Windows 8

[Sybex]

Reading today I hear that Windows 8 may be a 128 bit OS and is scheduled for 2012

http://www.pcadvisor.co.uk/news/index.cfm?newsid=3203583

[Mr Banana]

Reading today I hear that Windows 8 may be a 128 bit OS and is scheduled for 2012

http://www.pcadvisor.co.uk/news/index.cfm?newsid=3203583

It won't be a 128bit OS but it has plans to partially support it, while Windows 9 will have full support.

Although honestly it's just over-hype for something the desktop won't need for a long time. Will be useful for servers to handle workload of sites far before the desktop.

But still i long for the days OS's and Games/Applications that actually make proper use of many cores in a processor.

[Vic]

Reading today I hear that Windows 8 may be a 128 bit OS and is scheduled for 2012

http://www.pcadvisor.co.uk/news/index.cfm?newsid=3203583

Sadly, that article appears to be a triumph of buzzword-recognition over journalistic process.

The LinkedIn leak upon which the whole rumour is founded states that Robert Morgan's projects include "128-bit architecture compatibility with the Windows 8 kernel". "Compatability" just means that he's doing some work to make sure a port to 128 bits is possible, not that it's featured.

For Win8 to be 128-bit, there would need to be some 128-bit processors to run it. I haven't seen many of those in recent years, and if they haven't been announced yet, they're unlikely to be ready for 2012.

So this is actually good news - Microsoft appear to be architecting their kernel so as not to be painted into a corner again - but Windows 8 won't be 128-bit.

Vic.

[Vic]

Will be useful for servers to handle workload of sites far before the desktop.

Actually, I'm not even sure that's true. 32-bit kernels can handle 4GB *per process* (although Windows XP is specifically crippled to prevent this); most server apps could actually run in 32 bits.

Upgrading to 64 bits gives you a theoretical maximum of nearly 17 exabytes per process. That's 4 billion 4GB DIMMS per process - it's a lot of memory.

I can't see 128 bits being particularly useful except in large scientific calculations, where long-word calculations are already handled perfectly well. The overhead of upgrading to a 128-bit architecture for such applications would seem to be overkill...

Vic.

[Mr Banana]

My point is that it would be useful for servers BEFORE desktops, i wasn't saying that it will be useful anytime soon ;).

I know it is overkill for now, since more people still use 32bit than 64bit.

[Vic]

My point is that it would be useful for servers BEFORE desktops

Yes, I got your point - I was just disagreeing with it.

A 64-bit OS can address a *vast* amount of memory. Since I wrote my first application, the amount of memory available on a computer has increased by about 8 orders of magnitude. But it would have to increase another 8 orders of magnitude to approach the *per-process* (not per-CPU) limit for a 64-bit OS.

These are very big numbers; I firmly believe that the maximum addressable range of a 64-bit OS will never be exceeded. An exabyte is a very large amount of memory...

So what good is a 128-bit processor? It's no better than a 64-bit one for memory addressing, so the only advantage is in the data width. Having more bits allows you to handle larger ranges of values without resorting to accuracy-sapping floating-point, or it allows you to run more parallel operations in a SIMD sequence. That means that it is best suited for desktop applications, where these things are useful. Server apps tend to do many instantiations of a simpler process - so servers like memory and I/O, not number-crunching capabilities.

But these features are application-level; the OS doesn't need to know anything about them (you don't run your main application datapath through the OS!). So it is wise to ensure that you don't build incompatabilities into the OS, but that is a very long way from making it a 128-bit OS.

Vic.

[Mr. Lime]

Yes, I got your point - I was just disagreeing with it.

A 64-bit OS can address a *vast* amount of memory. Since I wrote my first application, the amount of memory available on a computer has increased by about 8 orders of magnitude. But it would have to increase another 8 orders of magnitude to approach the *per-process* (not per-CPU) limit for a 64-bit OS.

These are very big numbers; I firmly believe that the maximum addressable range of a 64-bit OS will never be exceeded. An exabyte is a very large amount of memory...

So what good is a 128-bit processor? It's no better than a 64-bit one for memory addressing, so the only advantage is in the data width. Having more bits allows you to handle larger ranges of values without resorting to accuracy-sapping floating-point, or it allows you to run more parallel operations in a SIMD sequence. That means that it is best suited for desktop applications, where these things are useful. Server apps tend to do many instantiations of a simpler process - so servers like memory and I/O, not number-crunching capabilities.

But these features are application-level; the OS doesn't need to know anything about them (you don't run your main application datapath through the OS!). So it is wise to ensure that you don't build incompatabilities into the OS, but that is a very long way from making it a 128-bit OS.

Vic.

And given the fact that the people who make the components seem to have their hearts set on doubling every year, just how long do you think it will take? lol

[Vic]

And given the fact that the people who make the components seem to have their hearts set on doubling every year, just how long do you think it will take?

That's exactly my point - I don't expect *ever* to have that much memory around.

Exabytes are big. They're a thousand petabytes each.

Petabytes are big. They're a thousand terabytes each.

Terabytes aren't small - they're a thousand gigabytes each.

I really cannot imagine what data you would hold in any process that would take up more than 16 exabytes. To add a little context: that's more than 2GB of address space for every human on this planet, and it's available to every process on the cmoputer.

And remember that these are *current* values. This is what 64-bit addressing gives you.

Vic.

[Mr. Lime]

Heck at a doubling rate Ram sizes themselves will have hit more than 16 exabytes by 2041, granted we dont need it now but when has that ever stopped us? :D We seem content to carry on making bigger and better things, whether we need them or not :D

[Vic]

Heck at a doubling rate Ram sizes themselves will have hit more than 16 exabytes by 2041

You're assuming a year-on-year doubling of RAM sizes; I'm contending that that will not happen.

There comes a point where you don't have enough information to put into that memory. I put it to you that if you were given a year to compile a 16 exabyte lump of data, you would be hard-pressed so to do - even if you were just to copy the same DVD image into your output, over and over again[1]. It's a *vast* amount of space.


granted we dont need it now but when has that ever stopped us? :D WE seem content to carry on making bigger and better things, whether we need them or not :D

But you're not talking about "bigger and better", you're talking about "expensive and unusable".

Vic.

[1] If we take a DVD as being about 5GB, you'd need to make 3 *billion* copies of it to fill 15EB. That's roughly 10 million copies a day,every day for a year...

[Mr. Lime]

You're assuming a year-on-year doubling of RAM sizes; I'm contending that that will not happen.

There comes a point where you don't have enough information to put into that memory. I put it to you that if you were given a year to compile a 16 exabyte lump of data, you would be hard-pressed so to do - even if you were just to copy the same DVD image into your output, over and over again[1]. It's a *vast* amount of space.



But you're not talking about "bigger and better", you're talking about "expensive and unusable".

Vic.

[1] If we take a DVD as being about 5GB, you'd need to make 3 *billion* copies of it to fill 15EB. That's roughly 10 million copies a day,every day for a year...

Well do we currently need quad core processors? No as someone earlier in this thread pointed out we do not currently use their full potential, and yet we have them, designers will try and keep doubling as long as they can

[jonbanjo]

The way I see it, we are not just doubling here. n is getting doubled in 2^n. I don't really see a sequence like this continuing to "double" and be useful:

..8: 256
.16: 65536
.32: 4294967296
.64: 18446744073709600000
128: 340282366920938000000000000000000000000
256: 11579208923731600000000000000000000000000000000000 0000000000000000000000000000

[snakedoc]

The way I see it, we are not just doubling here. n is getting doubled in 2^n. I don't really see a sequence like this continuing to "double" and be useful:

..8: 256
.16: 65536
.32: 4294967296
.64: 18446744073709600000
128: 340282366920938000000000000000000000000
256: 11579208923731600000000000000000000000000000000000 0000000000000000000000000000

The chairman of IBM was quoted to have said "The World will need about 5 computers"

Things change and very quickly in the world of computing. Who knows how much data we will be throwing around our systems in 30 years time? When I was a kid with my 1k zx80 a PC of today was nothing but the dreams of a madman.

[jonbanjo]

Things change and very quickly in the world of computing. Who knows how much data we will be throwing around our systems in 30 years time? When I was a kid with my 1k zx80 a PC of today was nothing but the dreams of a madman.

Point taken.

[Helior]

I think it's very improbable that 128 bit will be introduced to the mainstream any time in the foreseeable future.

As jon as pointed out, it's only conceivable use at present is for floating point calculations.

I won't say never ever, but currently a 64bit OS can potentially cache all data known to man in memory.

[IainK]

i think it's ridiculous to think that it's ridiculous that we'd ever need that much because we simply can't tell the future, however i think it's also ridiculous to assume that we will ever need that much for the same reason

IMO we've already experienced the fastest growth in computer technology that we're going to see

[PMM]

Its not really a memory thing and our processors actually have 128bit MMX registers within them

Currently these resisters are used in a way that you fit in say 2x64bit instructions or 4x32bit instructions or 8x16bit instructions.

In a sense it is what x64 is to the 32bit architecture when it was implement some years ago by AMD.

But to process in 128bit has a performance gain being able to do something in say 1 instruction cycle verse 2 or 4 instruction cycles to work with something stored in a 128bit register.

Therefore Microsoft have to approach next gen CPU's from 2 levels - 128bit instruction sets and future 128bit processing esp considering the already available chip within Sony Playstations the up and coming AMD Bulldozer/Fusion stuff and Intels Larrabee its very valid work they have to be doing now.

The whole GPGPU mechanics also evolving into up and coming CPU's and potentially the requirement to communicate 128bit words between the 2 natively also.

[IainK]

We all just got pwned by PMM :)

(i understood about 30% of that so i'll just believe you ;))

[Helior]

I wonder if we'll get a new kernal in windows 8?!

[Mike]

We'll need that kind of storage and even better processing to store the human mental state. Otherwise how're we going to upload ourselves for the online gaming of the future? ;)