Will we ever have 128-bit processors?
- Thread starter Fixadent
- Start date
Solution
Ecky
Illustrious
- Feb 3, 2013
- 7,592
- 2
- 38,960
Nice link RCFProd.
One pertinent point:
https://en.wikipedia.org/wiki/Advanced_Vector_Extensions
It's important to decide what exactly you want to be more than 64bit. CPUs are complex, and it's not all or nothing.
One pertinent point:
https://en.wikipedia.org/wiki/Advanced_Vector_Extensions
It's important to decide what exactly you want to be more than 64bit. CPUs are complex, and it's not all or nothing.
gamerk316
Glorious
- Jul 8, 2008
- 13,234
- 130
- 44,840
The number of "bits" a CPU supports has always referred to the Address Space the CPU is capable of addressing [EG: How much memory it can use at any one time]. From that perspective, there's really no need to go beyond 64-bit, as we aren't ever going to saturate the full 64-bit address space [16 Exabytes]
128bit will allow to support a larger memory subsystem. This requirement is not expected soon, but some guys as those behind the RISC-V ISA already include a 128bit version for future needs.
https://en.wikipedia.org/wiki/RISC-V
As stated in the wiki link, the 128bit support was added to the RISC-V specification for usage in a future supercomputer: "At historic rates of growth, it is possible that greater than 64 bits of address space might be required before 2030."
And then it would go down to general user level latter. Remember that a smartphone today has about 100x more memory capacity than older supercomputers like the Cray-1.
TechyInAZ
Titan
If anything, we should be moving off of x86 all together and run ARM as it is a much more efficient architecture. Fortunately Microsoft is making an ARM variant of Windows 10 (that is NOT like RT, you can run any programs you like (for x86 that'll be via emulator)).
gamerk316
Glorious
- Jul 8, 2008
- 13,234
- 130
- 44,840
ARM is more efficient due to being 10x slower. And yes, I've benchmarked. Make an ARM CPU with x86 performance, and you'll end up with a worse version of x86.
x86 is a horrid CPU architecture, but ARM isn't replacing it at the top end.
TechyInAZ
Titan
Not even close. In reality ARM is already better than x86 because provides nearly the same performance but on smaller size and power
http://www.androidauthority.com/apple-a10-fusion-chip-performance-723918/
One also has to recall Jim Keller talk about core microarchitectures, where he explained the advantages of the ARM ISA over the x86 ISA and why the efficiency provided by ARM allowed him to make K12 faster than Zen.
One has also to recall the several server ARM SoCs available, which are matching and/or beating Xeons in raw performance but with lower power and cost
https://www.top500.org/news/applied-micro-claims-third-generation-arm-chip-ready-to-take-on-intel-xeon/
http://www.hpcuserforum.com/presentations/santafe2014/Broadcom%20Monday%20night.pdf#page=5
It is more, there are persistent rumors that Intel is in panic, because their analysts predict that ARM will replace x86 sooner than expected, and engineers at Intel are already working in a hybrid CPU that can execute both x86 and ARM code.
Are you aware that the A10 is used in a phone? There is no way that a 1--2W dual-core phone chip provides the same raw performance than 65--220W desktop chips. Do you know some x86 phone with an i7-7700k inside?
But we can compare phone to notebook:
Note he is comparing a 2W chip to a 15W chip, and still the ARM core provides the same performance than the x86 core.
The ARM ISA is more efficient than the x86 ISA, and this efficiency (in transistor count) can be used in different ways. For high-performance servers:
Moar cores. This is the MACOM aproach. They are currently shipping 32-core server CPUs, but a 64-core CPU is in the plans. This is also the Qualcomm approach, with its 48-core chip.
Faster core. This was the K12 core approach. As mentioned by Jim Keller, K12 was wider (and better) than Zen.
A mixture of both. This is the Vulcan approach. The Vulcan core provides 90% of the single thread performance of Haswell core, but there are moar cores in Vulcan chip than in any Haswell Xeon.
weissschnee
Prominent
- Sep 20, 2017
- 1
- 0
- 510
I hope I'm still alive by then. I'm kinda old. Think of the games that can be made with 128 bits. Think of the graphics! The immersion! The sheer realism! Oh, I'm giddy just thinking about it.^_^
jvanasselberg
Prominent
- Nov 18, 2017
- 1
- 0
- 510
You are so right! WE won't ever need it, but once long ago when there was little life on the planet there was no need for much memory then either. And then came the Cambrian Explosion, with the "IoT Cambrian Explosion" it'll be the same. Instead of millions of years it'll be decades and they, our cyber children, will need it! Who knows 128bit might just be the tip of the iceberg. I do know that humanity won't get to see much of it. Carbon life on this planet is doomed.
- Dec 26, 2012
- 30,740
- 4,559
- 142,790
Ya'll are thinking small. You are seeing limitations on exactly what a cpu is or can do as it applies to current or the 'now' needs. If you think back just a few years, a computer was the size of a bedroom, then a desk, then a full tower AT, now you have HTPC and PS4 the size of a college text book. Just how long before a pc with the power we are used to now comes in a package the size of a coffee cup? . How long before the things in StarTrek become reality? How long before there is nanotechnology and cpus the size of a pin head.
And 1024bit...
It wasn't all that long ago when ppl saw only a need for the physical connection, puppets on strings, phones on a wired connection, even Tv's needed cable, and ppl rave about a slight adjustment to the now. All of which boils down to the single most important discovery ever, which is totally glossed over,... The battery.
Maybe ppl can't see a need for 128bit addressing now, tomorrow is a different story.
And 1024bit...
It wasn't all that long ago when ppl saw only a need for the physical connection, puppets on strings, phones on a wired connection, even Tv's needed cable, and ppl rave about a slight adjustment to the now. All of which boils down to the single most important discovery ever, which is totally glossed over,... The battery.
Maybe ppl can't see a need for 128bit addressing now, tomorrow is a different story.
gamerk316
Glorious
- Jul 8, 2008
- 13,234
- 130
- 44,840
All memory really is at the end of the day is external storage for the CPU. So when you say that we'll need a larger address space, you need to ask "what for".
The only thing that currently comes close to saturating the 64-bit address space is DNA modeling. In theory, dynamic physics can require that godly amounts of address space, but we'd need CPUs orders and orders of magnitude beyond what we have to be able to even run a simulation that needed that much physical address space.
64-bit is going to be around for a really long time, and I can't foresee any situation where a home user would need more then 16 exebytes. There's simply no way to actually use it all.
sebastianpalm7
Prominent
- Nov 15, 2017
- 15
- 0
- 520
Just to demonstrate what kind of numbers are involved when talking 128 bit memory addressing.
Imagine the very improbable future when we're able to pack 16 exabytes of memory into a gram of silicon. It would take an information density of just over 1100 atoms per byte to manage this feat, which may some day be theoretically possible, with the liberal applications of space magic.
At that density of information, it would take 19 *teratons* of silicon to store the full addressable memory of a 128 bit processor. If I got the numbers right, 19 teratons of silicon equals not one, but *three* cubes, each 10 km per side, made out of solid silicon. Even if you could bring it down to one byte per atom, you're looking at a solid block that's 10 x 10 x 3 *kilometers*.
We might, some day in the far future, need to build 128-bit computers because RAM has become dense enough to run out of 64-bit space.
We will categorically *never* need a 256 bit CPU, because there's not enough of basically any material laying around to build memory big enough to need one.
Imagine the very improbable future when we're able to pack 16 exabytes of memory into a gram of silicon. It would take an information density of just over 1100 atoms per byte to manage this feat, which may some day be theoretically possible, with the liberal applications of space magic.
At that density of information, it would take 19 *teratons* of silicon to store the full addressable memory of a 128 bit processor. If I got the numbers right, 19 teratons of silicon equals not one, but *three* cubes, each 10 km per side, made out of solid silicon. Even if you could bring it down to one byte per atom, you're looking at a solid block that's 10 x 10 x 3 *kilometers*.
We might, some day in the far future, need to build 128-bit computers because RAM has become dense enough to run out of 64-bit space.
We will categorically *never* need a 256 bit CPU, because there's not enough of basically any material laying around to build memory big enough to need one.
- Dec 26, 2012
- 30,740
- 4,559
- 142,790
Like I said, ya'll are thinking small. I see the word 'currently' used alot. 15 years ago 'currently' ppl couldn't really see a need for anything more than a single core Pentium, dual cores were exotic and pricey. 100 years ago ppl couldn't see the need for those noisy contraptions called automobiles that required a hand crank to start, were semi impossible to drive and were lucky to top 10mph. A good horse was preferable. 'Currently' silicon has issues with the 5nm process, next round of cpus will likely be 7nm at best. You assume that in the next 50 years or less that cpus will still be made from silicon. Could very well be that silicon will be ditched as a ram/cpu base and something entirely different take its place and have no issues with a 0.1nm process.
When the first computer was invented in the late 30's, I'm quite sure the creator had absolutely no idea his invention would end up a million+ times more powerful and go from the size of his parents living room to something smaller than a pinkie nail.
It took less than 30 years for someone with some serious imagination to take a Startrek communicator and end up with today's cellphone. What's going to happen in the next 30 years? Nanotechnology robots diving through the bloodstream destroying cancer cells? Oh, that's already been done by Starwars.
Doesn't take all that long to go from Science Fiction to Science Fact. Think outside the box and sooner or later it'll dawn on you that there really is no box, only one of your own making.
When the first computer was invented in the late 30's, I'm quite sure the creator had absolutely no idea his invention would end up a million+ times more powerful and go from the size of his parents living room to something smaller than a pinkie nail.
It took less than 30 years for someone with some serious imagination to take a Startrek communicator and end up with today's cellphone. What's going to happen in the next 30 years? Nanotechnology robots diving through the bloodstream destroying cancer cells? Oh, that's already been done by Starwars.
Doesn't take all that long to go from Science Fiction to Science Fact. Think outside the box and sooner or later it'll dawn on you that there really is no box, only one of your own making.
gamerk316
Glorious
- Jul 8, 2008
- 13,234
- 130
- 44,840
He wasn't thinking ahead enough; he believed the most general users would need is a decent word processor...
...oh wow, that kinda explains everything now doesn't it.
But seriously: I work in an industry that actually does dynamic physics calculations that legitimately eat up terabytes of RAM; I can categorically say that there is nothing out there that is going to come close to the exabyte scale within the next century, and I doubt ever.
TRENDING THREADS
-
-
-
[SOLVED] Is it safe to use a recapped PSU for a long time ?- Started by zwtch_17
- Replies: 7
-
-
Discussion What's your favourite video game you've been playing?- Started by amdfangirl
- Replies: 3K
Facebook
Twitter
Instagram