8-bit or 16-bit Architecture for a Hobbyist Computer

[LtIvan]

I am thinking of building a basic computer, similar to Ben Eater's 8-bit computer. I may be building it on a PCB. However, I intend to aspire to 16-bit bus computer, because it will allow larger numbers, larger computations, larger memory, because the address is larger.
However, is it worth it? 2^9-1 = 511, which is not a practical range to work with. I want my computer to perhaps handle large problems; so, it may be useful.

16-bit Architecture:
Advantages:

• Larger Numbers,
• Larger Computations,
• Larger Memory, because the addresses are larger, which allows more memory¹,
• Thus, it is more 'practical' and useful.

Disadvantages:

• Potentially Expensive,
• Large Physical Volume,
• More Complexity, which may allow more errors,
• More Power Consumption, which may require a complex power system,

Thus, I offer the enquiry to you.

Ic þancie eower,

¹ Yes, there are ways to increase memory for 8-bit computers, but I am suggesting generally.

 

[ISamson]

Are there any specific projects you would like to attempt that we could focus our opinions on?

 

[ISamson]

It really depends on the projects.
I have an Arduino UNO R3 and it is based on an ATmega328P microcontroller. Ah, do you want a microcontroller or microcomputer?
The Arduino UNO has 8-bit processing and seems alright to me. As I said, it really depends on the project. If you want fast computing and good processing then get the most powerful option. If you want to control many input/output sources then...it still depends. My current answer to the poll is... it depends...

 

[Mark44]

I am thinking of building a basic computer, similar to Ben Eater's 8-bit computer. I may be building it on a PCB.

If the goal is to simply build a computer from scratch, then using an 8-bit processor is reasonable, and might be simpler than working with a 16-bit processor.

However, I intend to aspire to 16-bit bus computer, because it will allow larger numbers, larger computations, larger memory, because the address is larger.
However, is it worth it? 2^9-1 = 511, which is not a practical range to work with. I want my computer to perhaps handle large problems; so, it may be useful.

With 8 bits, the largest number that can be represented is $2^8 - 1$ or 255, not 511 as you wrote. That would give you a range of unsigned numbers 0 through 255, or a range of signed numbers from -128 through 127..
If you really want larger numbers, etc., why not make the jump to 32 bits? The Raspberry Pi board comes with a variety of types of ARM processors, with prices ranging between $5 USD to $35 USD (see https://en.wikipedia.org/wiki/Raspberry_Pi). There are also the Arduino kits that ISamson mentioned.

 

[jedishrfu]

just because the computer is 8 bit doesn’t mean you are limited to 8 bit numbers. You can always use software to add and multiply larger numbers.

 

[rbelli1]

just because the computer is 8 bit doesn’t mean you are limited to 8 bit numbers. You can always use software to add and multiply larger numbers.

I have a project I worked on that used an 8-bit micro-controller. At one point in the calculations I found that 32-bit floats made the programming easier. So 32-bit floats I used. It took a relatively long time and a distressing amount of storage but was fast enough by far.

BoB

 

[ISamson]

I have a project I worked on that used an 8-bit micro-controller. At one point in the calculations I found that 32-bit floats made the programming easier. So 32-bit floats I used. It took a relatively long time and a distressing amount of storage but was fast enough by far.

BoB

8-bit microcontrollers are an average Arduino UNO...

 

[rbelli1]

8-bit microcontrollers are an average Arduino UNO...

True. And they like the micro I used (Cypress psoc3) can do 32 bit numbers. Actually they are only limited to 32 bits out of the box by the libraries available. Add more and you can do arbitrarily large numbers limited only be the flash and ram available.

BoB

 

[rootone]

are ways to increase memory for 8-bit computers, but I am suggesting generally.

The original IBM PC and similar had 8 bit processors, meaning that addressing more than one megabyte of memory was problematic from a programming point of view.
Things started to change with invention of 'overhead' memory hardware. you could have several banks of 1Mb which could be 'paged', but still was tricky.
When 16bit processors arrived it got a lot less messy.

I don't know what the details are for your project but I suggest go the 16bit way.
Parts are really not that much more expensive.

 

[rbelli1]

The original IBM PC and similar had 8 bit processors

They had 16-bit Intel 8088 processors with an extra 4 bits of address that could used via segment registers to get to 1MB of memory. Beyond that required external bank switching.

The 80286 added 4 additional bits of address that could be accessed by using additional features of the address unit which is an early version of an MMU. This gives 16MB natively still with a 16-bit CPU.

Finally the 32-bit 80386 comes out and addressing in user mode software gets "less messy" but the underlying MMU gets tons "more messy" for the OS.

Future generations (486, 586{Pentium}, Core, 64-bit, etc.) complicate the situation for the OS even more but provide GB and then TB address spaces.

BoB

MMU = Memory Management Unit

 

[rbelli1]

I don't know what the details are for your project but I suggest go the 16bit way.
Parts are really not that much more expensive.

32-bit is available too at a similar price point to the 16-bitters. Unfortunately all (as far as I can find) of the 32 bit DIP micros have been discontinued. 16-bit DIPs are still widely available. Some of these more capable micros are available Arduino ready.

Many of the manufacturers (NXP, ST, Cypress, TI, Silabs, FTDI, off the top of my head) offer their own ecosystem of beginner and wallet friendly boards.

BoB

Edit: Microchip makes some 32-bit DIP micros. The ARM based DIPs are discontinued.

 

[ISamson]

If the goal is to simply build a computer from scratch, then using an 8-bit processor is reasonable, and might be simpler than working with a 16-bit processor.
With 8 bits, the largest number that can be represented is $2^8 - 1$ or 255, not 511 as you wrote. That would give you a range of unsigned numbers 0 through 255, or a range of signed numbers from -128 through 127..
If you really want larger numbers, etc., why not make the jump to 32 bits? The Raspberry Pi board comes with a variety of types of ARM processors, with prices ranging between $5 USD to $35 USD (see https://en.wikipedia.org/wiki/Raspberry_Pi). There are also the Arduino kits that ISamson mentioned.

https://en.wikipedia.org/wiki/Arduino
:)

 

[ISamson]

8-bit computers and electronics/software are most popular. Most computers function on it.

 

[Mark44]

8-bit computers and electronics/software are most popular. Most computers function on it.

?
If by "computer" you mean desktop computer or laptop or tablet, what you said isn't true at all. Most computers that are sold right now are 64-bit computers, meaning that the registers in the processor are 64 bits wide. My Apple II computer that I bought in 1980 was an 8-bit computer, but all of the ones I've had since then have been 32-bit or higher machines.

 

[CWatters]

One option would be to build using the bit slice approach. Initially build it as an 8 bit then later double up the hardware to 16 bit if you still feel keen on the project.

 

[.Scott]

As @jedishrfu posted, the problem will not be with the arithmetic range - at least not directly.
In the same way that decimal digits don't limit you to 0-9, bytes won't limit you to 0-255 or -128 to -127. You just keep track of the carry bits, etc.

What is much more important is the address range. You want to be able to access much more than 256 bytes of data and program space.
This can be done with a 8-bit instruction length, either with paging, offset addressing, indirect addressing, multibyte instructions, or some combination.

 

[CWatters]

Ben Eater's page for reference..
https://eater.net/

 

[willem2]

What is much more important is the address range. You want to be able to access much more than 256 bytes of data and program space.
This can be done with a 8-bit instruction length, either with paging, offset addressing, indirect addressing, multibyte instructions, or some combination.

Note that all 8-bit processors had an 16 bit address bus, allowing 64 kByte. Having an adrress space of 256 bytes is really too limiting.
Having an 8-bit bus made execution really slow. An instruction like ld a, [memory] would take 4 memory cycles of at least 3 clockcycles each on an 8080 or z80.
1 reading the opcode
2 reading the low byte of the memory address
3 reading the high byte of the memory address
4 reading the contents of the memory location.
Having some 16 bit address registers (or at least one) is really useful, and at least the program counter would have to be 16 bit.
A full 16 bit ALU would increase the transistor count by a lot however.

 

[rcgldr]

8080 Z80

These processors have some 16 bit math, in the form of DAD, a 16 bit add, HL = HL + BC or HL = HL + DE. There's also an 16 bit increment "INX", 16 bit decrement "DCX", and a load "LXI" instruction. A later 8 bit processor, the Z8, has 128 or 256 8 bit registers, and any even/odd pair could be used as a 16 bit address for memory access instructions.

The 6502 was used in the Apple II, Comodore ?, and Atari 400/800/65XE/135XE "8 bit" home computers. Other than PC, it only has 8 bit index registers and an 8 bit stack register (the stack is located at addresses hex 100 to hex 1FF).