I assume that the 74LS181 (not 74181) didn't became common....

[yuhong]

until the microprocessor became common, right? It would be funny if the 74LS181 caught on instead of the microprocessor.

 

[yuhong]

(For those that are not well aware, 74LSxxx consumes much less power than the original 74xxx chips)

 

[Hugo Holden]

until the microprocessor became common, right? It would be funny if the 74LS181 caught on instead of the microprocessor.

A 74181 is an ALU, it cannot replace a CPU. It is a useful chip alright, but despite the Boolean logic it performs, it cannot on its own run a sequence of operations in the sense that uP can, it is better told what to do , either by a uP, or other hard wired logic to control it.

It makes a great IC for a calculator, or any time you need to perform the basic math functions on binary logic. It is better regarded nowadays as a "microprocessor support IC". But granted, it is an impressive IC. I have a number of these in my IC's collection.

You can of course make a simple uP free digital computer, with these to demonstrate 16 binary arithmetic operations on two 4 bit words. I think there have been some internet projects demonstrating it in this application. But these days an average programmer can code that into a fairly low generation uP, and still have resources to make it display data on a screen and even make the screen interactive.

 

[jplr]

> It would be funny if the 74LS181 caught on instead of the microprocessor.
For me it's not the same level of integration and it was not as fast as you would imagine. They had to be put in serial plus a carry-ahead CI, plus registers to buffer the output, so it was hard to achieve more than a few megahertz.
There are also microprocessors in slices such as the AMD2900. They were used in many computers and even game consoles.

 

[daver2]

The Honeywell Level 6 computers use the 74181 as an ALU.

The smaller machines had a fully TTL CPU - with the 74181 as the ALU. The later machines went with the AMD2900 bit slice processor - but still utilised the 74181 as an ALU (if I remember correctly). I also think they used the 74181 in another application on the CPU board - just for its addition functionality (possibly subtraction as well).

Dave

 

[MattisLind]

> It would be funny if the 74LS181 caught on instead of the microprocessor.
For me it's not the same level of integration and it was not as fast as you would imagine. They had to be put in serial plus a carry-ahead CI, plus registers to buffer the output, so it was hard to achieve more than a few megahertz.
There are also microprocessors in slices such as the AMD2900. They were used in many computers and even game consoles.

Serial? Few megahertz?

A delay of 44 ns should make it possible to run at 10 MHz with good selection of registerfile. Using the S variant improves the situation yet more.

A variant of 74181 (S or standard) were used in most TTL PDP-11.

 

[jplr]

> 44ns
Yes that's what is in the datasheet for LS181 for 16 bits with carry-ahead.
But that's only the time to propagate a result from inputs to outputs.
That's not a fair comparison with microprocessors.
A fair comparison would include the instruction register, the instruction decode circuit, the pipeline registers, and all buffers for input/outputs on data and memory busses.

The PDP-11/70, one of the last TTL PDP-11, took 2 µsec to process a single operand instruction.

http://www.bitsavers.org/pdf/dec/pdp11/1170/PDP-11_70_Handbook_1977-78.pdf

This includes the time to load the instruction, plus the time to process it.
In the document linked they allude to a duration of 300ns or more for EF time.

 

[MattisLind]

> 44ns
Yes that's what is in the datasheet for LS181 for 16 bits with carry-ahead.
But that's only the time to propagate a result from inputs to outputs.
That's not a fair comparison with microprocessors.
A fair comparison would include the instruction register, the instruction decode circuit, the pipeline registers, and all buffers for input/outputs on data and memory busses.

The PDP-11/70, one of the last TTL PDP-11, took 2 µsec to process a single operand instruction.

http://www.bitsavers.org/pdf/dec/pdp11/1170/PDP-11_70_Handbook_1977-78.pdf

This includes the time to load the instruction, plus the time to process it.
In the document linked they allude to a duration of 300ns or more for EF time.
View attachment 1265498

I think you are mixing up things here. Depending on the architecture and design of the machine instruction decode and fetch may overlap with execution called pipelining. So the speed of the datapath is just one parameter. The amount of parallellism is important.

11/70: The Execute and fetch time is 300 ns with register to register and cache hit. From where do you get the 2 microseconds?
The 11/70 has an internal 33 MHz clock which is divided by 5 creating five time states. During these five time state it enables things on to buses and strobes things into registers. The resulting 150 ns clock drives the microcode which is 64 bit wide. I.e a lot of things can be done in parallell the wider microcode you got. A register to register ALU op takes two microcycles. One fetch and one execute.

 

[Al Kossow]

The 45 and 70 are essentially the same microarchitecture and use 74S logic (fast and power hungry)
DEC stuck with TTL cpus (except the MOS QBus ones) only using 2900s for FPU and CIS accelerators.
Bell's "Computer Engineering" covers this.

There weren't a lot of PDP-11 clones built because of DEC's Unibus patent

The only use of the 74LS181 I can think of in games was the Cinematronics vector game CPU

 

[yuhong]

Serial? Few megahertz?

A delay of 44 ns should make it possible to run at 10 MHz with good selection of registerfile. Using the S variant improves the situation yet more.

A variant of 74181 (S or standard) were used in most TTL PDP-11.
View attachment 1265496

In this case I used the 74LS181, which also consumed less power.

 

[NeXT]

There was a famously large argument regarding weather or not an ALU could be used a s a microprocessor after Four-Phase made it work, however that required its own external glue logic and microcode ROM, plus they ran multiple ALU's in bitslice.

 

[Al Kossow]

It was a patent breaker for the TI Microprocessor patent, not just an 'argument'

The Texas Instruments TMX 1795: the (almost) first, forgotten microprocessor

The first 8-bit microprocessor, the TMX 1795 had the same architecture as the 8008 but was built months before the 8008. Never sold commerci...

www.righto.com

 

[Eudimorphodon]

You can of course make a simple uP free digital computer, with these to demonstrate 16 binary arithmetic operations on two 4 bit words. I think there have been some internet projects demonstrating it in this application.

Obnoxiously hagiographic Steve Wozniak biographies (which is most of them) usually mention the so-called ""Cream Soda Computer" that he and a buddy built in high school; there aren't any schematics or anything else other than some bad pictures documenting this thing, but it's been determined pretty definitively that the heart of it was a couple 74LS181s. Building a simple computer just like this was actually a standard exercise in computer engineering classes for many years and, no, the "Woz" didn't invent the idea.(*)

(* If you read his autobiography you get the idea that he believed that semiconductor manufacturers were just churning out silicon chips as some kind of blind cargo-cult exercise with no idea what they were good for until he came along and read the datasheet that they somehow inexplicably assembled despite having no idea what these mystery blobs were for.)

Anyway, focusing solely on how fast the 74181 and related bitslice components were on their own is kind of missing the point when comparing them to microprocessors. Yes, up until the end of the 1970's, and even into the early 80's, you could build a faster/more powerful computer out of bitslice components than you could out of microprocessors, which is why they continued being used in Minicomputers and early high-powered workstations like the Xerox Alto/Star. But to make them that fast you had to pair them with boards-worth of expensive bipolar memories and high-speed logic; yes, you can make a toy that adds two numbers together that fits on a big postcard:

("Cream soda")

But making a real general-purpose computer out of these is dozens and dozens of chips which, again, were very expensive at the time if you wanted "fast". Microprocessors were not about sheer speed (in the beginning), it was about making something that was "useful" while being optimized to use cheap supporting circuitry. These days you can make a TTL computer like the Gigatron on a board the size of a piece of paper which has pretty impressive capabilities, but the heart of it is a positively huge and impossibly fast for the 1970's microcode ROM, and its "RISC" architecture also makes it extremely RAM intensive compared to a more complex CISC design. Built out of 1970's components the Gigatron would be an order of magnitude or two more expensive and also about the same degree slower. The same amount of PCB space would go *much* further with an 8080 or whatever on it and also perform better. Up your space and budget to filling a box the size of a small file cabinet with parts then, sure, you can build something out of bitslice components that will outrun it.

 

[jlang]

It was a patent breaker for the TI Microprocessor patent, not just an 'argument'

The Texas Instruments TMX 1795: the (almost) first, forgotten microprocessor

The first 8-bit microprocessor, the TMX 1795 had the same architecture as the 8008 but was built months before the 8008. Never sold commerci...

www.righto.com

That's a good read thanks

 

[NeXT]

It was a patent breaker for the TI Microprocessor patent, not just an 'argument'

The Texas Instruments TMX 1795: the (almost) first, forgotten microprocessor

The first 8-bit microprocessor, the TMX 1795 had the same architecture as the 8008 but was built months before the 8008. Never sold commerci...

www.righto.com

Argument in that it did perform microprocessor tasks as to prove evidence of previous work but in such a way that it operated more as a multi-chip processor...and the argument therein of what does and does not qualify in that context, considering no CPU before or after could really operate independently without external glue (but the F-11 comes really close). Ken's opinion on that is hard to dispute in that the AL-1 seemed to work for the demo due to playing a song and dance and hiding that it was getting away with some processor-like functions externally through the rom and latches. How TI didn't catch that is a mystery to me and as a result I had to change my own opinion on the AL-1 being the world's first single-chip microprocessor, even though it is structured more like an ALU. I can't dispute his evidence or his findings beyond a petty "well, the entire industry was fluid at the time and standards were still being finalized, so it was different back then"

 

[yuhong]

Obnoxiously hagiographic Steve Wozniak biographies (which is most of them) usually mention the so-called ""Cream Soda Computer" that he and a buddy built in high school; there aren't any schematics or anything else other than some bad pictures documenting this thing, but it's been determined pretty definitively that the heart of it was a couple 74LS181s. Building a simple computer just like this was actually a standard exercise in computer engineering classes for many years and, no, the "Woz" didn't invent the idea.(*)

(* If you read his autobiography you get the idea that he believed that semiconductor manufacturers were just churning out silicon chips as some kind of blind cargo-cult exercise with no idea what they were good for until he came along and read the datasheet that they somehow inexplicably assembled despite having no idea what these mystery blobs were for.)

Anyway, focusing solely on how fast the 74181 and related bitslice components were on their own is kind of missing the point when comparing them to microprocessors. Yes, up until the end of the 1970's, and even into the early 80's, you could build a faster/more powerful computer out of bitslice components than you could out of microprocessors, which is why they continued being used in Minicomputers and early high-powered workstations like the Xerox Alto/Star. But to make them that fast you had to pair them with boards-worth of expensive bipolar memories and high-speed logic; yes, you can make a toy that adds two numbers together that fits on a big postcard:

("Cream soda")

But making a real general-purpose computer out of these is dozens and dozens of chips which, again, were very expensive at the time if you wanted "fast". Microprocessors were not about sheer speed (in the beginning), it was about making something that was "useful" while being optimized to use cheap supporting circuitry. These days you can make a TTL computer like the Gigatron on a board the size of a piece of paper which has pretty impressive capabilities, but the heart of it is a positively huge and impossibly fast for the 1970's microcode ROM, and its "RISC" architecture also makes it extremely RAM intensive compared to a more complex CISC design. Built out of 1970's components the Gigatron would be an order of magnitude or two more expensive and also about the same degree slower. The same amount of PCB space would go *much* further with an 8080 or whatever on it and also perform better. Up your space and budget to filling a box the size of a small file cabinet with parts then, sure, you can build something out of bitslice components that will outrun it.

Part of the reason I suggested the 74LS181 is that before the 74LSxxx chips were invented these 74xxx chips would also have consumed a lot of power as well.

 

[Eudimorphodon]

Part of the reason I suggested the 74LS181 is that before the 74LSxxx chips were invented these 74xxx chips would also have consumed a lot of power as well.

The dozens of "LS" chips and bipolar memory devices are still going to consume a shedload of power compared to a MOS microprocessor. And, again, the real point, take a few square feet of PCB space to build a computer with a similarly rich instruction set to most 8-bit microprocessors.

 

[Hugo Holden]

Part of the reason I suggested the 74LS181 is that before the 74LSxxx chips were invented these 74xxx chips would also have consumed a lot of power as well.

A typical board, one example off the top of my head, averaging the diffence across a large number of gate and flip flop & counter types, with about 65 74 series TTL IC's (Atari's arcade Pong) draws just over 1 Amp from its 5V supply. If you populate that same board with all 74LS TTL's instead, the current consumption drops to around 0.32A.

 

[Hugo Holden]

I'm not sure how an ALU on its own, can be compared with a CPU, because an ALU is already an integral part of the more complex structure of the CPU. Unless you could say, if you started with an ALU in isolation, added a lot of hardware around it, to create something resembling the functionality of a CPU. It appears to me that a CPU could replace an ALU, but not the other way around, without a lot of extra hardware help for the ALU.

 

[Eudimorphodon]

... Here's with a real example of what you're stuck with if you only have TTL:

Here are the schematics for a Wang 2200 computer that sold for $7,400 in 1973 money (That's about $50,000 today) for a minimal 4K system. (more about the Wang) This machine uses a single 74181 on its ALU board, and all together it looks like what we'd call the "CPU" (if we were comparing it to an 8080 or something) is spread across around six boards. It was clocked at 10mhz, but took around 16 cycles to complete most "microinstructions", which weren't really comparable to the "visible" instructions on a normal CPU; many microinstructions were necessary for even the simplest operation. In other words, this baby was pretty slow despite its relatively high input clock speed, and it lived inside a giant metal suitcase cabled to a power supply about the size of a microwave oven. (the terminal was separate) Switching from "straight" TTL to LS sped the machines up significantly and get rid of the microwave oven power supply; by 1976 denser memory devices finally enabled them to offer a low end "PCS" model that fit inside of a (very fat) desktop terminal case. (Although arguably the main line remained the giant suitcases, which started incorporating systems to support multiple terminals and act like baby Minicomputers.)

Anyway, this 1976 "portable" Wang still weighed a ton and cost $5400 for an 8K computer. (FWIW, this was considered "cheap" compared to the 16K configuration of the competing IBM 5100, which was priced at a cool $8,975.) At this point you could get a microcomputer of roughly similar capabilites for around a third of that price (including a monitor ), and by the end of 1977 the Wang would be competing with $599 TRS-80s. It's hard to compare the absolute performance of the Wang machines to early microprocessor-powered PCs because the Wang computers could literally only run BASIC; it was baked into their microcodes, but for most practical purposes they were "similar", maybe for some BASIC programs the Wang could be faster than a cheap micro running a software BASIC, but you can see how this isn't even a contest at this point in terms of "bang for the buck".

(Wang themselves baked their TTL CPUs into proprietary VLSI chips by the early 80's, essentially turning them into microcomputers, to get the margins up. And by the late 80's they started selling their legacy customers a system that had a 386 CPU in it running a software version of their BASIC "microcode", because at this point they were too niche to keep developing hardware themselves..)

Anyway, you can see how there is *no way* we would have had a "home computer revolution" in the 1970's with the cheapest machines costing as much as Cadillacs. (Or even houses.) Getting the scale of integration up was the only way that was ever going to bring the price down enough for that to happen.

A typical board, one example off the top of my head, averaging the diffence across a large number of gate and flip flop & counter types, with about 65 74 series TTL IC's (Atari's arcade Pong) draws just over 1 Amp from its 5V supply. If you populate that same board with all 74LS TTL's instead, the current consumption drops to around 0.32A.

By comparison, looking at a few datasheets the typical/max consumption for a mid-70's microprocessor would be in the 70-150 mA ballpark. Not bad considering it's *several* boards worth of LS TTL. And this is leaving out the consumption of the bipolar memories that you need to hold the microcode of a bit-slice computer. Some of that stuff ran *hot*.