Re: [oc] Beyond Transmeta...

["Marko Mlinar" <Marko.Mlinar@campus.fri.uni-lj.si>]

> That is true. I've figured latency would be an issue, that is why most
> sequential programs would need to be converted/compiled to a more network
> centric program. I do not think it would be impossible, although
difficulty
> and practicality are still unknown. But one othing that is interesting, is
> that with a network centric program, latency is not much of an issue
because
> things are only updated as needed because everything is persistent,
> sequential programs don't have this ability naturally they normally redo
> something even if they do not need to.
I don't know if I clearly understand your network centric program.
But - suppose that one action occurs. Mouse click or anything. Then you
have to execute let say 100 sequential instructions, ILP = 3.
Let's say your computer would take 1000cycles to do it, and average RISC
50c.
But due to symplicity you could have 50% higher clock speed, but it is still
a lot
slower (assuming communication cost is zero).

> > Do you think such large programs would fit into your network? If they
won't
> > you would have to change most of the network every cycle. That's the
> > bandwidth problem I was talking about. For general applications
> > particular procedures are not often used. Generally you can execute just
> > a few of them simultaneusly, so your network would be mostly idle?
>
> Well, that is an interesting situation. I would think that after a full
days
> work, the system could be left to "sleep" where it would figure out its
> capacity and ability and what is used often by the user and to reconfigure
> the network between a CISC, RISC, or persistent kind of network.
Persistent
> is the highest form, it is equivalent to having multiprocessor system a
> processor for everything (like I said having a network of operations for
> calculating a color of a pixel), while the lowest form is like CISC, where
> you use one processor to do everything, in betweens would be like having a
> processor do graphics and FPU operations and the main processor do other
> operations. The CISC one will not be as fast as a normal X86 kind of
> processor, but it can also decidedly do 2 CISC processors if the user
> commonly use many diffrent programs that are multithreaded in nature..
What
> might be an interesting experiment is to purposely create a network that
> works exactly like an X86/windows system, and then figure out a way to do
> network morphing to anothering kind of system (Like X86/Linux, or PPC/Mac,
or
> Alpha/Linux, etc), and see if that will morph the application and then
> capture that and have that application may work on the actual system....
Kind
> of a funny thought, but it may be possible some day.
That seems good. But I am still worried about latency and speed.

Marko