Sign in with
Sign up | Sign in
Your question

Is 64 bit necessary?

Tags:
Last response: in CPUs
Share
September 29, 2003 4:32:15 AM

CPU technology has moved from 8 bit (I think that's the oldest but I don't know) to 16 bit, from 16 bit to 32 bit and now from 32 bit to 64 bit. I don't know a lot about processors and this will probably sound stupid, but why not just skip 64 bit and go straight to 128 bit, saving the hassle of a new line of CPUs in 4-5 years?

Roses are <font color=red>red</font color=red>, violets are <font color=blue>blue</font color=blue>, post something stupid and I won't reply to you!

More about : bit

September 29, 2003 5:29:09 AM

maybe no program will support that kind of code yet.
128-bit.

<b><font color=purple>
The real troubles in your life are apt to be things that never crossed your worried mind.
</font color=purple></b>
---
<A HREF="http://www.mapua.edu.ph/" target="_new">
MIT
</A>
September 29, 2003 6:22:19 AM

Moving to a bigger default data set size is a double-edged sword. On the one hand, programs that use numbers that big can benefit. On the other hand, programs that do not use numbers that big will not and may even run slower, at the cost of extra hardware and complexity of design.

Currently, I know of no tasks (except for a few that consumers will never touch in the next 5-10 years) that use numbers so big that it requires 128-bit integer precision. So the question isn't really why we haven't jumped to 128-bit MPU's, the question is, "why would we?". There's little if any benefit to be had on any level at this point.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
Related resources
September 29, 2003 7:58:36 AM

I was also pretty much excited in getting an Athlon64 this holidays, but i decided to wait till AMD releases Athlon 64 "San Diego" core which is the second revision of K8 architecture.

I frequently avoid first-release new generation designs like the clawhammer (K8).

<b><font color=purple>
The real troubles in your life are apt to be things that never crossed your worried mind.
</font color=purple></b>
---
<A HREF="http://www.mapua.edu.ph/" target="_new">
MIT
</A>
September 29, 2003 8:00:43 AM

Quote:

I frequently avoid first-release new generation designs like the clawhammer (K8).


Like the WILLAMETTE pentium 4.
I do have the earliest version of northwood which is 2.0ghz and runs very damn coooool.

<b><font color=purple>
The real troubles in your life are apt to be things that never crossed your worried mind.
</font color=purple></b>
---
<A HREF="http://www.mapua.edu.ph/" target="_new">
MIT
</A>
September 29, 2003 8:05:02 AM

I think 64bits are most certainly a good thing. This at least takes us that 1 step closer to building the faster PC. If this does not happen now, it will in the future. Later on we may have performance bottlenecks with 32bits and at least with going 64bit now, that will not happen. We have to remember that everything we do now, effects standards we will use in the future.
September 29, 2003 9:22:02 AM

64bit is one of those things that very few people need NOW, but will probably end up being taken for granted in the future.

If I had built a computer with 1024MB of memory 5 years ago (probably not possible), most people would have thought I was insane. However, I have several applications (including some games) that make good use of the 1024MB that I have in my current setup.

"Some mice have two buttons. Macintosh has one. So it's extremely difficult to push the wrong button." - Apple ad. circa 1984.
September 29, 2003 9:44:37 AM

Well, 4 bit was the oldest actually, the Intel 4004 in 1971.

64 bit technology itself is big leap, it will take a few years till we actually make use of such power. It is only the scientific and engineering applications that need such immense computing power and accuracy, almost all of our day to day stuff runs fine on 32 bit. Of course, heavy graphics 3D games are one candidate for heavy computing platform that can take good advantage of 64 bit architecture.

Making a processor is a complex job, and its complexity increases exponentially with its "bitness" - the word size it can handle. A lot of money has gone into making 64 bit processors, and a lot more will go into making 128 bit ones. Why waste good money on things we dont really need right now?

girish

Every point I make has 'n' perspectives!
Boy do I need a <i>disclaimer</i> for my every word?
September 29, 2003 9:54:09 AM

Quote:
Like the WILLAMETTE pentium 4.
I do have the earliest version of northwood which is 2.0ghz and runs very damn coooool.


Bingo! I never use the first version of any software, esp Micro$oft ones. Remember Windows 98 First edition?

BTW P4 wasnt going to make any bang at less than 2GHz anyway.

Buying a 64 bit chip today is waste, there are no applications for it nor any need, and it ripss thru your nose. I'd wait for a year or two untill it comes down to some sane levels, and I can actually run some 64 bits apps.

girish

Every point I make has 'n' perspectives!
Boy do I need a <i>disclaimer</i> for my every word?
Anonymous
a b à CPUs
September 29, 2003 10:13:04 AM

Going from 16 bit to 32 bit increased addressable memory by not by 2, but by 65.000. Not bad, however 64 for bit is not twice as much as 32 bit either.. its <b>4 billion</b> times as much..

Let's just focus on memory addressing to get an idea of what we are talking about. Assume 1 GB Dimms are 5mm thick.
*) A 32 bit machine can address 4 of those, put them on a pile, and you get 2cm.
*) PAE allows 36 bit addressing, that gives you 64 GByte, 32 cm or a pile as high as a few good books.
*) Opteron (and I think Itanium) allow up to 40 bit physical addressing. That is one terrabyte, or the equivalent of over half a meter of 1 GB Dimms per cpu. Not very practical yet.
*) Opteron and Itanium allow 48 bit virtual memory addressing, 48 bit represents 256 terrabyte, or a tower off 131 meter of 1 GB dimms. should be enough for a while
*) If you'd fully exploit the 64 bit addressing that means 16 million terrabyte, or a pile off well over 8000 kilometer high !! most satellite orbit below 1000 km. See how ridiculous 64 bit addressing is already ? 32 bit doesnt always cut it anymore, but 64 bit is MASSIVE overkill.
*) Now, you are suggesting 128 bit. that allows 309.485.009.821.345.000.000.000.000 terrabyte addressing. Even using hypothetical terrabyte DIMMS, that is a pile of 16 million LIGHTYEARS (roughly 154.742.504.910.673.000.000,00 kilometer).

You see.. there are 10 kind of people; those that understand binary, and those that don't :D 

= The views stated herein are my personal views, and not necessarily the views of my wife. =<P ID="edit"><FONT SIZE=-1><EM>Edited by bbaeyens on 09/29/03 06:20 AM.</EM></FONT></P>
Anonymous
a b à CPUs
September 29, 2003 10:18:10 AM

afterthought.. :

>Even using hypothetical terrabyte DIMMS, that is a pile of
>16 million LIGHTYEARS

Imagine memory latencies on such a system.. 8 million years on average. A few hexabyte L2/3/4/5/6 cache would be welcome here :D 

= The views stated herein are my personal views, and not necessarily the views of my wife. =
September 29, 2003 10:23:51 AM

hmmm


1) Build a comp with 128bit processor/memory management
2) Make stackable DDR Dimms (TB/each)
3) Make a bridge across the galaxy

Is that what you are suggesting?
I like that idea...

-Sig-

nVidia didn't make me an nVidia fanboy, ATi, and lack of other serious competition made me an nVidia fanboy.
September 29, 2003 10:26:36 AM

didn't we get the same argument for 32 bit at one time?

Also, are there ANY 128 bit computer processors out? I know 64bit has been around a few years, but 128?

-Sig-

nVidia didn't make me an nVidia fanboy, ATi, and lack of other serious competition made me an nVidia fanboy.
September 29, 2003 10:39:14 AM

I know that 2 to the power of X has a lot of POWER but I wouldn't be caught saying how far technology will reach, too say the least about home PCs. Bill Gates said that 640K should be sufficient storage space for anyone.
Now, I know that the numbers sound, well, pretty big, but I still think that in 10-15 years computers will require that amount of system resources for uses not imaginable today (for example, maybe pinpoint simulation of real-life events) and the DIMMs probably won't be a tower high (they would evolve too).

Roses are <font color=red>red</font color=red>, violets are <font color=blue>blue</font color=blue>, post something stupid and I won't reply to you!
Anonymous
a b à CPUs
September 29, 2003 1:02:27 PM

>Now, I know that the numbers sound, well, pretty big, but I
>still think that in 10-15 years computers will require that
>amount of system resources for uses not imaginable today

I doubt it. Memory density and capacity roughly doubles every 18 months. That has been true for ages, both disk space and RAM. Bill should have seen that one coming, but instead he thought it would level off. Now apply this 18 month doubeling to my example:
2 GB Dimm available now and lets say 2 GB/CPU "common".
256 GB Dimm in over ten years
2048 GB dimms in 15 years, around the year 2018

So, we would have 2 terrabyte "Dimms" by 2018.. 64 bit cpu's could still address 128 times more.
128 bit cpu's could take 154742504910673000000000000 of those same dimms. that is still a pile of several million light years.

trust me, 64 bit (addressing) is enough for the next 15-20 years at least. And by that time, God knows how things will have changed; there is little use in designing a chip or ISA with that in mind.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
September 29, 2003 3:35:12 PM

64-bit computing is not faster than 32-bit computing. All it does is enable the use of a larger linear memory space.

Athlon 64 thanks its increased performance to the integrated memory controller and extra registers, not because it's 64-bit.
September 29, 2003 4:22:04 PM

> that is still a pile of several million light years.

I was about to reply and say "alright, it's a lot, but it's not THAT much..." But then I worked it out, and I'm coming up with about 814 million light years. Heck, that's a tall stack o'RAM.

-Anthony
http://nodivisions.com/
September 29, 2003 5:58:58 PM

And someone would have to find a way to convert light into matter, namely silicon, just for the resorces.
September 29, 2003 6:14:36 PM

actualy, I was talking about the integer calculation apspect of processors, I would probably be a while before we see a gazillion TB RAM.

Roses are <font color=red>red</font color=red>, violets are <font color=blue>blue</font color=blue>, post something stupid and I won't reply to you!
September 29, 2003 7:02:15 PM

I don't think you know what a light year is, it's not actual light, but the distance the light travels during 1 year (light gets to the moon in 2 secs and to the sun in 7 minutes, so you can understand it's no long walk to work).

Roses are <font color=red>red</font color=red>, violets are <font color=blue>blue</font color=blue>, post something stupid and I won't reply to you!
September 29, 2003 8:09:02 PM

I understand that the switch from 64 to 128 won't help in terms of memory allocating, but I read that games will gain from using 64 bit integers (when those games will be released) so I guess that there are also advantages for using 128 bit integers.
I know for example that when I work with Visual Studio 32 bit integers is usually not enough for seeing the computer work a second on calculating an equation, so when the instructor said that a program wasn't efficient, I had to take his work for it 'cause I couldn't see any difference (I had to use long or double, but I didn't bother).

Roses are <font color=red>red</font color=red>, violets are <font color=blue>blue</font color=blue>, post something stupid and I won't reply to you!
Anonymous
a b à CPUs
September 29, 2003 8:11:29 PM

>But then I worked it out, and I'm coming up with about 814 >million light years

Either you got your math wrong, or use a DIMM height that is other than mine, cause I got 150 million light year. Either way, waaaaaaaaaaaaay beyond what we can even begin to imagine.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
Anonymous
a b à CPUs
September 29, 2003 8:22:54 PM

You are mixing up a lot of things here..

> I read that games will gain from using 64 bit integers
>(when those games will be released)

Hardly. Games will benefit from the new "64 bit long mode", mostly because of the extra registers that mode exposes on these cpu's (16 instead of 8). Therefore you need less register renaming and shuffeling which takes up quite a few cpu cycles. Now, some apps do benefit just from being able to work with 64 bit ins; mostly mathematical things, encryption and such. Speedups can be as much as 4x in theory (64 bit multiply versus 4x32 bit multiply) But games, I wouldnt think so..

>I know for example that when I work with Visual Studio 32
>bit integers is usually not enough

Hu ? are you sure you are talking INTEGERS, and not floats ? I'd be curious to see what the hell you are coding that a 32 bit int isnt enough. Either way, the compiler has no trouble converting a 64 bit INT into 2 32 bits. It does impose a small performance hit.. but it should be a very rare occurance. Floating point numbers on the other hand, may require more precission / range. But x87 FP already supports 80 bit for decades..

= The views stated herein are my personal views, and not necessarily the views of my wife. =
September 29, 2003 8:27:40 PM

> Either you got your math wrong, or use a DIMM height
> that is other than mine,

I used a DIMM _width_ of 5mm.
September 29, 2003 9:02:06 PM

I was just programming a simple program that calculated the factorial of a number, but in recursion which is supposed to be a lot slower, and the factorial of a number gets very high, the maximum of a 32 bit integer is 2.147B and 13! is already 6.227B (20! = 2432902008176640000). So 32 bits isn't that big...

Roses are <font color=red>red</font color=red>, violets are <font color=blue>blue</font color=blue>, post something stupid and I won't reply to you!
Anonymous
a b à CPUs
September 29, 2003 9:26:22 PM

>I used a DIMM _width_ of 5mm.

LOL, yeah well, width, height.. if you stack them on top of each other you lay them horizontally, hence width becomes height :) 



= The views stated herein are my personal views, and not necessarily the views of my wife. =
September 30, 2003 7:36:08 AM

unsigned 32-bit ints is 4 billion roughly. Also, you'd have to consider what the critical loop of your program would be. Are you looping 10 billion times to find out the factoral of different numbers?
Most performance-limited applications today are FP-intensive. Hardly any require huge integer performance anymore.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
!