why did 70/80/90 only have 1 cpu if cpus where so slow?

[aqarwaen]

so my question is why did 70/80/90 only have 1 cpu if cpus where so slow?if i understand most computer only had 1 single main cpu.
for example why it was possible use for example multiple Intel 4004
instead single main cpu to make faster computers.
if money was not issue,what would prevent 4 same cpus running in single system?

[DavidAlfa]

Money was definitely an issue!
Technology was very primitive, slow buses, limited complexity...
Multicore systems existed, but only on very expensive mainframes.

[abeyer]

if money was not issue,what would prevent 4 same cpus running in single system?

As pointed out already, money absolutely was an issue. You could buy multiprocessor machines but they were expensive. (the CM-1 was commercially available in the mid-80s with up to 64k individual processors... but it wasn't exactly competing in the same market as a 4004  )

Also, you can't just throw two cpus on the same memory bus and let them duke it out. They need to be designed to share a bus, or need to have separate memory spaces and some sort of communications channel. Either way adds hardware and software complexity and cost.

[nightfire]

Back in the 90s, I was diving deeper into computing, and was in '99 proud owner of a Tyan Tomcat 4 with dual Pentium 233MMX and a whopping 128 MB RAM.
Was able to run only under Windows NT and later FreeBSD- where I had to compile myself the SMP kernel.

The SMP stuff was mostly the issue: It required some additional hardware, the hardware designers really had to be careful about timings of the various components, and then came the big issue: The software.
Most operating systems were not suited for SMP, and those that were, imposed quite a overhead over the operations there. Like having global locks for interrupts from the network card etc.
So usually the 2nd CPU gave about 50% performance increase, and 4-way boards in servers were only deemed fit for database operations, where massive parallelization of queries in big databases could be made.
Also the user software had to be fit for SMP to have some real performance boost, otherwise one would only have the advantage of being able to run a few more singlethreaded programs on the box simultaneously.

[Kleinstein]

Using more than 1 CPU needs extra effort. This is extra hardware to transfer data and also extra support from the OS or compiler to actually use more than 1 CPU. So the return for more than 1 CPU is not that great. AFAIR Linux only started to support more than 1 CPU core from around 2000 on and initially the support was somewhat limited. So much software could not use a 2nd CPU, even if present.
Even today 2 CPUs does not mean 2 x the speed.

There were a few systems with more than one CPU. A have an old CPU board with 2 x MOS6502A from the early 1980s / late 70s  (from a Comodore external dual floppy like 4040). So there were a few cases using more than 1 CPU, but that was more like the exception for a special task. AFAIK it is a somewhat crazy system with 2 CPUs capable of 2 MHz run at 1.5 MHz interleaving the access to a common memory (at least the RAM, the ROM part could be separate).
Other computers had 2 different CPUs like 6502 and Z80 to run different software, but usually not at the same time.

[Bud]

Jack Tramiel, the head of Commodore, said "I build computers for the masses, not for the classes".
Does this explain ?

[rstofer]

A couple of the Control Data CPUs had multiple processors.  The most powerful was CDC 6500, released in 1967.  In addition to 2 processors, it had 10 peripheral units to handle primarily IO and such but could also execute user code.

https://en.wikipedia.org/wiki/CDC_6000_series

There were later series like the 7600 and the 7700 was just two 7600s running together.  Very expensive!

Yes, they were expensive since a single flop probably required 2 discrete transistors.  Sixty bit registers were expensive.

Core memory wasn't very fast and it seriously limited the speeds for early machines.

I understand the Apple M2 Max chip has over 67 BILLION transistors on the chip.  I have one and it seems to scoot pretty well.

https://9to5mac.com/2023/01/31/m2-pro-max-vs-m1-pro-max

Can you even imagine a discrete implementation?

[Benta]

"why did 1930's Ford A not have turbocharging?"
"why did "Spirit of St. Louis" not have jet engine?"
I get a feeling of some chatbot here...

[floobydust]

In that era memory was also quite slow, expensive, and DRAM needs refresh - all of which gets in the way when you add cores.

[SiliconWizard]

Plus, multi-CPU designs with the CPUs they had back then was highly inefficient anyway, due to bottlenecks everywhere, from busses to RAM to a lot of other stuff, plus very few software developers who could have used that well.

That said, some even modest machines had two CPUs, like a main CPU and a secondary one for peripherals and such. But each was dedicated to its own tasks rather than potentially sharing the same tasks like we do now.

[Brumby]

... you can't just throw two cpus on the same memory bus and let them duke it out.

I didn't get past the title of this thread before this thought was screaming around my head.

They need to be designed to share a bus, or need to have separate memory spaces and some sort of communications channel. Either way adds hardware and software complexity and cost.

Indeed.

Asking this question is, as others have stated, an indication that there is a distinct lack of understanding on how technology evolves.

You could just as well ask why the Sopwith Camel didn't have a delta wing design.

[Kleinstein]

Many of the early private computers in the 1970s/1980s were not really made to be fast. E.g. they still sold lots of 1 MHz 6502 systems when a 2 MHz version was available for not much more money.  Adding a 2nd CPU add much most complications (e.g. SW compatibility) than just increasing the clock. So a faster clock was the more obvious way to more speed, not a 2nd parallel running CPU.

Sharing a single task over 2 or more CPUs is by no means easy from the software side. Even now only some software can make real use of mulitple CPUs is the modern PCs.
Without the SW support the second CPU is pretty useless and only adds costs and complications.
The early mainframes with multiple CPUs often did separate tasks (e.g. different users) on the added CPUs, as this was the easier part and the coputers were anyway shared between users.

To a certain degree I am wondering why the modern PCs do so many jobs with the main CPUs and not split of more tasks to dedicated parts. It would part with the old PC architecture, but it may be about time for that.

[Siwastaja]

For the same reason multicore performance still sucks: it is very hard to write truly parallel (efficient at that) programs. People think in "do this, then do that" sequences, and reflect this thinking in programming languages. In order to get any advantage of multiple CPUs, you need to write programs like "do this, and simultaneously do that", and the problem then is getting everyone do useful work all the time and not just wait for others to finish their part of the result.

It was way easier to just push performance up, to get job done earlier, so that next job can be started earlier; giving illusion of things being finished simultaneously. Only after physical limitations of how fast things can be done were hit, we had to start thinking how to do things simultaneously.

[m k]

if money was not issue,what would prevent 4 same cpus running in single system?

Check transputer.

[tooki]

so my question is why did 70/80/90 only have 1 cpu if cpus where so slow?if i understand most computer only had 1 single main cpu.
for example why it was possible use for example multiple Intel 4004
instead single main cpu to make faster computers.
if money was not issue,what would prevent 4 same cpus running in single system?

Gotta love questions that start with a premise that isn't even necessarily true, coming from the place of "I don't know anything about that topic, so I'll just make assumptions based on thin air".

Note that the question just said "computer", not "personal computer", so this query technically includes minicomputers, servers, and mainframes, which commonly did have multiple processors.

So there were multiprocessor systems back then, but they were expensive. Very expensive. Thus, "if money was not an issue" then very little prevented one from getting a multiprocessor system back then.

[mendip_discovery]

They existed, mostly on servers. I remember Apple had them on the G4 machines in the early 2000s and it's how they were able to outperform the windoes PCs for graphics for a while. It wasn't cheap. The cpu was always one of the more expensive parts and I remember having a 200mhz Citrix and a few years later the processor speeds had jumped so we were often quite happy with the speed as things moved quite quickly. We also had software and a OS that was rather efficient, not like all the bloat we have now.

[BillyO]

In the 70's Ohio Scientific had the C3.  It had 3 processors.  I'm sure there were others.

In the 80's DEC had many systems with multiple processors (VAX clusters) and even had a PC, the DEC Rainbow 100 that had 2 processors.

By the 90's there were multi-processor systems all over the place and they were quite common.  In 1998 I had a Dell machine under my desk that had two Pentium II processors in it.

[alm]

As a research project, the massively parallel (designed for 256 CPUs, but built with 64 CPUs) ILLiAC IV was already conceived in the fifties, designed in the sixties and delivered in the seventies. although not a successful project on its own, this certainly inspired other development of large multi-processor systems.

[xrunner]

so my question is why did 70/80/90 only have 1 cpu if cpus where so slow?if i understand most computer only had 1 single main cpu.
for example why it was possible use for example multiple Intel 4004
instead single main cpu to make faster computers.
if money was not issue,what would prevent 4 same cpus running in single system?

I was told by my friends back in the day I was crazy for buying a 20 MHz clocked 386 PC. Why would I need such a powerful CPU? 

[tooki]

They existed, mostly on servers. I remember Apple had them on the G4 machines in the early 2000s and it's how they were able to outperform the windoes PCs for graphics for a while.

And that was far from the first. The first dual-processor Mac from Apple* was the Power Mac 9500/180MP from August 1996.

And of course there’s the philosophical question of whether the 1980s-90s Macs with PC daughtercards — a full PC motherboard shrunk down into an expansion card plugged into a host Mac — count as multiprocessor, since they contained a separate Intel CPU just for DOS/Windows to run on. Third parties sold such cards, but Apple also sold “PC Compatible” versions of some Mac models, with such a thing factory installed.

*”From Apple” because the very first dual- and quad-processor Macs were actually licensed Mac clones by DayStar. It worked so well that DayStar actually made the dual-processor CPU daughtercard in the 9500/180MP, and it was DayStar’s multiprocessing API that became the classic Mac OS multiprocessing API.

[pcprogrammer]

There was also the Basis 108 that had a 6502 and a Z80 and if I recall correctly also had a 6809 extension card made for it.

Also the BBC micro was able to get a second CPU unit, and probably some more that offered this multiple CPU setup to allow running the different operating systems.

Sure nothing like what there is now and not intended to speed up processing, but available non the less.

[coppercone2]

they had dual access memory or whatever in some stuff, but the thing is
1) parallel programming problems were and are being solved
2) personal computers were deemed to be a solution to a particular problem (mostly relating to replacing paperwork), people saw that a single core is basically capable of this, since its replacing essentially a type writer and calculators


I would say that most people do not realize the complex engineering heritage of busses/protocols/datapath designs/etc.. fairly obvious digital circuits but all that stuff had to be figured out. And good interconnects and even PCB related stuff. Reliable cheap many layer PCB with internal vias and stuff. IMO people imagined the wire wrap version and thought 'there is no fucking way'. Even PCB trace minaturization is pretty complicated, sure you just etch it smaller, but the amount of trust you get from such fine pin spacings to make these systems portable? You need advanced knowledge of glue, silk screens, etc to make that kind of design decision for something you want to be durable. Tons of pioneers IMO.

If you wanted to implement that kind of density in 1970 on a large scale people would just think "manufacturing problems" and that they are gonna go out of business if they offer a warentee past 2 weeks. Low production for the air force maybe, but they would have a legion of technicians inspecting and checking everything, and probobly the throughput would be so pathetically low that for anything but national defense the 'efficiency' of the business would be beyond dismal. When you look at the advanced military semiconductor tech of the past, 'scared stupid' (rightfully so) comes to mind, because what they got for how much it cost is abysmal, thankfully we figured out how to use some of it.


Like if you never did it, you just think 'etching foils' duh. But if you do it you know that you need the right photo resist. If the awesome high reliability low cost good shelf life DOW corning blue whatever was not around for etching fine geometry, you would be stuck fiddling with the old process for a long time. And developments in PCB manufacturing/glue.

Like look at how dismal BGA reliability was for maybe 15 years when they start to use it for consumer electronics lol, its still a problem, and thats just some mechanical BS. I feel like there is a ton of invisible esoteric technologies that were coincidentally developed in different fields that allow this stuff to be made now. Its common to still hear that a design has 'too many parts' and that the design is viewed as unreliable for that reason.

[SiliconWizard]

I remember having read some 80's articles on multiprocessing and it all seemed like breakthrough stuff at the time and nothing really practical outside of research centers and extremely expensive hardware.
Usually the key issue was quickly diminishing returns due to the bottlenecks of having to share resources. Along with much more complex software design.
That made it all unpractical for personal computers, and even for pro computers outside of maybe "supercomputers".

As I and some others have said, you still had small computers with several CPUs, but each was usually dedicated to its own thing, so while they were still "multiprocessor" systems, that was much, much simpler. They often didn't even share main memory.

But several identical CPUs sharing the same resources in a typical multi-CPU architecture? That was rarely worth it except maybe in niche applications.
Say with a typical "consumer" CPU of the time, you'd be lucky if you could get about 1.5x the computing power of a single-CPU system with a dual-CPU one, all this with much higher cost and much more complex software. While getting 2x the power just by clocking at twice the frequency quickly became possible, for a much lower cost and complexity.

Interestingly, while many-core computing is now the norm, things are running a bit in circles - different types of cores in recent CPUs, instead of all the same, along with dedicated accelerators, are coming back.

[tooki]

Usually the key issue was quickly diminishing returns due to the bottlenecks of having to share resources.

This is a really good point that bears repeating. It’s easy to forget how back then, the CPU was just one of multiple potential bottlenecks all vying for the honor of limiting overall system performance. CPU was one, expansion bus throughout and memory bandwidth were others, but IO (including storage) and IO latency were huge bottlenecks back then (and in some ways, remained so until SSDs took over from hard disks).

Remember when memory access was so slow that we had to set hard disk interleave factors to artificially slow down the throughout of hard disks? :p

[Kleinstein]

In the early days I would consider the lack of suitable software and the still existing complications in using more than 1 core for many problems as the large hurdles than sharing other resources. The other bottlenecks still existed and this was in part limiting the useful clock speed. So the need for a much faster CPU was limited. If needed there were often better alternatives to multiple CPUs to share the load.
A thing was more havine a dedicated FPU and the first steps towards graphics acceleration.

We don't use PCs with multiple cores because it is a very efficient way, but more like because most of the alternatives (high clock, wider words) have reached there limits. The SW side is still limiting in how muliple CPU cores can be used. A 16 core CPU may not be any faster (maybe even slower) than a single core CPU if the SW does allow for parallel execution. The point is more than modern PCs go a bid overboard with just adding more cores that are rarely used.