Can a RAM-less computer run?

[xilman]

I'm with LaurV on this one --- it depends on how you define RAM. If a machine is Turing-complete it can emulate any other Turing-complete machine, including one which contains RAM. So the question boils down to the definition of RAM.

Personally I'd describe the single tape of a classical Turing machine as RAM. I would also regard disks, tapes, mercury delay-lines, magnetic cores, etc as RAM in that they are all forms of read-write storage which may be accessed in (apparently) random order.

[Dubslow]

Originally Posted by axn

Cache is, by definition, not visible to software.

Ah. Thanks for the edification.

[LaurV]

Originally Posted by cheesehead

No need to store anything?

From where, exactly, are you pulling the "1" and "2" values

Two rows of dip-switches?
And a couple of boolean gates to process, plus a row of LEDs to display the result in binary?

You all are spoiled by keyboards, touchscreens and UWUZWUXVGA displays....
(eventually 3D?)

edit: these "computers" had no ram...
edit2: the systolic arrays also have no ram, in fact they do not "store" any data, the "data" is always jumping from one place to the other, with each clock... and you can't say that such tools can't compute too much....

[axn]

Since the OP is asking for solutions based on "general purpose CPU", some of these discussions seem moot.

Anyway, rather than get into the minutiae of computer architecture, I'll leave you with this. (The article is full of linky goodness)

[xilman]

Originally Posted by axn

Since the OP is asking for solutions based on "general purpose CPU", some of these discussions seem moot.

Anyway, rather than get into the minutiae of computer architecture, I'll leave you with this. (The article is full of linky goodness)

I would argue that a UTM is by far the best definition of a general purpose CPU available.

[only_human]

When I worked at Northrop, we had 8085 computers designed for airborne use that had a ROM monitor that worked perfectly without any RAM. The ROM monitor had the normal debug functions. Later we tried doing the same with an 8086 and it was mostly successful but I couldn't get the goto address instruction to work. Oh it could successfully get the hex address from user input but I had trouble changing the full address without using stack. Changing the CS register directly didn't seem reliable even with a known IP offset. I was considering various hacks but I then transferred to a department that was in imagery transmission instead. I think the goal of having the monitor function without any RAM installed was dropped.

[jnml]

Originally Posted by jasong

Mind you, when I ask this question, I'm not talking about a general purpose computer. But I am talking about a general purpose cpu.

Basically I'm wondering if it's possible to boot a cpu and, without having any RAM, and without thrashing the hard drive or flash drive, have it do useful work. In my head I have a mental vision of a cd that boots into truly bare-bone Linux and runs distributed.net stuff, but you can come up with your own idea if you want.

A general purpose CPU executing programs w/o RAM is feasible.

What is RAM used for in an universal machine, like the ones etched in silicone CPUs? For storing state. State, implemented in RAM or whatever, but not the specific implementation is the important thing. W/o state there's no general computing possible. And RAM is just one of the options for having state.

As a proof, there's for example the Parallax Propeller chip. It has RAM, but that RAM is not required for running programs in the Propeller. Propeller's RAM is kinda external, like a very small flash disk, in a way.

The Propeller CPU (there are 8 actually on the chip) has 512 registers and no (directly accessible) RAM (above the hub memory discussed above). The program running in this "cog" (that's how Parallax calls it) is completely stored in the CPU registers. So is any other _state_ - as long as hub memory is not used (that's really just an option), mainly for too big problems to fit in the 2 kilobytes == 512 32bit cog registers.

BTW: The Propeller has experimental (in beta now) GCC support. It runs via a small kernel stored in the RAM-less cog, which interprets opcodes stored in the hub memory.

[xilman]

Originally Posted by jnml

A general purpose CPU executing programs w/o RAM is feasible.

What is RAM used for in an universal machine, like the ones etched in silicone CPUs? For storing state. State, implemented in RAM or whatever, but not the specific implementation is the important thing. W/o state there's no general computing possible. And RAM is just one of the options for having state.

As a proof, there's for example the Parallax Propeller chip. It has RAM, but that RAM is not required for running programs in the Propeller. Propeller's RAM is kinda external, like a very small flash disk, in a way.

The Propeller CPU (there are 8 actually on the chip) has 512 registers and no (directly accessible) RAM (above the hub memory discussed above). The program running in this "cog" (that's how Parallax calls it) is completely stored in the CPU registers. So is any other _state_ - as long as hub memory is not used (that's really just an option), mainly for too big problems to fit in the 2 kilobytes == 512 32bit cog registers.

BTW: The Propeller has experimental (in beta now) GCC support. It runs via a small kernel stored in the RAM-less cog, which interprets opcodes stored in the hub memory.

One could reasonably claim that a set of addressable registers is RAM.

[jnml]

Originally Posted by xilman

One could reasonably claim that a set of addressable registers is RAM.

Of course it is. Either the Propeller can be viewed as RAM-less or register-less. That's the point - asking if a general CPU can live w/o RAM is not a good question b/c that's an implementation detail.

[jasong]

This forum is so awesome. Most people who I know in meatspace who might be able to answer this question would probably ask a version of,"Who frickin cares?"