Re: Would green arrays produce something with a web browser? like a cheap appliance?

From	Bernd Paysan <bernd.paysan@gmx.de>
Newsgroups	comp.lang.forth
Subject	Re: Would green arrays produce something with a web browser? like a cheap appliance?
Date	2012-01-19 23:27 +0100
Organization	1&1 Internet AG
Message-ID	<jfa5d2$9jp$1@online.de> (permalink)
References	(7 earlier) <1b326644-7c60-480c-8fdf-681a23b1111c@34g2000yqm.googlegroups.com> <7x7h0qomxi.fsf@ruckus.brouhaha.com> <bc039b77-257d-43ac-a933-8ee2206f4ceb@m2g2000vbc.googlegroups.com> <jf7ooa$m9$1@online.de> <694c58a1-c5a6-4ea7-be90-834b83f5ca4d@do4g2000vbb.googlegroups.com>

Show all headers | View raw

rickman wrote:
> I don't see the comparison to GPUs as they are power hungry devices
> and do not lend themselves to applications outside of PCs.  I've never
> seen them used in any other designs.  Although they can be separately
> programmed, they don't run an OS and don't work standalone.

Well, GPUs as such are power hungry devices, but in terms of 
power/compute power, they are prett damn'd good.  A teraflops (single 
precision) for 200W?  Well, if you only need a gigaflop, a GPU is too 
big for you.

> I think your comparison with FPGAs is not too far off target.  I
> question the price issue as the Cyclone 5 devices are quite large,
> even the smallest.

Well, it's a 28nm device.  It isn't large, even the largest.  It just 
has a lot of gates on it.  Because it uses the latest available 
technology.

> But in general the GA144 is prices in the same
> ballpark as low end FPGAs.  I don't think it would be easy to get the
> same processing speed in a low end FPGA, but then at GA144 won't be so
> easy to get working with 100+ MHz signals.

A GA144 in 28nm would probably a much more impressive device than a 
Cyclone 5.  But it isn't available.  I think the Cyclone 5 with the 144 
multipliers will beat a GA144 hands down in whatever task you can 
imagine.

I've done my b16 in similar processes as Chuck did his stuff.  But that 
was because my b16 always was just an add-on to some custom-specific 
ASIC with analog and power components, and analog+power usually is added 
to decade old processes (there is a current counter-trend to this, 
because people want SoCs, and the power+analog should not go into 
another die).  The main reason to deploy the b16 never was speed, it 
always was area and power consumption (be quick to finish is good for 
power consumption, too).

The IO stuff always was dedicated Verilog, not bit-banging in software.  
It always cooperated with the CPU, or used DMA to write directly into 
memory.

I heard Chuck talking at EuroForth about 10 years ago, when he started 
this multi-chip endeaver.  He clearly said that he thinks this is a cool 
idea, but did not know what kind of product that would go into.

This is the definition of research project.  You have a cool idea, but 
no product.  Research is necessary.  But to make it a product, you have 
to think about how people will use it.  In a product-driven development, 
the requirements are first, the solution (the actual implementation) 
comes next.

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

Back to comp.lang.forth | Previous | Next — Previous in thread | Next in thread | Find similar

Thread

csiph-web