128-bit architechture | The Coffeehouse

exoteric
I : Next

The recent discussion/speculation around a possible Windows 8/9 with a 128 bit architechture sounds interesting if there is any truth to it, which of course none of us know. What are the possible requirements that would drive this design? It reminds me of ZFS which is also 128 bit.

So how much is 64 bit?

18,446,744,073,709,551,616 bits

2,097,152 terabytes

2,048 petabytes

That's a pretty big number. No single computer is likely to have that much space for a long time to come.

On the other hand, if we talk about this globally, we're talking about 2 million people with a 1 terrabyte disc. That threshold has already been exceeded.

128 bit on the other hand is so perverse that it can't possibly be exceeded

Then there's cloud computing.

Is this about creating a shared virtual memory address space for the whole world wide web or a really big cloud?

Sounds cool.

Is there a global memory counter for the world? A mechanical (and possibly a biological) one (although I don't believe we know the memory power of the brain yet?)?

spivonious

I can't wait for my 2 exabyte system to be released.

One big shared address space....that's a little scary sounding. Would a buffer overrun let me modify someone else's cloud?

figuerres
???

In reply to spivonious

#Oct 13th @ 6:49 AM

Perplexed

reminds me of an old 60's or 70's song.

STones I think?

Hey hey You Get Off of MY CLOUD! .... Big Smile

Blue Ink
C you

Sorry, but I don't get the logic behind your numbers. Why would you want to address single bits? Using 64 bit addresses you should get up to 16 exabytes.

And using 64 bits to address the clusters in a HDD, the number becomes humongous: with 4KiB clusters, we would get 64 zettabyte... but there is no reason why we shouldn't use, say, 1MiB clusters instead (if we were able to manufacture storage devices that large, the waste would be irrelevant), for a grand total of 16 yottabytes. Anything larger and we will have to invent a new SI prefix.

I don't know if a 128 bit CPU has some practical use at the moment, but I'm pretty sure it has nothing to do with the larger addressing it provides... quadruple precision floats would be nice, though.

exoteric
I : Next

In reply to Blue Ink

#Oct 13th @ 8:31 AM

D'oh, you are absolutely right. Hm, are there applications that require this precision outside of scientific computing?

Dodo
I'm your creativity creator™ :)

In reply to Blue Ink

#Oct 13th @ 8:31 AM

Yup, precision and unicycle performance. 256bit and 512 instruction cores @1GHz and we're talking. Smiley

Blue Ink
C you

In reply to exoteric

#Oct 13th @ 8:39 AM

Probably not... but that's good enough for me Smiley

On second thought, 128 bit would also speed up business calculations (the .NET decimal data type is 128 bit, if I remember correctly).

CreamFilling512

What is 128-bit? The data or address bus? 128-bit addressing makes no sense. Using the full extent of 64-bit addressing isn't even practical, the huge amount of paging tables needed to support it.

dneomatrix

I think be cool if we could some sort of instruction scheme that allows us to do something like 2 64 bit operations, (or even 3 32 bit operations plus overhead bits) at the same time. Then if you could use that as some sort of insane hyper-threading... possibilities are pretty nice. Say running your main machine and 2 virtual machines at the same time -> literally having the instructions execute concurrently, as if on different cores. This turns a quad-core machine into a 12 core machine... =). Wishful thinking, but a cool thought.

exoteric
I : Next

In reply to CreamFilling512

#Oct 13th @ 10:22 AM

See this the cached LinkedIn profile page. Then see if it makes any sense whatsoever in any way. I don't know.

CannotResolveSymbol
{insert caption here}

In reply to dneomatrix

#Oct 13th @ 10:56 AM

That would be SSE (in IA32) or Altivec (in the POWER architecture).

Intel also does some trickery to "simulaneously" execute instructions on a single core; that's what Hyperthreading is.

Sabot
My name is Dave Oliver. I'm a Technical Architect.

In reply to exoteric

#Oct 13th @ 11:22 AM

The most obvious reason why you would want to increase the bit length and think about 128bit, 256bit and so on is Cloud Computing!

Cloud computing is about seemlessly allowing a program and it's data to expand and contract, you pay for as much or as little you use.

Cloud computing relies on the art of distributed computing. Lots of machines working together. Increasing the bit length means that you won't need so many machines in your cloud data-centre which will drive down costs especially as servers have such short lives (in comparison to other electrical devices such as the alarm clock or the TV) the move by hardware vendors will be looking to entice Cloud Computing vendors into justifying the costs of the new machines on reducing costs (not so many machines needed) and complexity justification (again not so many machines needed)

CreamFilling512

Processors have executed instructions in parallel since the advent of pipelining. It's not threading though.

exoteric
I : Next

In reply to Sabot

#Oct 13th @ 2:40 PM

I understand the benefits of cloud computing to be dynamic scalability (on-demand elasticity), global redundancy (globally dispersed data centers), transparent failure (the cloud OS allocates nodes and resources), etc. In particular dynamic scalability seems like a crucial characteristic of cloud computing. This will of course help drive down costs as you pay for what you use, not more and the infrastructure costs are shared amonst all cloud users, so there is a low barrier to entry. I see virtualization (sub-machine / virtual machine allocation) and dynamic scalability as ways to decrease costs.

On the other hand, I can't quite see how an 128-bit address space would help reduce costs or the need for more machines. As Blue Ink writes, 64-bit allows you to address 16 exbibytes or almost 17 mn terabytes.

Now enter 128-bit land and this expands to 3,4028236692093846346337460743177e+38 in scientific notation. Now multiply that by 8. So 256-bit would appear out of this world - if not universe.

CannotResolveSymbol
{insert caption here}

In reply to exoteric

#Oct 13th @ 3:14 PM

Fun statistic: one estimate of the number of the number of atoms on earth is ~8.87*10^49 (first result from Google).

2^256 (the maximum number of bytes that would be addressable by a 256-bit processor) is 1.158*10^77. That's almost 30 orders of magnitude larger than the number of atoms on the earth-- it would be physically impossible to build enough memory to take full advantage of a 256-bit addressing scheme.

That said, there is some value in being able to do computations quickly on a 128-bit or 256-bit number-- it would be useful for some types of simulations, among other things. You wouldn't need a whole new processor architecture to be able to do this, though-- IA32 has included 128-bit registers since SSE was introduced, and (although an instruction's not there to do so now as they are intended to be vector registers) one could imagine SSE5 (or later) including instructions that operate on all 128 bits together in the future.

Sven Groot
My name has 9 letters. Coincidence? I think not...

In reply to CannotResolveSymbol

#Oct 13th @ 5:31 PM

That reminds me of a really, really old COM tutorial on MSDN, written by Dr. GUI, which claimed that the number of possibilities for a GUID (2^128) was larger than the number of atoms in the universe. Which isn't even remotely true. Smiley Iirc, he explained in the next article that he got an e-mail demonstrating that even just the sun would have far, far more atoms than that.

Minh
WOOH! WOOH!

In reply to Sven Groot

#Oct 13th @ 5:45 PM

Hey, bit-ness used to mean how many bytes you can copy in one operation...

I'm not sure if SIMD has made that obsolete... but wouldn't it be advantageous to be BIG-bit?

edit: damn you CannotResolveSymbol... what he said

Blue Ink
C you

In reply to Sven Groot

#Oct 13th @ 5:45 PM

Apparently calculations at that scale tend to confuse even the brightest minds. Erring on the other side, Andrew S. Tannenbaum, in his "Computer Networks", speaking of IPv6 claims that: "While it was not the intention to give every molecule on the surface of the earth its own IP address, we are not that far off".

Actually, with 128 bits we could have a density of about 7000 IP addresses per square Ångstrom (oceans included)...

Now, that's overkill!

Sven Groot
My name has 9 letters. Coincidence? I think not...

In reply to Blue Ink

#Oct 13th @ 8:04 PM

IPv6 addresses are structured though, which makes it not quite possible to have 2^128 distinct meaningful addresses.

magicalclick

I don't get it. If you have a cluster or cloud, each machine can have 2^64 memory already. Meaning if you have 10 PC in a cluster, you get 10x2^46 memory in totall (logically) if you divided the work properly to each machine. I don't see why you need a giant big ass 2^256 machine to do all the stuff.

exoteric
I : Next

In reply to CannotResolveSymbol

#Oct 13th @ 5:31 PM

It's true what they say: Wolfram|Alpha is your friend. And one should listen to ones friends. Begone random guessing.

It appears 2^266 is enough to count to the number of atoms in the visible universe.

Stanford researchers have demonstrated with Electronic Quantum Holography how to store 35 bits per electron.

Did someone say overkill?

AndyC

In reply to magicalclick

#Oct 13th @ 8:45 PM

But there are advantages to being able to address memory on other machines as if it were local to the current node, see things like Infiniband or Myrinet. And then there are the advantages of a vast reduction in virtual address space fragmentation: it's much easier to allocate a multi-gigabyte array when you can be pretty sure of finding a suitable window in the address range.

Dodo
I'm your creativity creator™ :)

In reply to exoteric

#Oct 13th @ 9:43 PM

35 bits per electron? And what do you use to read or write to the storage device (the electron) and how many nuclids does that contain?

exoteric
I : Next

In reply to Dodo

#Oct 14th @ 8:25 AM

Full details on the Stanford press page

magicalclick

In reply to AndyC

#Oct 13th @ 11:30 PM

It certainly offers more flexibility. But we also need to be careful not getting spoilded. If we are getting lazy and forget about compression and so on, we will be in trouble.