Saturday, April 17, 2010

foxchip to gavino

On Tue, Oct 27, 2009 at 7:26 AM, foxchip wrote:
On Oct 27, 12:33 am, gavino wrote:
> http://technoninja.blogspot.com/2009/10/i7-intel-sickness-but-s40-is-...
>
> is the s40 really like 10 times the speed of an intel chip?
>
> how can this be? Intel has what billions?

Have you ever given Intel money by purchasing one of their
processors? ;-)

I have noticed that more and more people are now calling you a
troll for asking the same questions over and over and over
and insulting the people who answer them by ignoring what they
say and asking the same question again. You rule as our
resident clueless troll.

c18 does not just run at a max of 700mhz, it represents 80%
of the dynamic execution of Forth programs as one cycle
5-bit opcodes that operate at several times the speed of
memory and make programs very small. Because there is not
an instruction pipeline or cache timing is simple, and
consistent. Code triggered by an external event can run
a few hundred picoseconds after an event. Cost and power
consumption and speed are all records and Forth programming
is easier than on complex machines.

SEA24 had 24 c18
SEA40 had 40 c18
GA4 has 4 c18
GA32 has 32 c18
GA40 has 40 c18
GA144 has 144 c18

duh

GA4 is the low end, low cost, low power uw to about 40mw,
with a total of 2.8B maximum forth opcodes per second,
and <$.10 manufacture cost

GA144 costs about $2 to manufacture and has 144x 700
Mhz processors for a total maximum throughput of 100,800
forth mips. so far SEA40, ga32, 40, and 144 have been
put in the same packages for testing.

You have asked before and been told before. I don't
know why you ignore answers to your questions unless
it is the obvious answer that you are just trolling.

How is it possible that Intel chips have some numbers
that are lower? Pentium has varying models with different
numbers of transistors. This computer has a Pentium M
with 400,000,000 transistors. That is roughly the same
as 20,000 c18 or one Pentium core. Yield is low when
you need all of 400,000,000 transistors to work or you
have to throw away the silicon. Manufacture cost is
high, and so are prices. The highest prices are for
high end Pentium chips: you can buy 5000 of them
for $3000 each if you have $15,000,000.00 plus tax handy.

A Pentium core is really something like 20 pentium core
connected inside using pipelining so it looks like one
Pentium core that can internally do 20 cycles of an opcode
and make it look like it took one cycle where thoughput
is concerned. This is one way a pentium uses parallelism,
in its pipelined core, it is part of why they are so big.
Then add on-chip cache, and instruction branch prediction
etc. and backwards compatibility with many generations of
previous chips to a processor and it gets REALLY BIG.

c18 needs none of that because it just does simple Forth.

c18 has 64 words of ram and rom each on each core. That's
not much. Most processors in the world are like that and
cost ~$.10. People make many billions of them a year. But
most of them are very slow, about 10k Forth instructions
per second, and very small, less than 100 bytes of memory.
c18 is unusual in that as small cores go it is still very
small and very low power and VERY fast, 700M vs 10K...

This helps keep c18 very small and cheap so that they cost
about a penny each to manufacture and packages typically
add a half a cent per pin. On the other hand Pentium have
hundreds of much more expensive pins, thousands of times
more silicon, and of course a power supply thousands of
times bigger and a heat sink etc.

But Pentium are designed to pull heavy loads at high speed.
They can address 64 gigabytes of RAM! 64 GIG VS 64 WORDS.
Of course only Anton actually has 64GB of ram on his PC.

Pentium is terrible for realtime. Sure this machine has a
2GHz Pentium but at any time it can stall for hundreds of
cycles after a cache miss or pipeline stall. The average
speed is 2 billion instructions per second but because
realtime performance is unpredictable one must assume that
occasionally a function will take 100 times as long as it
takes on the average (see Koopman's paper back in Embedded
Systems journal years ago). This is why a 2GHz Pentium
can only process a couple hundred thousand realtime events
per second. Despite running at 2GHz it might take a ms
or more to get around to processing a realtime event on a
pin.

c18 has no interrupts at all. the idea instead is to
dedicate a c18 to an event such as a change on a pin
so that processing can being in 100ps instead of
1,000,000,000ps like on a Pentium.

Pentium and c18 based computers are about as different
as computers can be.

GA4 characterizes a c18 based computer well, designed
for apps where you want an <$.10 processor. Pentium
are for PCs costing $500 to $20000

Chuck Moore's company licensed multiprocessing and clocking
patents to Intel and AMD. Now they make multicore processors.

But when core cost $20-$1000 to manufacture and are BIG
they can only fit a few on a chip. But they are using the
state of the art fab while Chuck's c18 designs have been
done in .18u or .13u like Intel was using a decade ago.

So if you level that playing field then a GA144 would cost
about what a GA4 does and you could get about 20 cores for
a penny. But those kind of fabs cost billions of dollars and
forth chips are not so well funded.

Now a c18 can go in package where some of them have enough
pins to connect to an external memory. You only need a few
pins for flash, more for ram or dram. On the haypress creek
board there are 9 SEA40 each with 32MW of DRAM on the back
of the board.

Maximum density for .18u is a little less than what you
see on the SEA40 wafers. You could think of a wafer as
a BIG chip, bigger than Pentium but costing about the
same because it uses older .18u fabrication and has a
much higher yield since ever 20k transistors without a
bad transistor is a working core instead of needing
hundreds of millions of perfect transistors per core.

This has all been explained a dozen times, more like a
hundred times, in c.l.f. People really wonder why
you ask the same questions over and over.

Just trolling I guess.

You know that Pentium have to be backwards compatible
with thirty years of Intel chips, that they need to
address huge amounts of memory, that they need giant
packages with hundreds or thousands of pins, that
they need large caches and power supplies and heat
sinks. You know why they are big and expensive and
why they are fast at what they do.

They are like a super freighter. They can carry 100,000,000
tons of stuff and are pretty fast but need many miles to
speed up, slow down, or make a turn. Pentium are like
a freight train too.

c18 is like a racing motorcycle. ultralightweight, doesn't
carry much, but can make turns at speed and stop and start
very quickly.

Really a Pentium to a c18 comparison is like a freight
train to 20,000 dirt bikes. If I need to haul 10,000
refrigerators across the country a freight train would
be better than 20,000 dirty bikes. But to move 20,000
people in 20,000 different directions as fast as
possible a freight train would not do very well.

I guess it must be fun for you to troll.

No comments: