>128 bit cpu will it ever happen?

>>58116678
>3d
kill youself and go back to kikebook normie nigger reeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee

you have 60 days to explain how its benefits would outweigh the costs

get cracking

give me one reason it needs to.

>>58116693

vr gaming

aliens who program their spaceships and AI components would probably need a 128bit cpu

>muh arrival

>>58116710
>lets blow tons a money on something pointless

>>58116705
>cpu

>>58116750
>or lets spend tons of money on something that will increase performance
:^)

>>58116765
>increase performance
lol

>>58116693
128 bit atomics

>>58116777
Thank for the (You) kouhai i was just shitposting

>>58116678
We've had them for a while, along with 256 and 512 bit see MMX (and later) instructions

>>58116796
well I get some (You)'s in return so thanks ;)

>>58116678
Theres no benefit in speed. Only precision of floats.
64bit is well within range of consumer use. Thats why we didnt move on. Theres literally no reason to.

If 64bit is so perfect then why didn't we just use it from the start? Hurf durf fucking idiots

File: morfius.jpg (76KB, 1279x548px) Image search: [Google]

76KB, 1279x548px

>>58116693
128 bit CPU would be twice as fast as 64 bit, idiot

>>58116765
https://www.youtube.com/watch?v=mlRWqqWay7c

>>58116825
>Theres literally no reason to.
this is a sad thought. why to get up in the morning, then. Theres literally no reason to.

File: an angel.jpg (12KB, 291x300px) Image search: [Google]

12KB, 291x300px

>128 bit cpu
>will it ever happen?

We could make more sometime, if we need them.

>>58116678

Probably not in a general purpose CPU. There's no point.

64-bit doesn't magically make your programs way faster. The reason why 64-bit mattered in the x86 world was for a couple of reasons:

1. 64-bit chips operate on 64-bit integers natively, and the size of the native integer type is important when you're dealing with memory addresses. 32-bit integers can only address 4 gigabytes of RAM, while 64-bit can in theory address 16 exibytes of RAM. In practice, limitations of current x64 CPU's limit that amount to around 256 terabytes.
2. You can do arithmetic on on 64-bit integer types quicker. If you're dealing with lots of 64-bit integers in your programs, that's a plus, but most software written for x86 didn't use 64-bit integers if they didn't have to, so it took a while for programs to catch up, plus they had to be recompiled anyway.
3. You can pass 64-bit floating point numbers back and forth between memory slightly quicker. The x86's FPU operated with 80 bits of precision internally, so the calculations themselves were no faster.
4. 64-bit gave programs access to WAY more general purpose registers at the same time, as part of the upgrade. Compilers can take advantage of those extra registers to make programs faster.
5. 64-bit means that you can make assumptions about the existence of certain CPU instructions that you couldn't with x86. Compilers can take advantage of these extra instructions to make programs faster.

Of these five improvements, only the first was the one that actually mattered to upgrading users. Only #2 and #3 could be considered direct performance improvements from the "bitness" of the CPU, and the gains were relatively small. #4 and #5 could have been done without any change to the "bitness" of the CPU.

So...128-bit CPU's. We don't need anywhere NEAR that much RAM yet, and 128-bit integers and floating point numbers are incredibly rare in practice. So probably not, and even if it did it wouldn't really matter.

>>58116678
>will it ever happen?

Probably not in our lifetimes, nor in the lifetimes of our grandchildren.

The 64-bit address space is 16 EiB, which is a billion times more RAM than a 16 GiB computer holds.

Assuming that computer RAM size doubles every 5 years, then it would take about 150 years to reach 16 EiB. (30 doublings is one billion.)

There's a small possibility that you might have a grandchild who's still alive 150 years from now, but it's not likely.

You've also got to wonder how long we'll need to keep doubling our RAM capacity every 5 years. Right now we're in a huge expansion phase, but it's not clear that it needs to keep going at the current pace. For example, CPU clock speeds hit a brick wall about 10 years ago, and now they're set to double only once every 20 years, at best. And yet, that hasn't hurt computing much. Once we get to 10K monitors, we'll be pretty much done with our need to keep increasing the resolution, because of the limits of our retinas. The same thing might happen with RAM size -- we might get to a similar "plateau" phase where most people finally have far more RAM than they could ever use, and the size doublings are only needed every 10 or 20 years -- which could easily extend the life of 64-bit CPUs for yet another 100 years beyond the original 150 year projection.

>>58117237
What about bandwidth tho? Sure we're not using 128bit that often, but it would mean you can run more 64bit stuff parallel right? More bus space to fill with datas!

>>58116765
Are there really people who think that increasing the bit width somehow increases performance by any considerable amount?
Fucking off yourself you retard.

I don't think they will be made or if they are, there will be limited runs of them. Quantum computing is more realistic than a 128-bit CPU.

The only way that happens is is address start to become proto headers, with metadata. Sort of like Rec 2020 12 bit color and IPv6.

>>58117320
No you fucking retard

>>58116678
If your CPU is a dual core CPU then you're technically using 128-bit.
(64x2, each core has 64-bits, 128-bits multi).
My CPU has a technical 512-bit.

>>58117320

Breaking a 128-bit word back down into 64-bit words to work on them either costs you processor cycles or specialized silicon that will increase cost, heat and power consumption.

File: 1481693571516.jpg (247KB, 1224x1445px) Image search: [Google]

247KB, 1224x1445px

>>58117421
>My CPU is technically 512-bit

>>58116678
even 128bit cpu isn't enough

there are an estimated 10^80 atoms in the universe

a 128 bit cpu can count to about 3.4^38
you would need a 512 bit cpu to be useful.
even with a 512bit cpu, an atom can only be represented by a single byte

>>58117320
Modern x86 CPUs have this; it's called SIMD.

>>58117459
It is possible for CPU to be useful even if it does not attempt to emulate the universe.

File: 1468145943600.jpg (136KB, 1125x480px) Image search: [Google]

136KB, 1125x480px

Now that we have a huge 64bit address space, why aren't we mapping entire disk drives into the address space? It should be possible with SSDs, not sure about HDDs.
Why are we still using interfaces like ATA?

>>58117405
No lol.

>>58117320

You have basically thought up of SSE2 a decade and a half after the instruction set was first released on the Pentium 4. The registers are 128-bits wide and can operate on 2 64-bit types at once.

Is there a matching 128-bit bus to go with it? I'm not sure, but the #1 issue with an idea like yours would be space - these chips are fucking tiny and space is at a premium, so any decisions about what goes on the chip and where has to make sense from a size AND performance point of view.

>>58117588
It's kind of like saying we should abandon DNS with IPv6.

>>58117459
Why do we need as many addresses as there are atoms in the universe?

>>58117588
>why aren't we mapping entire disk drives into the address space
Because we are already used to the way it works now.
Because disks are block devices and mapping them to memory is not going to come for free.
Because almost no one needs direct access to disk; everything is done through filesystem.
Because mmap exist and allows you to do that when you feel the need for it while still keeping the benefits the centralized filesystem provides.

>>58117609
DNS can be useful for security, but I see no reason not to map storage devices directly into the address space.

>>58117657
Can you think of a single reason to do that though?

>>58117657
Also how the fuck is DNS useful for security?

>>58117609
Wait I got mixed up with NAT, did you mean NAT? otherwise your post doesn't make sense.

>>58117664
Simpler interface (just memory access), simpler driver.
Better performance?

>>58117596
Who knows m8. With the whole moore law coming to an end. Deelopers might move onto fancier controller/buses. Which might mean fancy headers. Might as well use what you have until you are back to practically 32.

>>58117694
No. it won't happen.

>>58117682
>otherwise your post doesn't make sense.
Your post makes just as little sense as his (and unlike you he made it nonsensical on purpose).

>Simpler interface (just memory access), simpler driver.
So we just rewrite all drivers we have and we're good to go, right? Easy!

>Better performance?
I can't tell if you're baiting. We just change the interface and somehow we get better performance doing the actual thing. Fascinating.

>>58117723
Whatever, it was just an idea. I just wanted to see what experts on the subject thought about it.

>>58117682
I know it doesn't make sense. How do you think programs address storage today?

Pro tip: they interact with a file system.

IPv6 practically does NAT natively anyway, since there is an innate private/public dichotomy.

>>58117719
One day someone will get the bright idea to do session management within IPv6, and then IPv6 will beomce the new TCP/UDP port. Then we will have to come up with IPv8. It can happen to anyone.

Trump literally got elected president of the US.

>>58116890
Theres plenty of benefit to getting up in the morning. None for 128bit.

>>58117772
>Pro tip: they interact with a file system.
Your fucking point? I never said to get rid of filesystems. I just suggested replacing ATA/whatever with direct memory access since the address space is large enough to fit entire storage devices now.
The filesystems would still be there.

>>58117819
Direct memory access is not fairy dust. You still need to connect to the other device somehow. ATA and direct memory access are not mutually exclusive. You need to stop posting for a moment and think really hard about what you are proposing.

maybe once we get actual 64 bit processors OP
the address bus only allows for 52 bit physical addresses and 48 bit virtual addresses

>>58117860
That seems kind of stupid, should be the other way around.

>>58117876
why? a single process can address up to 48 bits while the entire OS can address up to 52 bits

>>58117887
Seems kind of pointless. I feel 52 bits virtual addresses would be more useful than 52 bits physical addresses.

>>58117901
the virtual memory needs to be addressable in the physical memory, you can't have an address that is outside what's actually possible to be physically backed, it'd be inaccessible by the processor

File: _81738062_risitas.jpg (34KB, 1024x576px) Image search: [Google]

34KB, 1024x576px

>>58116678
just install windows 7 x64 twice to get 128bit, are you fucking new here?

>>58117955
Then make both either 52 bits or 48 bits. I don't get why physical addresses need to be 52 bits while virtual addresses are 48 bits.

Only reason I can think of for 128-bit is faster code that uses arbitrary size integers, and Intel's introducing a new extension to accelerate these use-cases anyway.

>>58117977
It's so a single process can allocate up to 48 bits worth of RAM (256 tb)
With a 52 bit physical backing, you can have 4 PiB of ram total, meaning you can have 16 processes running 256 tb of ram each

>>58118054
Sounds stupid. Just give each process full access to RAM. I see zero downsides to that.

>>58118077
development time & cost, it's cheaper to do this and will be good enough for the next few years

>>58118054

Does the distinction actually matter for any modern computer? I'm pretty sure nothing out for any x86_64 platform even approaches that amount of RAM, and that once we do the buses will just be expanded to accommodate and that will be that.

>>58116678
It already has, there's just no benefit to consumers.

>>58116678
The Dreamcast and PlayStation 2 both had 128-bit CPUs.

If we ever need more than 16 exabytes of RAM, then yeah. 128-bit RISC-V is already a thing. Nobody's put it to silicon though because there's literally no need.

>>58118114

No it hasn't. There is no such thing as a 128-bit general purpose CPU.

There are CPU's with 128-bit registers, but there is no architecture that uses native 128-bit integers because there's no reason to.

>>58118102
servers are getting pretty insane amounts of ram nowadays, especially memcached boxes. it's currently a lot, but in 5-10 years who knows, for the foreseeable future it's enough

>>58118117

The primary CPU in the Dreamcast was a Hitachi SH-4, which was actually just 32-bit. The primary CPU in the PS2 was a MIPS R5900, a 64-bit chip.

File: 1466015849270.jpg (29KB, 640x480px) Image search: [Google]

29KB, 640x480px

>mfw the PSP had a 64-bit CPU in 2004 while the first phone to have one didn't come out until 2013

File: Capture+_2016-12-21-18-39-57.png (80KB, 1266x400px) Image search: [Google]

80KB, 1266x400px

/r/askscience/comments/2ke0o5/we_have_32_and_64bit_cpus_why_not_a_128bit_cpu/

>>58116678

I have one OP. My dad bought it for me for xmas

Its awesome.

>>58116693
Porn is always the answer... porn...

>>58118543
Porn doesn't benefit from 128bit CPUs.

>>58116678
No, because you don't need more than 2 TiB of RAM.

>>58118594
Well technically it uses compressed image formats which run fast due to 128b+ SIMD decompression instructions (which i know it's not the same but...)

A tl:dr version of why we dont have 128 bit cpus and why nobody is working on one is simple
>we dont need it

The big thing is that it can adress a bunch more shit. But its tens of thousands of times more than what we use now.
Think 20 years back with things like early pentium processors. They could support up to 3.4gb of ram but most systems had 16-48mb of ram

64 bit can do 16 exb and the average user only needs 4-8gb

Why move to 128 bit now?
Even when 64 bit cpu's were new people were still mainly using under 2gb of ram
There are still new 32 bit cpus because some people dont even need that, 3.4gb is enough

Nothing anyone does now or will do for probably quite a long time will ever need 128 bit processor

>>58116678
Not for hundreds of years, no.

>>58118707
>not for hundreds of years
meanwhile a few decades ago 128K ram was considered a huge amount

>>58116678
IBM already holds the patent

>>58118249
Really?

"Announced in October 2011,[7] ARMv8-A (often called ARMv8 while the ARMv8-R is also available) represents a fundamental change to the ARM architecture. It adds an optional 64-bit architecture"

>>58116678
>what is bit slicing
already happened

>>58118625
Except if you watch porn you're probably also on windows which means 95% of your software is 32bit which means no SIMD instructions are being used because the program cannot assume the presence of them.

>>58118657
>3.4gb is enough
You mean 4GB.

>>58116678
Never heard of Itanium, right?
You must be underage

File: no gods or kings.png (1MB, 1000x699px) Image search: [Google]

1MB, 1000x699px

>too smart to wait for the singularity

File: surrender.png (1MB, 810x1318px) Image search: [Google]

1MB, 810x1318px

>>58116693
go to bed Terry

>>58118810
64-bit ARM may have existed since 2011 but the iPhone 5s was the first phone with a 64-bit CPU and it came out in 2013.

>>58119300
There's no need for a phone CPU to be 64 bit just yet, and the iphone doesn't even have more than 4 GBs of RAM, I think only a few android phones have 6GBs of RAM

>>58119323
I'm not saying it's necessary, I'm just pointing out that the
PSP beat phones to 64-bit by almost a decade. Also the original PSP had 32MiB RAM and later models had 64MiB.

128 bit CPUs have existed for more than 45 years.

https://en.wikipedia.org/wiki/IBM_System/370

>>58119323

>le 64 bit if for more ram meme
windows sure cucked everyone good

>>58119323

The advantage of 64 bit processors isn't just for addressing more memory.

>>58116678
We would need to start building bigger chips and that would decrease performance. We cant pack transistors any tighter on silicon

>>58118077
if you want identity mapped memory, install templeos :^)

>>58119510
>If we have address changers it's same amount of ram.

There are plenty of disadvantages with this

File: 1341408996286.jpg (18KB, 409x393px) Image search: [Google]

18KB, 409x393px

>>58117421
>If your CPU is a dual core CPU then you're technically using 128-bit.
>(64x2, each core has 64-bits, 128-bits multi).
>My CPU has a technical 512-bit.
Quite of the year right there

>>58119209
>assume
cpuid

>>58116678
what applications are you currently working with that utilize integer values exceeding 64 bits enough that the additional load/store instructions required stack up noticeably?

>>58116705
none of that shit is going to be handled by the CPU when GPUs do it far better
even in the case of graphical calculations that require a shit ton of accuracy to the point of overflowing a 64-bit register, floating point units, SIMD units, whatever else have long since incorporated registers as large as 512 bits to handle it

>>58116765
there is none, you're just adding a bunch of extra empty space that will rarely be utilized to the point of a meaningful benefit

>>58118249
>PS2 had a 128bit CPU
>Could actually keep up with the PS3 and could run GT4 at 1080i whereas GT5 ran at 1280x1080p and didnt have fog or decent realtime reflections and alot of other effects due to the CPU being a pile of garage
is the Cell the worst CPU ever made?

>>58116868
>this is what retards look like

>>58119642

>PS2 had a 128bit CPU
No it didn't. It had a 64-bit address space and general purpose registers. It had 128-bit SIMD, but pretty much everything has that these days, or better.

>>58119323

x86 and ARM both had fundamental architecture changes in the switch to 64-bit. As a result, even if one has only 1 GB of RAM, it may often still be worth it to use 64-bit applications on these architectures.

>>58118249
PSP's CPU was a modification of a 64-bit CPU, and one of things they modified was they gutted it to be a 32-bit processor, if my memory serves me right.

>>58121455
Here we go, found it:
>The R4000 is a microprocessor developed by MIPS Computer Systems that implemented the MIPS III instruction set architecture (ISA). Officially announced on 1 October 1991, it was one of the first 64-bit microprocessors and the first MIPS III implementation.
>The PSP's main microprocessor is a multifunction device named "Allegrex" that includes a 32-bit MIPS32 R4k-based CPU (Little Endian),

>>58119619
>there is none, you're just adding a bunch of extra empty space that will rarely be utilized to the point of a meaningful benefit

No one will need more than 637 kB of memory for a personal computer.

>>58116678
64 bit computers can address up to 16384 PB of RAM. Maybe we will exceed that limit some day, but right now, it's looking extremely low-priority.

>>58116678
That filename tho, had a good kek

>>58116825
>Only precision of floats.
Float precision is not tied to your machine's address size. We already have 128-bit, 256-bit and 512-bit wide integer and floating point operations. It's called SIMD (SSE/AVX etc.)

The problem is that nothing needs operations this wide. Modern CPUs have already hit the point where the parallel ALUs don't get exhausted anymore.

>>58117256
>There's a small possibility that you might have a grandchild who's still alive 150 years from now, but it's not likely.
this is /g/ so the possibility of having children is 0

>>58116868
Nope but it would be able to handle a greater amount of data. Is useless as there are already GPUs that can do this.

>>58117256
>You've also got to wonder how long we'll need to keep doubling our RAM capacity every 5 years.
Don't worry, “programming” hipsters and code artisans will find more ways to bloat up websites

>>58116678
>he fell for the 16 exabyte meme

>>58118782
You understand how exponential growth works? Adding one bit doubles the amount of memory you can address. Doubling the number of bits squares the amount of memory you can address.

The leap from 16 bit to 32 bit increased the addressable bytes from 65536 to 4294967296

The leap from 32 bit to 64 bit increased it from 4294967296 to 18446744073709551616.

The next leap will increase it to 340282366920938463463374607431768211456.

The numbers very quickly start getting ridiculous.

File: 1460751587984.png (125KB, 453x433px) Image search: [Google]

125KB, 453x433px

>>58117421