Sep 04, 2017 10:04 pm

From this article, http://www.adapteva.com/andreas-blog/adapteva-status/ Andres indicates he stepped down way back in January.

So I am guessing it's on autopilot? It is because of this statement which reads in part, " at this point the platform will need someone else to step up to the plate in my absence to push the technology forward. If you still believe in Parallella and Epiphany and want to contribute or otherwise collaborate with Adapteva send an email to..."

I am not an Electrical Engineer, but instead went to school for Computer Science. I do have an interest in seeing this move forward, and would like to see a new board with at least the Epiphany V processor, newer ARM v8 CPUs, a lot more memory, and even include the ARM Mail GPU on a desktop version. Possibly do this in a new Kickstarter with a lot more press coverage. I dd not even find out about Parallella until way after the Kickstarter had long finished.


So, what can be done to press this forward, and even get out a newer Epiphany beyond the 5th one?
User avatar
KNERDY
 
Posts: 13
Joined: Thu Jul 20, 2017 9:23 pm

Re: Who is running Adapteva now? New Parallella Kickstarter?

Postby dobkeratops » Wed Sep 06, 2017 6:21 am

KNERDY wrote:From this article, http://www.adapteva.com/andreas-blog/adapteva-status/ Andres indicates he stepped down way back in January.

So I am guessing it's on autopilot? It is because of this statement which reads in part, " at this point the platform will need someone else to step up to the plate in my absence to push the technology forward. If you still believe in Parallella and Epiphany and want to contribute or otherwise collaborate with Adapteva send an email to..."

I am not an Electrical Engineer, but instead went to school for Computer Science. I do have an interest in seeing this move forward, and would like to see a new board with at least the Epiphany V processor, newer ARM v8 CPUs, a lot more memory, and even include the ARM Mail GPU on a desktop version. Possibly do this in a new Kickstarter with a lot more press coverage. I dd not even find out about Parallella until way after the Kickstarter had long finished.

So, what can be done to press this forward, and even get out a newer Epiphany beyond the 5th one?


I get the impression it's unlikely to ever appear, but dreaming..

Given that the epiphany 5 has more on chip memory (nearly 64mb pending deadcores) and GPIO, I would like to have seen it go the other way: a much more minimalist device, intended to be paired up with a raspberry pi or arduino or whatever - like the [https://developer.movidius.com] Movidius USB stick, 'on steroids';
dont both trying to compete with RPis in making a general purpose computer system, but rather make an AI accelerator to work with existing devices.

I wonder how much it would cost to make a single board with a 2x2 configuration (imagine that competing with the multi-GPU setups for training NNs
dobkeratops
 
Posts: 189
Joined: Fri Jun 05, 2015 6:42 pm
Location: uk

Re: Who is running Adapteva now? New Parallella Kickstarter?

Postby KNERDY » Fri Sep 08, 2017 4:54 pm

The Pine64 group has taken a similar approach on one of their models with what they call a "Compute Module," where the basic items of a computer are stuck on a SODIMM type module.
They do sell the base board to put it on. This is so one can evaluate it, then maybe build their board to to hold multiple modules.

Something similar could be done with the Epiphany chips.

https://www.pine64.org/?page_id=1491

The cost of making a board is not all that much really. There is a video on YouTube from around 2012, or 2013 where Andreas Olofsson is discussing the costs. The costs for making the Epiphany chips is where it is at. Back then, he was saying the cost of the 16 core run would be $1 Million USD (for the number of boards/chips they ordered), while the 64 core would be $3 million USD. He was indicating the much higher cost was due to the die size. Smaller die has a much larger cost. Because you are putting more cores in the same area, your die size has to shrink.

I would expect costs to have dropped since then. Since they are not posting the schematics of the CPU online, maybe it would be hard to get a cost analysis, unless we try to go with the specs we know of and ask a manufacture for a cost estimate.

The ARM processors are already being mass produced, thus of course they are cheaper, and readily available.


I guess I will have to study being an Electrical Engineer on my own time, when I have time.
User avatar
KNERDY
 
Posts: 13
Joined: Thu Jul 20, 2017 9:23 pm

Re: Who is running Adapteva now? New Parallella Kickstarter?

Postby dobkeratops » Mon Sep 11, 2017 1:14 pm

I would expect costs to have dropped since then. ..
The ARM processors are already being mass produced, thus of course they are cheaper, and readily available.


I think the up-front costs rise exponentially each generation (the masks for each process get harder to make) .. so you need bigger volumes for the final product to be cheap; that's what seems to kill this. The GPUs (and ARMs of course) grew a market gradually over decades.. whereas this has to grow a new market from scratch.. chicken/egg situation with software.

and yet AI is all the rage now, and has data-flow programming models (like tensor flow). There must be a way for the world to get this done..
dobkeratops
 
Posts: 189
Joined: Fri Jun 05, 2015 6:42 pm
Location: uk

Re: Who is running Adapteva now? New Parallella Kickstarter?

Postby claudio4parallella » Thu Sep 14, 2017 6:23 pm

I'm running two Parallellas just now: two Porcupines and next step eLink interconnected Parallellas.

Anybody else is still working / playing with Parallella just now?
claudio4parallella
 
Posts: 55
Joined: Thu Aug 10, 2017 3:48 pm

Re: Who is running Adapteva now? New Parallella Kickstarter?

Postby KNERDY » Sun Sep 17, 2017 5:19 pm

dobkeratops wrote:
I would expect costs to have dropped since then. ..
The ARM processors are already being mass produced, thus of course they are cheaper, and readily available.


I think the up-front costs rise exponentially each generation (the masks for each process get harder to make) .. so you need bigger volumes for the final product to be cheap; that's what seems to kill this. The GPUs (and ARMs of course) grew a market gradually over decades.. whereas this has to grow a new market from scratch.. chicken/egg situation with software.

and yet AI is all the rage now, and has data-flow programming models (like tensor flow). There must be a way for the world to get this done..



I could be wrong, but my thinking is, the previous larger die process would be a lot cheaper once the ability to make a smaller die has been established, and the return on that investment has been obtained. At this point, the Epiphany V could probably be made a lot cheaper now than it could of been a while back.
User avatar
KNERDY
 
Posts: 13
Joined: Thu Jul 20, 2017 9:23 pm

Re: Who is running Adapteva now? New Parallella Kickstarter?

Postby dobkeratops » Mon Sep 18, 2017 2:09 pm

> At this point, the Epiphany V could probably be made a lot cheaper now than it could of been a while back.

it's a chicken egg thing.. my understanding is it needs a large production run (I can't recall the exact numbers, 10,000? 100,000 units?) they didn't quite get enough through the first kickstarted to to the 64core version (they were stung in the crossover between 60->28nm)

we'd have to get 10,000+ people to put in money upfront and wait many months for a finished product. I argue there should be a market for this (deep learning researchers, maker community - the new chip has it's own GPIO) however we appear to be a minority. as such we get more 'bang per buck' from traditional mass produced chips (the RPi version 3 .. the nvidia jetson boards or GPUs for PCs etc etc)
dobkeratops
 
Posts: 189
Joined: Fri Jun 05, 2015 6:42 pm
Location: uk

Re: Who is running Adapteva now? New Parallella Kickstarter?

Postby jar » Mon Sep 18, 2017 8:05 pm

dobkeratops wrote:it's a chicken egg thing.. my understanding is it needs a large production run (I can't recall the exact numbers, 10,000? 100,000 units?) they didn't quite get enough through the first kickstarted to to the 64core version (they were stung in the crossover between 60->28nm)


See slide #33 for some chip production cost analysis: https://www.darpa.mil/attachments/Defen ... Summit.pdf
The slide states $1.8M for a 14nm, 125 mm^2 chip with 10,000 units (or $180/chip). At 100,000 units, the cost per chip is approximately halved (<$100) since the full mask is purchased. Extrapolating using marginal costs, at 1M units, cost is ~$20/chip. At 10M units, cost is approximately halved again (~$10/chip).

The chart implies full mask sets cost about $9M on a modern process. With more math, you can approximate the largest photomask size and wafer cost. Multi Project Wafer (MPW) runs reduce the cost of the mask for small production quantities (less than 100,000 units).

This does not count other product engineering costs.
User avatar
jar
 
Posts: 293
Joined: Mon Dec 17, 2012 3:27 am

Re: Who is running Adapteva now? New Parallella Kickstarter?

Postby KNERDY » Wed Sep 20, 2017 12:09 am

Upon doing some research, I may have found a possible solution to overcome the high cost. Rather than produce millions of units to get a very low cost, I was reading about something called "Multi-project wafers." It's a way for different groups to combine their lower quantity projects to make a larger low cost one.

A short article about it is available on Wikipedia. https://en.wikipedia.org/wiki/Multi-pro ... er_service

With that, has anyone done a performance comparison with application-specific integrated circuit crypto-currency chips to the Epiphany III 18 core? Those of the same era, of course for accurate measurement. Those currency miners are very expensive. If a similar performance can be shown, I am sure those who purchase those miners would jump on board for a possible cheaper alternative, especially if they can be obtained for a much lower cost.

I do like the idea of modularizing a Epiphany system where you can add more chips via a board onto a main board.
User avatar
KNERDY
 
Posts: 13
Joined: Thu Jul 20, 2017 9:23 pm

Re: Who is running Adapteva now? New Parallella Kickstarter?

Postby jar » Wed Sep 20, 2017 1:54 am

KNERDY,

You didn't read last my post (right above yours) where I identified exactly what you found and exactly how much it would cost for chips. The Epiphany-V is 117 mm^2 which is close to the 125 mm^2 for the cost numbers cited. Please read it and the source link.

A general purpose chip like Epiphany will not beat crypto ASICs with the same manufacturing process.
User avatar
jar
 
Posts: 293
Joined: Mon Dec 17, 2012 3:27 am

Next

Return to General Discussion

Who is online

Users browsing this forum: No registered users and 8 guests