Hypothetical question... Giant CPU

Associate
Joined
3 Feb 2012
Posts
2,267
Location
Bath
Just had a strange thought. It happens occasionally.

If it were possible (and I assume it would be, technically) to build a working CPU out of discrete components- i.e off-the-shelf transistors, bits of wire, resistors etc instead of etching onto silicon at microscopic level, how big would the result be?

Say I wanted to build a 'full-size', working Ryzen 3700X out of stuff from the RS catalogue, what sort of aircraft hangar would I need to house it?

And more importantly, are there enough years left to build it, before the sun swells up and engulfs the inner planets?

Told you it was a strange thought...
So GD can have it.
 
Soldato
Joined
1 Apr 2014
Posts
18,532
Location
Aberdeen
Something like that has been done, though not at that level. I remember seeing a Youtube video. The thing took the whole room, though that was for educational reasons as it was full of blinking lights to show every stage. I think it did 1 ips or was manual.
 
Soldato
Joined
17 Mar 2009
Posts
6,589
Location
Nottingham
Something like that has been done, though not at that level. I remember seeing a Youtube video. The thing took the whole room, though that was for educational reasons as it was full of blinking lights to show every stage. I think it did 1 ips or was manual.

I assume you mean this?

 
Associate
Joined
8 Oct 2008
Posts
260
This site, https://monster6502.com/, describes building a 6502 out of discrete components. Scroll down and they give an example of building an Apple A8X CPU found in the iPad 2 would be 286m square. Wouldn't mind buying the 6502 version.
 
Associate
OP
Joined
3 Feb 2012
Posts
2,267
Location
Bath
That's pretty cool!

So 118 acres...
477529 square metres. Lets say we have a 40m tall hangar. 477529/40 gives 11,938, divide that by two for a square-sided build... So it'd fill a room nearly 6km square and 40m tall.

Very high chance my maths is crap.
 
Man of Honour
Joined
5 Dec 2003
Posts
20,997
Location
Just to the left of my PC
That's pretty cool!

So 118 acres...
477529 square metres. Lets say we have a 40m tall hangar. 477529/40 gives 11,938, divide that by two for a square-sided build... So it'd fill a room nearly 6km square and 40m tall.

Very high chance my maths is crap.

Looks a bit wrong to me. If you had a "room" 6km by 6km the floor area would be 36 square Km, not 12. You'd be looking at a "room" about 4km by 3Km.

Except that you wouldn't. You'd need far more space than that for several reasons:

1) That volume was an estimate for transistors only. You'd need a lot more space for the wiring to connect the transistors together.
2) Heat generation would also be scaled up. You would need a lot more space to provide sufficient cooling.
3) With hundreds of billions of macroscopic components there would be frequent failures requiring replacement. So you'd need a lot more space to enable people to move around inside the giant CPU in order to replace parts. That would help a lot with the cooling issue, so at least you can kill two birds with one stone there. But you'd still need massive cooling systems. A couple of dozen 120mm fans won't do the trick here :)
4) You'd need a lot of people to service the giant CPU, so you'd need space to allow for staff transport to and from the site and staff transport within the site. Since it's so big, you'd need internal transport. Small EVs, probably, something like golf carts, so technicians can get to faults in a reasonable amount of time with spares required.
5) You'd need storage space for spare parts too, probably several such storage spaces given the size of the place. Otherwise downtime would be pretty much all the time.

So at a really rough estimate, I'd say if you were restricting your building to 40m tall you'd probably need it to be at least 5Km by 5Km. Probably more. Also, it's probably impossible in practice to get such a thing to work. You can do it for really simple processors, but not ones with tens of billions of transistors.

Early electrical computers were as you describe, but with tens of thousands of components rather than tens of billions. They filled buildings and required constant staffing by technicians. It's even theoretically possible to construct a computer, a bona fide programmable general purpose computer, that's entirely mechanical. Sir Charles Babbage designed one in the 1830s and called it an Analytical Engine. It was never built because the cost would have been enormous, nobody with control of enough money was absolutely certain it could be built with mid 19th century technology and Babbage kept changing the design to improve it. It's technically possible, though. One of his earlier mechanical processors (Difference Engine number 2) was built in the 1990s and works perfectly well. A few years back there was a project started to build Babbage's Analytical Engine. Presumably it hasn't happened yet. IIRC, they were finding it heavy going trawling through Babbage's surviving notes and sorting out the many revisions he made to his designs in order to get a single complete, compatible and functional design to work from. It would be a bummer to be building version 5 of one part and version 8 of another part and then find out that v8 of the second part was only compatible with v6 of the first part.
 
Commissario
Joined
17 Oct 2002
Posts
32,996
Location
Panting like a fiend
Looks a bit wrong to me. If you had a "room" 6km by 6km the floor area would be 36 square Km, not 12. You'd be looking at a "room" about 4km by 3Km.

Except that you wouldn't. You'd need far more space than that for several reasons:

1) That volume was an estimate for transistors only. You'd need a lot more space for the wiring to connect the transistors together.
2) Heat generation would also be scaled up. You would need a lot more space to provide sufficient cooling.
3) With hundreds of billions of macroscopic components there would be frequent failures requiring replacement. So you'd need a lot more space to enable people to move around inside the giant CPU in order to replace parts. That would help a lot with the cooling issue, so at least you can kill two birds with one stone there. But you'd still need massive cooling systems. A couple of dozen 120mm fans won't do the trick here :)
4) You'd need a lot of people to service the giant CPU, so you'd need space to allow for staff transport to and from the site and staff transport within the site. Since it's so big, you'd need internal transport. Small EVs, probably, something like golf carts, so technicians can get to faults in a reasonable amount of time with spares required.
5) You'd need storage space for spare parts too, probably several such storage spaces given the size of the place. Otherwise downtime would be pretty much all the time.

So at a really rough estimate, I'd say if you were restricting your building to 40m tall you'd probably need it to be at least 5Km by 5Km. Probably more. Also, it's probably impossible in practice to get such a thing to work. You can do it for really simple processors, but not ones with tens of billions of transistors.

Early electrical computers were as you describe, but with tens of thousands of components rather than tens of billions. They filled buildings and required constant staffing by technicians. It's even theoretically possible to construct a computer, a bona fide programmable general purpose computer, that's entirely mechanical. Sir Charles Babbage designed one in the 1830s and called it an Analytical Engine. It was never built because the cost would have been enormous, nobody with control of enough money was absolutely certain it could be built with mid 19th century technology and Babbage kept changing the design to improve it. It's technically possible, though. One of his earlier mechanical processors (Difference Engine number 2) was built in the 1990s and works perfectly well. A few years back there was a project started to build Babbage's Analytical Engine. Presumably it hasn't happened yet. IIRC, they were finding it heavy going trawling through Babbage's surviving notes and sorting out the many revisions he made to his designs in order to get a single complete, compatible and functional design to work from. It would be a bummer to be building version 5 of one part and version 8 of another part and then find out that v8 of the second part was only compatible with v6 of the first part.
Point 4 would probably be easiest solved with the sort of cargo lift units they use in some factories/warehouses, where there is a lift on rails that runs down an isle so you can reach anywhere on two vertical faces from it, except you'd be moving one or two people in rather than a pallet of materials (if you set it up right it could automatically take the technician to exactly the right location at the right height so they never had to bend/stretch but just work at a comfortable height). It could also lift to/from a bay where spares could be delivered as needed*.
It would also potentially help with the cooling as having the structure in vertical rows would mean the access area the rails use would act like chimneys for the hot air.

Something I don't think has been mentioned is that for anything with critical timing, the size of the thing would mean you'd run into issues with how fast signals can move that distance (doesn't the PCIE bux has quite tight limits on track lengths, and memory busses even more so?).
I remember reading a book many years ago about one of the early transistor computers where they had an issue with intermittent crashing until they changed the routing of some wiring so it took a shorter route (which required completely changing the layout of a couple of boards to create room for cabling to go straight up rather than along the board to an edge).
You also run into issues with interference on the lines due to the length of tracks and wiring, which means you need to start using shielded wiring/shielding tracks etc.


*Potentially a better system would be to have space for humans, but design such changes of transistors etc to be done by a robotic system by default, or where possible pull entire sub assemblies if that's faster.
 
Soldato
Joined
23 Jul 2009
Posts
14,083
Location
Bath
You're also missing the part where the distance between components is important for timing reasons. So unless it was exactly to scale, then it might throw the timing off
 
Man of Honour
Joined
5 Dec 2003
Posts
20,997
Location
Just to the left of my PC
A tangentially relevant anecdote that is probably amusing:

The word "bug" to describe an unexpected fault has been in use since well before the first computers, probably dating back to the early days of the telegraph system in the 19th century and in common usage by the late 19th century. In one incident in an early computer (building-sized with macroscopic components, hence the relevance to this thread) in the 40s or 50s, technicians traced the cause of an incorrect output from a program to a signalling fault between two vacuum tubes. The fault was caused by the body of a moth that had flown in and been electrocuted in just the wrong place - the first known case of a bug being caused by a bug, the first known literal bug.

And no, it wasn't Grace Hopper despite what you'll see claimed on some sites. She wasn't there at the time and said so herself, publically and repeatedly. She later mentioned it as an amusing story in an after dinner speech and it became attached to her because she was famous and the technicians weren't.
 
Back
Top Bottom