I thought of one Application Idea for a GA144

[Jason Damisch]

Hi Guys,

I thought of an application for a GA144, which I thought that I would
post here for public scrutiny. I remember when I was a kid, going to
a friends house and seeing a stand along chess game run by a
computer. It had a chess board, physical chess pieces, and the pieces
would move by themselves. I assume no that there was an electro
mechanical arm inside of the chess board, which would move the pieces
using magnets. Anyway, if the GA144 has so much horse power, then why
not just use all of the nodes to create a mean ass chess game and
drive an LED screen, or an electro mechanical arm. You could even
make it a chaotic-evil genius chess master like Bobby Fisher. Have it
throw random logic into the game to screw with you and trick you into
thinking that it was 'alive'.

Well, such an idea is probably closer to the mark than a multipurpose
toy ( game console ) anyway.

Jason

[Rod Pemberton]

"Jason Damisch" <jasond...@yahoo.com> wrote in message
news:fe32e45f-1c31-499e...@y12g2000yqc.googlegroups.com...

> [...] stand along chess game [...]

stand-alone ?

> Anyway, if the GA144 has so much horse power, then why
> not just use all of the nodes to create a mean ass chess game
> and drive an LED screen, or an electro mechanical arm.

We already have PCs, smartphones, and now Internet capable TVs, etc ... Why
do we need even more hardware? Why would we want dedicated hardware instead
of a multi-purpose generic device?

> You could even make it a chaotic-evil genius chess master
> like Bobby Fisher.

Most people can't beat the simple levels on stand-alone computer chess games
from the 1980's. Modern PC chess games are even more powerful. At some
point the difficulty level will be so great that only people able to play
the top levels against your software or machine will only be expert level
chess masters. How is that going to be profitable? I.e., why would anyone
buy something that far exceeds their needs? That's like a rich guy buying a
Stradivarius for a toddler, or a Lamborghini for his teenager with a
temporary license. If I know I can only play the lowest skill levels of
computerized chess, why do I need the toughest available computerized chess
competitor? That's assuming that chess is even popular enough anymore to
support sales of software or hardware ... I see more about bowling and
curling on TV and the Internet than I do on chess, i.e., almost nothing.
Like Arnold Doray said "Angry Birds" is where it's at for kids. Apparently,
"Words with Friends" is where it's at for middle aged people ...


Rod Pemberton

[Bernd Paysan]

Rod Pemberton wrote:
> We already have PCs, smartphones, and now Internet capable TVs, etc
> ... Why do we need even more hardware? Why would we want dedicated
> hardware instead of a multi-purpose generic device?

Haha, last EuroForth, people asked me why I have a GPS logger, a camera,
a camcorder, and a cheap phone, I should buy a smartphone. Now I have a
GPS logger, a camera, a camcorder, and a smartphone, because the
smartphone's GPS is not good at tracking for 20 hours, the smartphone's
camera is not nearly as good as the camera and the camcorder, and for
standby time, the cheap phone is way better (which matters when your
primary reason to take a phone with you is to be reachable and reach
other people in an emergency). The smartphone fills a niche which none
of the dedicated devices would, it's for gaming, maps, email, and
surfing the web - and for listening to music (but that's something my
old MP3 player could do, too). I like it, it's another special purpose
device that can do a particular thing well (cheap games). But it has
difficulties to replace any of the special purpose devices.

We clearly need even more hardware ;-).

> Like Arnold Doray said "Angry Birds" is where it's at for kids.
> Apparently, "Words with Friends" is where it's at for middle aged
> people ...

The games today are casual games, often in novel ways that older games
couldn't do. They should be easy and fun. "Words with Friends" or
"Gardens of Time" have clear relations to pre-computer-era games, but
certainly not something where the computer has to think a lot.

And for some of the remaining domains of the human brain, like Go, I
have to note that playing Go with real people next to you is a hell lot
more fun than playing against a computer, even if winning against the
computer is easier.

--
Bernd Paysan
"If you want it done right, you have to do it yourself"
http://bernd-paysan.de/

[Arnold Doray]

On Thu, 12 Jan 2012 07:57:36 -0500, Rod Pemberton wrote:

> "Jason Damisch" <jasond...@yahoo.com> wrote in message
>
news:fe32e45f-1c31-499e...@y12g2000yqc.googlegroups.com...
>
>> [...] stand along chess game [...]
>
> stand-alone ?
>
>> Anyway, if the GA144 has so much horse power, then why not just use all
>> of the nodes to create a mean ass chess game and drive an LED screen,
>> or an electro mechanical arm.
>
> We already have PCs, smartphones, and now Internet capable TVs, etc ...
> Why do we need even more hardware? Why would we want dedicated hardware
> instead of a multi-purpose generic device?
>

AIUI, the point Jason makes is that you could add a lot of intelligence
to old toys or even create new ones with a cheap, powerful and low power
processor like the GA144. AI toys are very interesting idea, IMO. Toys
that learn through interaction.

For example, let's take a popular toy, the robot dog. Right now, these
have the actuators to do a set of actions. For example, a basic dog might
sit,walk,bark and beg. The sequence of these actions are pre-built into
the toy's logic. No chance to teach the dog any tricks. All you need to
make it more interesting is the ability to understand feedback (ie,
understand positive feedback -- "Good Doggie" and negative ones -- "Bad
Dog!" as spoken commands or touch), and to label combination of actions
as a new command (eg, walk and bark as "sic" ). If you could add a cheap
camera to the toy, even more might be possible - "chase ball", etc.


Cheers,
Arnold

[A. K.]

Bite carpet!
:-)

[A. K.]

I mean, computational learning requires magnitudes more than just
selecting a processor -- if you don't count adaptive controls as robotics.

[Arnold Doray]

That's a valid point. This brings the discussion a little farther: Does
the GA144 present a good hardware platform for AI?

I'm guessing "yes". The fact that the cores work independently, can
communicate (relatively) easily and are fast means that it is a good map
to AI solutions. I think the GA website says as much. Remains to be seen
of course.

Further, if 144 cores seems like a low ceiling, in CM's 2011 fireside, he
says that they've used a conservative fabrication process (10 years old).
Using a modern process would mean very much higher densities - 50x as
much cores in the same space. That's some serious computing power.


Cheers,
Arnold

[rickman]

I have a GPS, a second hand camera and a dumb phone primarily because
the smart phone needs to be recharged every 12 to 24 hours and I don't
want to spend the bucks an expensive toy that will be obsolete in six
months. I am watching the tablet computers and I think in a year or
so when the internal memory is 16 GB Flash, 4 GB RAM I might jump and
get one. I also want them to work with up to 128 GB Flash cards vs.
the current 32 GB ones. That will replace my current laptop and I may
never go back, depending on available EDA software. but...

Even then, like you, I will still have a camera (likely a new water
proof one) a handheld GPS but I don't know if I'll still have a dumb
phone. Dumb phones are cheap, but the smart phones are pretty
tempting. If the battery life becomes an issue with more users and
the makers emphasize run time rather than processing speed they can
make good progress and provide for two days between charges I hope. I
get tired to trying to keep the charge topped off on all my gadgets.

Rick

[rickman]

On Jan 12, 10:46 pm, Arnold Doray <inva...@invalid.com> wrote:
> On Thu, 12 Jan 2012 07:57:36 -0500, Rod Pemberton wrote:

> > "Jason Damisch" <jasondami...@yahoo.com> wrote in message

That is just a matter of software, not hardware. I don't think the
GA144 brings anything to the party that you can't do with other
processors for these sort of apps.

The GA144 provides a wealth of processing that can replace dedicated
hardware, like I/O interfaces. Otherwise the processing is hard to
utilize with the limited memory access. The apps where the GA144
brings something new to the party will require a lot of processing
relative to the I/O and memory requirements.

AI might be an interesting app for the GA144, but I don't know much
about AI. I keep seeing the GA144 as similar to an FPGA, lots of
small processing elements which don't need to be used optimally. But
as FPGAs have found, it is hard to create large designs without some
specialized blocks such as RAM, EEPROM, special processing elements
and flexible I/Os. I don't think the special processing elements are
needed in a GAxxx device, but certainly memory could be useful, likely
separate blocks like in an FPGA and the I/Os are hurting. 1.8 volt
interfaces with no differential I/Os are just too restrictive.

If you design a specialized chip it will only get used for a few
applications. Add more flexible features and it can be used more
widely. Building a chip with more cores without addressing the other
aspects of system design will only make the application of the device
more limited.

Rick

[Rod Pemberton]

"Arnold Doray" <inv...@invalid.com> wrote in message
news:jeo9ed$cos$1...@dont-email.me...

>
> AIUI, the point Jason makes is that you could add a lot of intelligence
> to old toys or even create new ones with a cheap, powerful and low power
> processor like the GA144. AI toys are very interesting idea, IMO. Toys
> that learn through interaction.
>

Today, people make them. There is a large "maker" or DIY community using
robotics, animitronics, and microcontrollers, like Arduino.

Hack-a-Day
http://hackaday.com/

Makezine
http://makezine.com/

Instructables
http://www.instructables.com/

Kipkay (electronics and gadgets)
http://kipkay.com/

There are a number of channels on Youtube by individuals for various topics.
E.g.,

GreenPowerScience (green energy)
http://www.youtube.com/user/greenpowerscience

FPSRussia (guns)
http://www.youtube.com/user/FPSRussia

PhotonicInduction (high current transformers, lasers, power lamps)
(Sorry, recently closed)


Rod Pemberton

[Rod Pemberton]

"Bernd Paysan" <bernd....@gmx.de> wrote in message
news:jen14j$5sn$1...@online.de...

> Rod Pemberton wrote:
> > We already have PCs, smartphones, and now Internet capable TVs, etc
> > ... Why do we need even more hardware? Why would we want dedicated
> > hardware instead of a multi-purpose generic device?
>
> Haha, last EuroForth, people asked me why I have a GPS logger, a camera,
> a camcorder, and a cheap phone, I should buy a smartphone. Now I have a
> GPS logger, a camera, a camcorder, and a smartphone, because the
> smartphone's GPS is not good at tracking for 20 hours, the smartphone's
> camera is not nearly as good as the camera and the camcorder, and for
> standby time, the cheap phone is way better (which matters when your
> primary reason to take a phone with you is to be reachable and reach
> other people in an emergency). The smartphone fills a niche which none
> of the dedicated devices would, it's for gaming, maps, email, and
> surfing the web - and for listening to music (but that's something my
> old MP3 player could do, too). I like it, it's another special purpose
> device that can do a particular thing well (cheap games). But it has
> difficulties to replace any of the special purpose devices.
>
> We clearly need even more hardware ;-).
>

Even if your smartphone doesn't let you replace your current camera and
camcorder, future generations will, except for maybe professional
photographers. They'll continue to become more powerful and useful.
Currently, sales of cameras and camcorders are declining because of
smartphones. I don't have a smartphone, since I don't like phones, but my
brother does. He can watch movies, listen to music, browse the web, send
emails, text messages, make phone calls, etc. So, for most people, they've
probably already replaced MP3 players and the primary uses of PCs, and
probably PDAs too ... How many people need a dedicated GPS logger? Most
people don't want GPS anywhere near them, i.e., they rightly fear government
tracking. The exceptions are to find places, like restaurants, on
interactive maps, and for locating their position during an emergency.
Although, it's likely they won't be able to escape it's omnipresence. GPS
and RFID is being integrated into just about everything. Your automobile
tires (RFID), your automobile's crash recorder (GPS), ... The list goes on
and on.


Rod Pemberton

[Albert van der Horst]

In article <jeoo40$js8$1...@dont-email.me>,

I'm guessing no. AI requires considerable memory. Interfacing memory
to a GA144 , especially if it has to feed several processors is a
crime.

>
>Further, if 144 cores seems like a low ceiling, in CM's 2011 fireside, he
>says that they've used a conservative fabrication process (10 years old).
>Using a modern process would mean very much higher densities - 50x as
>much cores in the same space. That's some serious computing power.

I have this box (50 vintage refrigerator) with 64+ transputers and
20 fans and 3 5 V 100 A power supplies.
It would be something to rebuild it with GA144's.
What some people don't realize is that pJ per operation is important,
because you can have only some much dissipation in a tight spot.

>
>
>Cheers,
>Arnold

Groetjes Albert
--
--
Albert van der Horst, UTRECHT,THE NETHERLANDS
Economic growth -- being exponential -- ultimately falters.
albert@spe&ar&c.xs4all.nl &=n http://home.hccnet.nl/a.w.m.van.der.horst

[Arnold Doray]

On Sat, 14 Jan 2012 11:51:34 +0000, Albert van der Horst wrote:


>>I'm guessing "yes". The fact that the cores work independently, can
>>communicate (relatively) easily and are fast means that it is a good map
>>to AI solutions. I think the GA website says as much. Remains to be seen
>>of course.
>
> I'm guessing no. AI requires considerable memory. Interfacing memory to
> a GA144 , especially if it has to feed several processors is a crime.

I don't think AI (to be specific, neural networks) require much memory --
it really depends on the size of your network and the type of algorithm
you use. To be more specific, for every connection you make in your net,
you could get away with a single 32 bit cell. Or a single byte if you're
clever. If you us a more local topology, then you only need a few cells.
The update algorithms are not complicated and are local. The update
algorithms are trickier to implement in a small space though. But you
could do this in a few dedicated "backprop" nodes that inteorrogate other
nodes and updates them. Tricky but doable.

I think the programming will be the trickiest parts. Some popular
newtorks (eg the Kohonen network, my favourite) have global update rules,
or update rules that won't fit on one node. This would bottleneck the
computation (thereby negating the advantages of the GA144) if done
serially. So there might be a need to create new local update rules, but
this is not impossible.

It would be interesting for someone to write a simple backprop-type NN
for the GA144. What would help is for the guys at GA to share their code
as they use it, eg the code to make a node act as a smart wire, etc.


Cheers,
Arnold

[Ian Osgood]

I've thought seriously about this, even as just a one-off project. A
little history:

The strongest dedicated chess computer was made in 1993. The TASC R40
had a 40MHz ARM6; 256K ROM for program, opening book, and tables of
coefficients; and 1024K RAM for transposition tables (a necessity for
good endgame play). The program was The King (predecessor of the
ChessMaster programs). This was sufficient attain a master-strength
2400 ELO program; similar hardware won the 1991 world computer chess
championship.

TASC managed to sell twenty of these units.

The market reality is that dedicated chess computers nowadays are only
sold as children's toys, and only the cheapest embedded processors
(Hitachi H8 is popular) running 20 year old engines are required for
beginner's strength units.

Those who want portable strength can download the open-source
Stockfish for their smartphone for $5, and obtain a 3000 ELO opponent,
far above the strongest grandmaster.

On the other hand, there is a boutique market of "oldies" collectors
who are willing to pay big bucks for historical exemplars and nice
wooden boards. One maker, Phoenix Chess Systems, sells $4000 units to
this market using a 500MHz XScale with 16MB RAM. A selling point is
that his "Resurrection" units contain multiple engines, both modern
freeware and strong commercial engines, as well as 6502 and 68k
emulation and front panel emulation for a number of old dedicated
units for nostalgia. You might squeeze into this market using a GA144
simply for the novelty value of 1) a unique processor 2) multi-core
and 3) rather speedy. You'd have a guaranteed market of at least a
hundred european purchasers, even at a thousand euro price point. It
would be quite a challenge to fit a standard chess program into the
unique program-limited architecture of the GA cores, not to mention
safe concurrent access to whatever RAM you'd use for transposition
tables.

Ian