Re: [oc] Beyond Transmeta...

[Suboner@aol.com]

> Hi ALL,
> After so many emails, I want to pose my considerations with this
> aggressive subject.

Is it really that agressive?... :)

> I think mainly Transmeta owns two advantages. The first one is its 
compatibility
> with X86 ISA. The second one is her low power technology. How does the
> net-connected processor conquer the two issues at the same time?

The compatibility is in the ability to configure a network of bits that would 
work exactly like an X86 processor. It may even be possible to even create 
something that would be able to convert a processor layout diagram (of the 
traces and other parts) into a network of bits. But the other thing is, that 
programs could probably be incorperated into the network more deeply 
depending on how ofter a user uses a program, by incorperating it more deeply 
the user may even be given the ability to alter the program in diffrent ways, 
by blending features together from other applications, like blending a search 
tool a compiler a help system and a text editor to create a sort of IDE 
(literally integrated developement environment :)).

Power consumption would be in that the 1 bit processors are only used when 
bits change. It would probably have many of the advantages (if not more) then 
Crueso because it would only execute the relational instructions when a bit 
changes, but does not execute any more, while most processors do a lot of 
redundant work.

> I think basically many current processors, even what Intel has, have already
> the performances. But Transmeta finds the power niche.

I've heard though, that Transmeta actually says their Crueso processor has a 
diffrent kind of performance that makes it possible for them to do some 
things faster, while other things slower. I'd imagine the same would be true 
for a 1bit processor.

> So, I guess you have the following issues to be solved.
> 1. Hardware architecture: net-connected processor?

Yeah, that is still a big part of it (number 1), because this will effect 
performance and abilities of the network big time.

> 2. Compilers or translators or code-morpher for both native ISA and X86 ISA.

Yeah, that is definetly an issue, but there may be a way, by first building a 
sort of simple assembler from a network, then use the assembler to create a 
more higher level and complex assembler, and then using it to build maybe an 
x86 kind of assembler, and then from there you may be able to compile an 
early x86 C compiler, and then use that to compile more advanced one. It 
would take a bit of generations to occomplish it, but at some point it will 
be advanced enough that you would have a network built for compiling 
networks. And at some point this can start being used for self modifying 
programs (other pieces of software will have access to it, and sort of 
request for changes, or the compiler may sit idly by and wait for a chance to 
optimize the network).

> 3. What's your native ISA of the net connected machine?

Hmm I'm not sure what you mean by native ISA... I did mention that I am not a 
hardware engineer, but maybe an over eager software engineer.

> 4. Power Issue? How to let net-connected processors consume least power?
>     In deep submicro technology, routing is the worst issue. But 
net-connected
>     processors need routing most....

Like I was saying above, the processor only needs to do minimal operations, 
instead of calculating every bit over and over again, it can calculate only 
the bits that have changed, and of course the network could possibly be 
configured for less power consumption and altered performance. So if it only 
needs 10 1bit processors for one period of time and 32 1bit processors 
another, then the ability to use less power may be possible.

> 5....any more?

There probably is quite a bit of them, in details and such. But the main one 
is the hardware which basicly over shadows most of the others.

Its probably best if I mention that all these possibilites that I mention may 
not be practical, or may have the oposite effect (like causing more power 
usage or more heat, or lower yeilds), causing speed to decrease way to much 
that there is little to no benefit at all. But if there is some technology 
that would increase its performance would make a larger diffrence, so if 
people are looking, then there is a higher chance of finding it, if people 
are to busy looking for ways to enhance serial processors they may be missing 
something, if they look at things diffrently they may find a way to enhance a 
1bit processor, or maybe even find a way that speeds up a serial based 
processor.

One way to imagine it, is if you made an application in hardware, not using 
instructions but using logic gates. That seems tuff, but if you had a 
compiler that would convert your code into a network (and vice versa), then 
you would benefit from it.

One of the big reasons I am here, is that I want to figure out or get a good 
idea how hardware would deal with this, from that I could probably build a 
simulator that would allow me to start experimenting with creating networks 
that act like processors, and start experimenting with creating an assembler 
within the system. If I start compiling one now

> Of course, I hope better processors than Crusoe could be born in the future 
:)

I'm not sure it would be aimed at being better then Crusoe or any X86 
processor, I mean in terms of being a better X86/serial processor, but maybe 
support features and abilities that none of them have before, and be faster 
under certain circumstances. There is potential, but I'm not sure if it will 
come down to it, there maybe another of other Crusoes to follow that will 
take Crusoe even closer, and at some point it may be right next to a 1bit 
processor that they may just convert it over.

Leyland Needham