Why don't we see 128bit CPUs?

[Seminole]

I know that compared to 32bit, 64bit CPUs are ridiculously better at simple tasks and processing larger numbers. We really don't have a practical need for 128bit. But what is stopping us from making CPUs that run soley on a 128bit architecture? Is it too costly?

[terrytek]

No point.

[Clanscorpia]

It would require new standards, new software, new chip designs. 64 bit is to widely accepted. And we have no use for it

[RedWulf]

You know how Bugatti just doubled the V8 for a 16 cylindar engine then got its ass kicked around by single V8s even though V8s are far better than 4 cylinders, just doubling the structure doesn't mean its better at some point you reach a balance and maybe one day when the arch becomes the limiting factor we'll double it. 

[Nowak]

No need for it right now.

[Seminole]

5 minutes ago, Clanscorpia said:

It would require new standards, new software, new chip designs. 64 bit is to widely accepted. And we have no use for it

But 128 bit would be backwards compatible just like every previous iteration... so there must be something stopping us on a micro-architectural level then, right?

[don_svetlio]

Just now, Seminole said:

But 128 bit would be backwards compatible just like every previous iteration... so there must be something stopping us on a micro-architectural level then, right?

The reason is that 64-bit CPUs can support up to 16 million GB of RAM. We are VERY far away from that. Literally 0 reasons to go 128-bit

[Clanscorpia]

4 minutes ago, Seminole said:

But 128 bit would be backwards compatible just like every previous iteration... so there must be something stopping us on a micro-architectural level then, right?

It wouldn't be completely backwards compatible I believe. Plus 64bit which almost everyone has wouldnt be compatible so there would be lots of complaints

[Seminole]

4 minutes ago, don_svetlio said:

The reason is that 64-bit CPUs can support up to 16 million GB of RAM. We are VERY far away from that. Literally 0 reasons to go 128-bit

I'm not asking why we shouldn't, I'm asking what is stopping us?

[don_svetlio]

Just now, Seminole said:

I'm not asking why we shouldn't, I'm asking what is stopping us?

Common sense

[siddharthk4498]

@Seminole 16.8 terabytes of ram not enough for you, m8?

[harrynowl]

I saw a great post on reddit about it a while ago- found here: https://www.reddit.com/r/askscience/comments/2ke0o5/we_have_32_and_64bit_cpus_why_not_a_128bit_cpu/clkf8sd

 

Bit more informative than the responses here which are basically more people who don't know themselves why only that they don't need it.

[zMeul]

because we barely saw 64bit CPUs

and I don't mean the x86-64 kind - those aren't 64 bit CPUs, they're 32bit CPUs with a instruction set extension

 

as for higher bit extension, Intel's AVX, AVX2, AVX-512

[Seminole]

1 minute ago, don_svetlio said:

Common sense

That doesn't answer my question at all. I just want to know why we haven't done it. My big question was "is it too costly?" And I'd like to reiterate by asking how costly would it be. You have 21,000+ posts so I know you are good at communicating so you can think with me  

 

3 minutes ago, siddharthk4498 said:

@Seminole 16.8 terabytes of ram not enough for you, m8?

Haha well I'm sure if computers are still popular next century, we will have to worry about the limitations of 64 bit processing. The average 8gbs is almost 1/100th of the way there

[don_svetlio]

Just now, Seminole said:

That doesn't answer my question at all. I just want to know why we haven't done it. My big question was "is it too costly?" And I'd like to reiterate by asking how costly would it be. You have 21,000+ posts so I know you are good at communicating so you can think with me  

 

Haha well I'm sure if computers are still popular next century, we will have to worry about the limitations of 64 bit processing. The average 8gbs is almost 1/100th of the way there

They would have to spend several million in R&D for it and in the end it would make 0 difference in the real world use case scenarios - it would essentially be intel or AMD shooting themselves in the foot for no apparent reason.

[ThinkWithPortals]

26 minutes ago, Seminole said:

I know that compared to 32bit, 64bit CPUs are ridiculously better at simple tasks and processing larger numbers. We really don't have a practical need for 128bit. But what is stopping us from making CPUs that run soley on a 128bit architecture? Is it too costly?

You answered your own question. No one's bothering because it's not necessary, and won't be for a WHILE.

[Tosa]

64 bit CPUs can do calculations up to 2^64 (18446744073709552000) and could support up to 16 exabytes of RAM, while 32 bit CPUs only can go as far as 2^32 (4294967296) and supports 4 gigabytes at most. Well, they can go higher, but not natively.

 

I'll humor you; Let's say we switch to 128 bits. It would go as high as 3.402823669209385e+38 (move the decimal separator 38 places to the right) and support up to 576460752303423500 exabytes of RAM. Unless we have to deal with currencies which has suffered inflation far beyond extreme, I don't think we'll benefit form it anytime soon.

 

Interestingly, my 64 bit CPU was able to do the maths when I tried to make calculations waaay beyond 2^64.

[siddharthk4498]

6 minutes ago, Seminole said:

That doesn't answer my question at all. I just want to know why we haven't done it. My big question was "is it too costly?" And I'd like to reiterate by asking how costly would it be. You have 21,000+ posts so I know you are good at communicating so you can think with me  

 

Haha well I'm sure if computers are still popular next century, we will have to worry about the limitations of 64 bit processing. The average 8gbs is almost 1/100th of the way there

just can't wait for the the samsung galaxy s70 with a 64 k screen,  quantum processor and 16 terabytes of ram and 64 terabyte of rom.

[vanished]

2 minutes ago, Tosa said:

64 bit CPUs can do calculations up to 2^64 (18446744073709552000) and could support up to 16 petabytes of RAM, while 32 bit CPUs only can go as far as 2^32 (4294967296) and supports 4 gigabytes at most. Well, they can go higher, but not natively.

 

I'll humor you; Let's say we switch to 128 bits. It would go as high as 3.402823669209385e+38 (move the decimal separator 38 places to the right) and support up to 576460752303423500 exabytes of RAM. Unless we have to deal with currencies which has suffered inflation far beyond extreme, I don't think we'll benefit form it anytime soon.

 

Interestingly, my 64 bit CPU was able to do the maths when I tried to make calculations waaay beyond 2^64.

It's exabytes actually

[Tosa]

1 minute ago, Ryan_Vickers said:

It's exabytes actually

Oops, my bad. I'll fix it.

[Gikero]

5 minutes ago, siddharthk4498 said:

just can't wait for the the samsung galaxy s70 with a 64 k screen,  quantum processor and 16 terabytes of ram and 64 terabyte of rom.

Ugh. 16TB of ROM. Seriously, I'm gonna have to buy at LEAST a 128TB NanoSD card for my music and photos! =P

[Seminole]

Lol well thanks guys, I kind of see exactly how pointless it is now

[Coaxialgamer]

No point really.  There are two main advantages of a cpu with more bits :it can process larger numbers (in this case,  a 64 bit cpu can process numbers up to 1.8E19 in one cycle.  The key word here is one cycle.  Back in the days of 8 bit computers,  if you wanted to use a number larger than 256, it would take several cycles to do so.  However if your cpu has the registers capable of handling you number (lets say 1000),  there is no advantage in adding more bits,  as it will still take one cycle to do that operation.  In this case,  a 32 bit cpu would be just as good as a 64 bit cpu to do that task ( for the same power,  in reality no powerful 32bit cpus are being made today). 

The second main advantage is a larger data bus.  This means a processor can address more ram,  because,  well,  it has the addresses for it. 

But,  despite popular belief,  a modern typical processor cannot handle 2^64 bits of ram.  Since most procesors actually use 48bit data bus.  Meaning they can adress 2^48 bits of ram. For example,  the intel 8086 was a 16 bit cpu, but it had a 20bit data bus.  Meaning it could adress 2^20 bits of ram,  or ~1MB.

This is the principal behind the PAE extension.  You actually are just increasing the data bus size.  For example,  when the pentium pro launched,  it was a 32 bit cpu.  However,  thanks to a 36bit data bus,  it could address 64gb of ram. 

 

Hoped this helped.  However i simpligied the explanation here. And since i am not an computer engineer,  if i made a mistake,  feel free to correct me. 

[Seminole]

23 minutes ago, Coaxialgamer said:

No point really.  There are two main advantages of a cpu with more bits :it can process larger numbers (in this case,  a 64 bit cpu can process numbers up to 1.8E19 in one cycle.  The key word here is one cycle.  Back in the days of 8 bit computers,  if you wanted to use a number larger than 256, it would take several cycles to do so.  However if your cpu has the registers capable of handling you number (lets say 1000),  there is no advantage in adding more bits,  as it will still take one cycle to do that operation.  In this case,  a 32 bit cpu would be just as good as a 64 bit cpu to do that task ( for the same power,  in reality no powerful 32bit cpus are being made today). 

The second main advantage is a larger data bus.  This means a processor can address more ram,  because,  well,  it has the addresses for it. 

But,  despite popular belief,  a modern typical processor cannot handle 2^64 bits of ram.  Since most procesors actually use 48bit data bus.  Meaning they can adress 2^48 bits of ram. For example,  the intel 8086 was a 16 bit cpu, but it had a 20bit data bus.  Meaning it could adress 2^20 bits of ram,  or ~1MB.

This is the principal behind the PAE extension.  You actually are just increasing the data bus size.  For example,  when the pentium pro launched,  it was a 32 bit cpu.  However,  thanks to a 36bit data bus,  it could address 64gb of ram. 

 

Hoped this helped.  However i simpligied the explanation here. And since i am not an computer engineer,  if i made a mistake,  feel free to correct me. 

I thought I knew everything I needed to know about data busses but wow that was interesting! You really explained it quite well. Thank you for that explanation.

[AlTech]

7 hours ago, Seminole said:

I know that compared to 32bit, 64bit CPUs are ridiculously better at simple tasks and processing larger numbers. We really don't have a practical need for 128bit. But what is stopping us from making CPUs that run soley on a 128bit architecture? Is it too costly?

We don't need it. And nothing supports 128 bit. Windows doesn't, mac doesn't, Linux doesn't. 100% of all programs do not support 128 bit. Let alone 64 bit.