How fast could you make a CPU run if you made it entirely out of these and similarly sized components?

How fast could you make a CPU run if you made it entirely out of these and similarly sized components?

Other urls found in this thread:

megaprocessor.com/
youtube.com/watch?v=n3wPBcmSb2U
homebrewcpu.com/
github.com/szoftveres/ttlcpu
hackaday.io/project/11012-risc-relay-cpu
leo1cpu.puntett.net/main.htm
megaprocessor.com
hackaday.io/project/665-4-bit-computer-built-from-discrete-transistors
joewing.net/projects/q1/index.shtml
youtube.com/watch?v=5uCkrEzVjDw
youtube.com/watch?v=HDMkw6lnzmI
twitter.com/NSFWRedditImage

Not very fast

Intel's Core i7 has 731,000,000 of those things, so you do the math.

God awfully slow. You'd be lucky to get TI-89 performance out of a room sized computer.

For reference to this statement. Honestly, I'm not even sure that those types of transistors are even feasible to build anymore.

Pretty sure they even stopped producing the 80486 which used to be the go to chip for electronics and replaced it with the Atom.

GaAs cpus when?

You'd be better off buying a calculator at the dollar store.

megaprocessor.com/

>Using transistors
youtube.com/watch?v=n3wPBcmSb2U

Bigger circuit = need lower frequencies to use electronics laws

36MHz

TTL uses BJTs granted they are integrated into various packages on a die but they still use bipolar transistors so you can achieve speeds on par with the maximum speeds of the 74S and 74F families of logic chips. If you are using discrete BJTs you will not matched the speeds of integrated designs even using fast transistors. You probably won't even match the speeds of the slowest TTL chips or even RTL or DTL.

You might be able to achieve stupid speeds with heterojunction bipolar transistors (HBTs), metal-semiconductor field effect transistors (MESFETs), or high electron mobility transistors (HEMTs) however to my knowledge these are only really used in RF applications. I'm not aware of them being used in any digital electronics. HBTs have a pretty insane unity gain bandwidth of around 700GHz. Imagine using them to create clock speeds in excess of 100GHz along with ultra-high speed digital logic. Of course there are a whole host of issues operating at such a high frequency. Loads of physics and transmission line theory comes into play. That's probably why it's only used in RF, it's probably not practical for digital logic but we can dream right?

CPUs aren't bottlenecked by transistors' speeds though.

...

Modern microprocessors use FETs, not BJTs numbnuts. Also, OP asked about *discrete* transistors, not transistors on an integrated circuit.

When they can no longer squeeze more performance gains out of silicon.

Probably about 15-25 mhz.
It's hard to go much faster because the distance between the transistors becomes a huge issue at those frequencies.
Like, at 25 mhz, electrical signals can only travel about 100cm before the next cycle.
This greatly limits the size, and therefore complexity, of a macroscale processor like you're describing.

Bunch of undergrad level replies here.

Say it with me now: 7 inches per nanosecond. Make a real basic ISA and look at the longest path (probably in the tens), multiply by propogation delay of the component plus leads (we’ll say a nanosecond), take the inverse and add some buffer.

50 MHz at the absolute most.

I can't make a CPU. Neither can you, or anyone else in this thread.

Bullshit. Take three courses on computer architecture and making your own ISA will seem reasonable.

You are wrong.

Making a CPU is doable, especially if you plan on using a FPGA rather than actually fab a chip.

>How fast could you make a CPU run if you made it entirely out of these and similarly sized components?
very slow, because even small pieces of conductor gonna behave like parasitic coils and antenas so your cpu gonna end working in the Mhz order.

homebrewcpu.com/
github.com/szoftveres/ttlcpu
hackaday.io/project/11012-risc-relay-cpu
leo1cpu.puntett.net/main.htm
megaprocessor.com
hackaday.io/project/665-4-bit-computer-built-from-discrete-transistors

Wow, look at all of these non-existant homebrew CPUs!

>Bunch of undergrad level replies here.
>Measures speed in MHz

Cool now we have a junior high level reply.

Those are terrible and useless, and no one in this thread designed or built any of them.

Idiot.

Maybe a few kHz

>Useless
That's a compliment here. Here the common opinion is bing-wahoo is more useful than decentralized currency and a better allocation of compute resources.

Yes frequency is how you measure the speed of a synchronous system.

Depends on what you mean by CPU.
I can make an 8-bit adder. That's a CPU.
I can't make an x86 compatible CPU with discrete logic though.

It would be fun as hell to try and make an IBM PC 80286 compatible machine with off-the-shelf discrete logic though.

As fast as a kenyan

Only if you're measuring two otherwise identical systems. How the hell are you going to say your 50MHz discrete bjt cpu is faster than my arm32 going 40? Go do your homework.

>Being this retarded
OP is clearly asking how high of a clock a discrete transistor CPU could run at.

OP isn't clearly talking about any metric. MHz is a shit choice for speed. Go do your homework.

Try (M)IPS. Still, you have to determine whether the computer is CISC or RISC.

My former coworker (also the guy who made JWM) made this thing:
> joewing.net/projects/q1/index.shtml

I don't remember it clocking very high (on the order of 100kHz maybe?), but it could do stored program execution including prime solving by Sieve or Eratosthenes:
> youtube.com/watch?v=5uCkrEzVjDw

The only thing he cut cheated on was using a commodity SRAM memory array rather than soldering thousands of flip-flops by hand.

Yeah you’re a fuckin idiot my dude. instructions per second? That’s a shit metric too. Oh your data width is one bit? Shut the fuck up. You’re just trying to sound smart when you have nothing to offer. Get back to coding and let the real engineers talk.

...

MHz is the only objective measure of CPU speed since the design and intended application of the CPU can make any other comparison meaningless. Of course, the comparison of clock speeds won't tell you much in many situations, but it is a valid comparison to make.

But in reality, when you're talking about how fast a processor would work based on the transistor technology used, it's obviously a conversation about switching speed. The same architecture built on two different transistor techs will perform in vastly different ways. One of the reasons for the difference, and a very important one at that, is switching frequency.

Aka bogomips

>6502 cpu
>3500 transistors
>60khz
youtube.com/watch?v=HDMkw6lnzmI

You are wrong.
t. 3rd year comp eng.

Jesus Christ.

(Although I kinda want to make one.)

>Can I buy one?
>No. (Not yet, at least.) This is very much a hobby project at the moment.
Lame.

Isn't the quark the new replacement? Which is a 80586?