The creation of the IBM PC - a review

[RetroAND]

Good afternoon,

I would like to open a detailed thread about the 8-to-16 bit transition at IBM hardware during the early 80s, that had the PC and the PC-XT as a result. My statements written in here would be the result of observation at hardware, as well as pictures and other documents. In case of requesting more information or to inform of a discrepancy, feel free to answer. I won't put everything in an inmense post, but I will be publishing my knowledge in parts.

However, it would be very presumptous of me to just feed it all, so I thought a Q&A approach would work better. In case I don't know the answer, I might be able to relay it.

Please, could you tell me what would you like to know?

Thank you very much in advance.

 

[Zare]

Hey, I have an instant one; I was engaged in a discussion on a comment thread of Youtube video where the video author and myself parted ways on the bitness of PC.
The author specifically claims IBM PC was a 8 bit machine at start, because of 8 bit peripheral and memory bus.
My opinion is that IBM PC was never designed as 8 bit machine in prospect, but for current cost cutting reasons.

Digging through IBM material of the day, they never specifically refer to bitness of the machine, but of course, bit width of the CPU vs bit witdh of the bus.

So, I would like to know, whether it is correct to call IBM PC an 8 bit computer.

 

[RetroAND]

Hello @Zare ,

As in many aspects of the PC, the CPU was constrined by elements derivated from inheritance from the Datamaster. They wanted it to be as compatible to the existing expansion bus at the same time the intel peripherals, the 8253, 8255, 8259 and whatever the DMA device was.

That being said, the PC was never an 8-bit computer as it has a 8088 at its heart. However the main reason to stick to the 8-bit accesses was the foresaid compatibility. It is true that this decision also managed to keep costs lower, but this is more of a byproduct than a cause.

The PC was not the only or the first IBM computer to employ x86 processors. The first x86 machine from IBM was the Displaywriter, which featured a full-fledged 8086. This CPU is also found in a PC relative, the 5247.

My justification for this answer is because during my ongoing investigations about the Datamaster line of computers, I contacted Dr. Bradley and he kindly answered many questions I made him. Others are from my own observation. He told me the choice was made basically with the MCS-85 peripheral ICs in mind, because as the project was constrained to a single year, they wanted to work on parts they had already plenty of experience.

I hope this clears your doubt.

Regards

 

[lowen]

So, I would like to know, whether it is correct to call IBM PC an 8 bit computer.

Motorola called their 68000 processor a 16/32 bit processor; registers are 32 bits, but the ALU is 16 bits and the data bus is 16 bits.

The 8088 is an 8/16 bit processor using the same reasoning about the data bus. The PC and XT are 8 bit data bus designs with an 8/16 bit processor. It is a matter of semantics as to which bittedness 'counts.'

The Z80 could be termed a 4/8 bit processor if ALU size is considered. Or an 8/16 if some of the 16 bit operations with the HL register pair are considered. Bittedness is even more fuzzy for the Z280, which can be configured with an 8 bit data bus or a 16 bit data bus, but still uses 8 bit opcodes but 8, 16, and even 32 bit register (the 32 bit accumulator DEHL is used in multiplication and division operations).

The Pentium is an oddball, a 64/32 bit processor, with a 64 bit data bus and a 32 bit architecture. The 64 bit data bus is to feed the superscalar operation of the processor.

 

[jlang]

I've always used the width of the instructions to determine the "bitness" of a system.
Data bus, ALU, registers can all be smaller or larger.

By my logic the Mc68000,8088, 8086 are all 16 bits and that just doesn't sound right.

 

[Ruud]

The main registers that can output data to the data bus are 16 bits in case of the 8088. That it is done in two steps in some cases doesn't matter IMHO. In case of the Z80 it can only output 8 bits. That it has some internal 16-bit registers doesn't matter. But again, this is my personal opinion.

 

[RetroAND]

In any case, if they wanted to employ the 8086, they would have done so as they already had experience with the IC. The 8088 was just the ideal part that would allow easy reuse of most of the Datamaster architecture without adding effort. However, they would later make that extra effort for the 5247... strangely the expansion bus of that computer still had an 8-bit data bus, so unless they designed it to ignore half the memory map, there might be some kind of stepping mechanism. Maybe we should take a look at the XT/286 in order to guess what was done in the 5247.

 

[Zare]

Thank you for all your answers.

What Ruud says is for me the most important angle, the 8088 deals with 1 bit machine words.

Besides, there were 16 bit mem/peripheral bus PCs from clone vendors before AT. Like M24.

And then there were PC builds in early 386 age which used 32 bit CPU, 16 bit mem/io bus with all the IO cards being 8 bit.

One thing I find important is the target market, the goal. 8 bit home computers were heavy on the IO. All about the graphics, the sound, but very limited data processing capability.
IBM made PC from business point of view. Having a fast video system and sound was not the goal. The goal was to input the data, process it and show it on screen, print it to paper.

Aforementioned M24 had different market in mind, graphics workstation, hence it had 16 bit memory and graphics card and 8086 CPU to handle it. The platform is still made of COTS logic chips.

If IBM decided not to take Datamaster route and go directly to 8086 the platforms cost would be higher but still in the brackets. But then all the extra oomph would be wasted.

 

[RetroAND]

If you take a look at the early drawings and specs of what they expected the PC to be, you will see that they intended to employ the 8" FDC card as-is (with no important modifications other than form factor or connector removal). They also initially intended to externalize the original Datamaster video subsystem - it can be seen that the expected pixel clock was 18.432MHz and no RAM on it - so this points that it would be DMA-based and therefore it would have carried an 8275. What they expected of the video system is to be able to be easily replaced - that was one of the best choices they could do as it would delay obsolescence and give it longer product life.

 

[carlos12]

From a practical point of view (not literally), the 8088 and 8086 are identical. Yes, there are minor differences such as the different size of the prefetch queue or that the have one different pin. But in practical terms they are 100% compatible. The 8088's bus is 8 bit, right, but it behaves as a 16 bit bus because the fact it uses two trips to send a word is completely transparent to both the programmer and the end user. This just involve the 8088 being slightly slower than the 8086. So an 8088 running at 10Mhz would more or less equate an 8 Mhz 8086. To me in my personal opinion the 8088 is a full 16 bits processor.

 

[RetroAND]

From a practical point of view (not literally), the 8088 and 8086 are identical. Yes, there are minor differences such as the different size of the prefetch queue or that the have one different pin. But in practical terms they are 100% compatible. The 8088's bus is 8 bit, right, but it behaves as a 16 bit bus because the fact it uses two trips to send a word is completely transparent to both the programmer and the end user. This just involve the 8088 being slightly slower than the 8086. So an 8088 running at 10Mhz would more or less equate an 8 Mhz 8086. To me in my personal opinion the 8088 is a full 16 bits processor.

Of course, if we had a microprocessor with a serial interface, it wouldn't make it a 1-bit CPU. The interface does not define the internal architecture of the CPU.

 

[lowen]

This question is one of semantics for sure, but the PC design external to the 8088 is an 8 bit design. An 8085 could be plugged in without a lot of modifications (banking logic to generate 20 bit addresses, mostly); a 68008 could be plugged in just about as easily with a bit of glue. Then you'd have an 8/16/32 bit computer. (8 bit data bus; 16 bit opcodes; 32 bit operands.)

The 8088 has 16 bit opcodes and operands; it's 16 bits internally no question. But the IBM PC and XT are both 8 bit computers with 16 bit processors; hybrid designs, bridging the 8 bit to 16 bit gap.

The PC/AT is the first full 16 bit design in the PC-compatible world.

Similarly, 80386SX designs: 16 bit computers with a 32 bit processor (my IBM PS/2 model 30/286, for instance, has a 386SX processor upgrade; it's a 16 bit computer with a 32 bit processor). Hybrid; I don't know why that concept is so strange for people. Part of the computer is one bittedness (ROM and RAM, for instance) and another part is a different bittedness.

The original PC is both 8 bit and 16 bit.

 

[RetroAND]

The PC/AT is the first full 16 bit design in the PC-compatible world.

Actually, while true for total compatibility, from a architectural design perspective this claim has become challengeable recently.
The first IBM device with a PC chipset, an 8086 and an ISA-like expansion bus was the 5247 (1982).
I imagine they took from there and made a port reassignation. Also, with the HDD controller board, the 5247 also has two DMA devices.

 

[Zare]

The original PC is both 8 bit and 16 bit.

I ask because of this video

However sat at your modem machine, you might not released it started life as an 8bit micro in the early 80s, and has been built on top of 40+ years years of development.

From the description verbatim.

Did the IBM PC start the life as 8 bit micro? My opinion, is that most certainly it did not.

 

[lowen]

Actually, while true for total compatibility, from a architectural design perspective this claim has become challengeable recently.
The first IBM device with a PC chipset, an 8086 and an ISA-like expansion bus was the 5247 (1982)..

Well, I'm also not considering the Tandy 2000, which is one of the first full 16 bit MS-DOS computers, but it is not hardware compatible with the PC nor was it an IBM design. The Olivetti M24/AT&T 6300 wasn't considered, either.

Nor am I considering Seattle Computer Products' S-100 8086 systems, which predate the 5150.

But I guess the core of my point is that while the IBM PC ran 16-bit software and can be considered a 16 bit computer at that level of abstraction, the hardware implementation tells a different story, and at the hardware level it's an 8 bit computer with a 16 bit processor.

IBM must simply have thought the bittedness didn't really matter, or they would have mentioned it. Maybe there was disagreement internal to IBM even as to the "true bittedness" (whatever that means) of the machine. It's still being debated today, after all.

But I'll add this point; what about that 4004 system posted about in September of 2024 that runs a MIPS R3000 emulator and boots and runs 32 bit Linux? It's running 32 bit software, but on a 4 bit computer. Does that make it a 32 bit computer? (EDIT: maybe it does, and maybe it doesn't: it depends upon what level of abstraction we work at. I'm actually ok calling the complete unit, including the MIPS emulator 'microcode' module, a 32 bit computer even though it's running on a 4 bit microengine. It's all in the semantics.)

There are other hybrid designs; the best known in vintage circles is probably the LSI-11, which is a really interesting hybrid design that runs 16 bit code. 8 bit internal data path, 18 bit microcode width, 11 bit microcode addressing, and 16 bit external addressing.

And then there's the famous LGP-30 of "The Story of Mel, a Real Programmer" fame; bit serial architecture with a 31 bit word.

 

[RetroAND]

Well, I'm also not considering the Tandy 2000, which is one of the first full 16 bit MS-DOS computers, but it is not hardware compatible with the PC nor was it an IBM design. The Olivetti M24/AT&T 6300 wasn't considered, either.

Nor am I considering Seattle Computer Products' S-100 8086 systems, which predate the 5150.

But I guess the core of my point is that while the IBM PC ran 16-bit software and can be considered a 16 bit computer at that level of abstraction, the hardware implementation tells a different story, and at the hardware level it's an 8 bit computer with a 16 bit processor.

IBM must simply have thought the bittedness didn't really matter, or they would have mentioned it. Maybe there was disagreement internal to IBM even as to the "true bittedness" (whatever that means) of the machine. It's still being debated today, after all.

But I'll add this point; what about that 4004 system posted about in September of 2024 that runs a MIPS R3000 emulator and boots and runs 32 bit Linux? It's running 32 bit software, but on a 4 bit computer. Does that make it a 32 bit computer?

There are other hybrid designs; the best known in vintage circles is probably the LSI-11, which is a really interesting hybrid design that runs 16 bit code. 8 bit internal data path, 18 bit microcode width, 11 bit microcode addressing, and 16 bit external addressing.

You can ignore all those easily because they are third-party designs. But if the same manufacturer (IBM) made it, and it predates other designs, you can't brush it off that easily.

Both the 5324 and the 5247 from the Datamaster line of products show feedback from the PC prototype or its proposals. Some of those didn't end up in the 5150 but have a vestigial form in those models.

Actually, the 5247 is what the 5150 could have been if design decisions were made over different criteria. Please, I invite you to take a look at it here.

 

[lowen]

Did the IBM PC start the life as 8 bit micro? My opinion, is that most certainly it did not.

First, thanks for the link; love this guy's accent.

He obviously ascribes to the 'data bus width is "true bittedness"' school of thought. Within that framework his statement is correct. And I'm ok with that.

Does this 8 bit data bus machine have a 16 bit word size? Sure it does; that makes it a hybrid 8/16 bit machine.

I'm fully ok with both points of view, and due to the different level of abstraction the two points of view don't have to be contradictory.

Actually, the 5247 is what the 5150 could have been if design decisions were made over different criteria. Please, I invite you to take a look at it here.

But a System/23 isn't a PC-compatible. It is a fascinating machine for sure, and it is a good look at what could have been had the developers waited just a little bit longer. But the external market forces would not have waited, and we might be in a very different world today, where the TRS-80 Model 16 with its different take on a hybrid 8/16 bit architecture took over and we all were running a descendent of Xenix.

Man, I truly wished they had waited, because a world where the TRS-80 Model 16 set the standard would have been fabulous. (The 16 was my first Unix system, and I still use such systems daily, using Windows only when absolutely necessary)

 

[RetroAND]

First, thanks for the link; love this guy's accent.

He obviously ascribes to the 'data bus width is "true bittedness"' school of thought. Within that framework his statement is correct. And I'm ok with that.

Does this 8 bit data bus machine have a 16 bit word size? Sure it does; that makes it a hybrid 8/16 bit machine.

I'm fully ok with both points of view, and due to the different level of abstraction the two points of view don't have to be contradictory.


But a System/23 isn't a PC-compatible. It is a fascinating machine for sure, and it is a good look at what could have been had the developers waited just a little bit longer. But the external market forces would not have waited, and we might be in a very different world today, where the TRS-80 Model 16 with its different take on a hybrid 8/16 bit architecture took over and we all were running a descendent of Xenix.

Man, I truly wished they had waited, because a world where the TRS-80 Model 16 set the standard would have been fabulous. (The 16 was my first Unix system, and I still use such systems daily, using Windows only when absolutely necessary)

I never said a Datamaster was a PC compatible. What I am stating is, while the S/23 greatly influenced the PC, it also received feedback from the 5150 development.

For instance, the 5324 got its keyboard detached and also got a 110V line for the monitor in all regions, as stated for the PC prototype notes in 1980.

And as the other one, it is an equilibrated mixture between a Datamaster and a PC.

 

[Zare]

He obviously ascribes to the 'data bus width is "true bittedness"' school of thought. Within that framework his statement is correct. And I'm ok with that.

Well I find this school of thought inapt for PC affairs. Apple, Atari, Commodore did not fluidly transcend the 8/16 line. They ran the max with 8 bit platforms and then created totally new ones. You can say here is Apple II and here is Mac. You can see the evolution between 8 bit platform and 16 bit platform.

For PC it is rather fuzzy. Would a 640kB 5160 with EGA compare better to a Mac or to Apple II?

How does a 10 MHz V20 + VGA card + Sound Blaster Pro compare to a C64?

The reason why PC remained with 8 bit data bus for a rather long time - because it wasn't important at all. The 8 bit architecture is a key factor in those other machines, a limiting factor.

 

[lowen]

Well I find this school of thought inapt for PC affairs. Apple, Atari, Commodore did not fluidly transcend the 8/16 line. They ran the max with 8 bit platforms and then created totally new ones. You can say here is Apple II and here is Mac. You can see the evolution between 8 bit platform and 16 bit platform.

Apple's IIgs has entered the chat.

As much as I don't care for Apples, the IIgs smoothly bridged the 8/16 bit gap, with even a upgrade board for existing IIe machines being available.

But Atari, Commodore, and Apple all decided 16 bits was too small and limiting, and went big with the 68K machines, and had good success with them.

And whether you agree with that mindset or not, there is a large contingent who do agree with it. As for me, I don't think it's worth arguing about; both points of view are equally valid as far as I'm concerned.

What I am stating is, while the S/23 greatly influenced the PC, it also received feedback from the 5150 development.

True enough.