[iboy]

When will the MacOS be an 64-bit operating system? In 1 year, 2 years...????

gr�essli

yob

[Millennium]

First of all, do you actually know what it means to be a 64-bit operating system, and exactly how such an OS would benefit you?

As it currently stands, Smeagol -a special build of Jaguar which shipped with the very first Power Mac G5's- had some 64-bit optimizations. Panther has more, obviously, so you could say that we're probably about 25% of the way there.

I would guess that we won't see a 64-bit-only OS for at least two years; I doubt Apple will go that far until the entire line uses G5 processors. Since G4 processors are 32-bit by design, a 64-bit OS would run much more slowly on these chips than a 32-bit one, if it could be made to run at all.

[iboy]

yes, you are right... Panther has some 64-bit optimizations BUT is the system overall much faster on a G5 compared to a G4?

I'm asking because I want to buy a new Powerbook next year (maybe spring, maybe summer) and now the question is: when the System isn't that much faster on a G5 then I will stick to the next revision Al G4 (???the last???). Beside, I don't like "first-generation-products" anyway!!!

gr�essli

yob

[Chuckit]

Originally posted by iboy:
I'm asking because I want to buy a new Powerbook next year (maybe spring, maybe summer) and now the question is: when the System isn't that much faster on a G5 then I will stick to the next revision Al G4 (???the last???).

The system will be faster on a G5 than a G4 just by virtue of the fact that the G5 is a faster chip than the G4. I doubt making it 64-bit native would have anywhere near as big an impact on speed as that. "64-bit" does not equal "faster."

[Arkham_c]

The only feature that I know of 64-bit operating systems that OSX does not currently support is allowing more than 4GB of RAM per process. While this is a limitation, and one I expect Apple to overcome, it's not one of great concern. You can access memory in OSX in 64-bit chunks, and can use 64-bit memory pointers to get past the 2GB RAM limit, so some would argue that it it already a 64-bit OS.

[Eriamjh]

It will be 64-bit eventually. But if I said it was on June 1, 2005, would it really matter?

[iboy]

Is there a competition G5 <---> G4 somewhere? I already checked http://barefeats.com/ but I didn't find any shotout :-(

Yes I know that the 64-bit optimazion will be done step-by-step... but when it is this way that I can't use the full performance out of a G5 because the operating system isn't optimized enough then I will stick to a G4!

gr�essli

yob

[Angus_D]

Originally posted by iboy:
Yes I know that the 64-bit optimazion will be done step-by-step... but when it is this way that I can't use the full performance out of a G5 because the operating system isn't optimized enough then I will stick to a G4!

What do you mean by "optimized", though? I suspect that you don't really know. Speed is not directly related to whether or not the OS is 64-bit native. The G5 is faster than the G4 not just because of the chip but also because of the entire system architecture. There have been comparisons made in several places. See for example:
http://www.insidemacgames.com/review...?ID=423&Page=4
http://www.pcmag.com/image_popup/0,3...d=28726,00.asp
http://www3.macintouch.com/g5bench.html
http://www.macworld.com/2003/09/revi...g5testresults/

[Detrius]

Originally posted by iboy:
Yes I know that the 64-bit optimazion will be done step-by-step... but when it is this way that I can't use the full performance out of a G5 because the operating system isn't optimized enough then I will stick to a G4!

There is a big difference between the OS being fully 64-bit and being optimized for a processor. For example, if the OS did not take advantage of Altivec, there would be a big speed hit on the G4s and later because entire pipelines will not be used. However, a 64 bit processor is not by definition faster than a 32 bit processor. In fact, 64 bits is twice as much data as 32 bits and therefore will take twice as much time to move around. If you don't have more than 4GB of RAM, you will see no benefit from having a fully 64-bit OS. You will not be losing any performance from the bit-ness of the OS.

If the G5 is beating the G4 on the exact same code, the G5 would be the logical choice -- especially if the software is not optimized for it.

[Catfish_Man]

Note also that 64 bit (barring other changes) can be slightly slower than 32 bit due to increased cache usage by 64 bit pointers. 64 bitness isn't a magic bullet for most tasks. The G5 is much faster than the G4, just not because it's 64 bit.

[Millennium]

Originally posted by Detrius:
In fact, 64 bits is twice as much data as 32 bits and therefore will take twice as much time to move around.

Actually, this would only be true on a 32-bit chip, such as the G4.

What makes a chip 32-bit, or 64-bit, or whatever, is how long a word is. You may know that eight bits is a byte; a word is a collection of bytes of a specified length. On a 16-bit computer, a word is 16 bits. On a 32-bit computer, it is 32 bits. And on a 64-bit computer, a word is 64 bits long.

Now, what is the point of this? The point of this is that the system bus on a chip is one word wide. This is more of a common convention than a formal definition, but it remains true in most architectures, and the PowerPC architecture is no exception. The width of a bus indicates how many bits travel down it at the same time. A 64-bit computer can process 64 bits at once, and so it does this in about the same amount of time that it could process 32.

Here's a metaphor. Let's say that I have 20 people that I want to get into a building as quickly as possible. Let's say that one person can move through a door in one second.

If I have one door into the building, it will take 20 seconds. If I have five doors, it will take four seconds, because five people can enter at once.

Now, let's say that 20 more people gather, so I need to get 40 people into the building. If I still have five doors, it will take twice as long: eight seconds. But if I open five more doors, then it will take only four, because now ten can enter at once.

This is what it means to be a 64-bit computer: it has twice as many "doors" as a 32-bit computer, so it can move twice as much data in the same amount of time. That's really all there is to it. There's nothing magical about it at all.

So, why do people tend to base so much around the system bus? For example, why can 32-bit computers only access 4GB of RAM?

The truth is, the limitation is somewhat artificial: you could make a 32-bit computer access more than 4GB of RAM. But to do it, you would need to use more than one word to provide each memory with an address, and that would slow everything down (it would take at least twice as long to access memory in this way). Similarly, you could use less than one word to address memory (thus reducing the amount of memory you could put into the machine) but this wouldn't make any sense; each memory access would waste some bits.

But words aren't just used for accessing memory. There's the matter of numbers as well. At the moment, integers are usually stored in 32 bits of memory, because that's the common word size for most computers today. This will get you up to about 4 billion if you only want to count positive numbers; if you need negative numbers then you'll only get 2 billion or so each way. If you want bigger numbers than that, you need to use two words. 64 bits will get you some 18 quintillion numbers -9 quintillion each way if you want negatives- which is almost enough to express the current estimated age of the universe in seconds. That makes it the most common choice for big numbers nowadays, though some databases use 128 bits. Anyway, using multiple words slows things down; making the word size bigger speeds it up again. Under a 64-bit machine, 64-bit integers are as fast as 32-bit ones, and even 128-bit integers get a significant speed boost. This may not seem like much; after all, who uses numbers that big? The answer: many graphics-heavy applications, such as games, use numbers that big.

So it's really a matter of economics, not in terms of money but in terms of computing resources. You get the most bang for your buck by using exactly one word for things like this; there is no benefit to using less than a word, and things will slow down dramatically if you use more than one.

[Disgruntled Head of C-3PO]

Originally posted by iboy:
When will the MacOS be an 64-bit operating system? In 1 year, 2 years...????

gr�essli

yob

Why do you need it?

[Powaqqatsi]

Originally posted by Millennium:
64 bits will get you some 18 quintillion numbers -9 quintillion each way if you want negatives

How many zeros is that ?

[Link]

Originally posted by Powaqqatsi:
How many zeros is that ?

Quint

x,000,000,000,000,000.00

Tril

x,000,000,000,000.00

Bil

x,000,000,000.00

Mil

x,000,000.00

Thousand

x,000.00

Hundred

000.00

[Tsilou B.]

Originally posted by Link:
Quint

x,000,000,000,000,000.00

Tril

x,000,000,000,000.00

Bil

x,000,000,000.00

Mil

x,000,000.00

Thousand

x,000.00

Hundred

000.00

Sorry, that's not correct.

You forgot quadrillion, so quintillion is:

x,000,000,000,000,000,000.00

[Detrius]

Originally posted by Millennium:
Actually, this would only be true on a 32-bit chip, such as the G4.

What makes a chip 32-bit, or 64-bit, or whatever, is how long a word is. You may know that eight bits is a byte; a word is a collection of bytes of a specified length. On a 16-bit computer, a word is 16 bits. On a 32-bit computer, it is 32 bits. And on a 64-bit computer, a word is 64 bits long.

etc...

There are many different definitions for what makes a processor 32 bit or 64 bit. The generally accepted method is the largest size of an integer. It's possible to have a 32 bit processor whose system bus is 64 bits wide. If I remember correctly, the Pentium and every processor after it had at least a 64 bit wide system bus. This bus does NOT have to match the size of the integer registers. In fact, virtually none of the busses have to match the size of the integer registers. In many cases, it would be a good idea, but there are always other possible methods. 64 bits is still twice as mutch data as 32 bits. In large quantities, regardless of the size of the bus, 64 bits takes twice as much time.

[King Bob On The Cob]

Wouldn't the bus on the G5 be 128 bits wide because Altivec handles 128 bit operations?

[Graymalkin]

No.

[Angus_D]

Originally posted by King Bob On The Cob:
Wouldn't the bus on the G5 be 128 bits wide because Altivec handles 128 bit operations?

http://developer.apple.com/technotes/tn/tn2087.html

[Eriamjh]

Can you point out exactly what part of OSX needs to be 64-bit?

Remember that bits are really word size. The more bits, the higher the numbers. What part of OSX needs to reference numbers greater than 2^32 (that's over 4 billion to you and me)?

I think the whole 32/64 bits thing is more misunderstood than the MHz thing.

The only thing OSX needs to do is allow 64 bit apps to function and it already does that.

And according to that Apple Developer's note, the G5 has only a 42-bit address bus and can only access up to 16GB of RAM. So even though the G5 can only access 8GB right now, there isn't much higher it can go without moving to the G6.

[power142]

Originally posted by Eriamjh:
And according to that Apple Developer's note, the G5 has only a 42-bit address bus and can only access up to 16GB of RAM. So even though the G5 can only access 8GB right now, there isn't much higher it can go without moving to the G6.

OK - so here's a stupid question - how difficult is it for Apple to tweak Panther to support processes being able to allocate greater than 4GB memory? The way I understand it is that this capability would be handled at the kernel level, but would require application recompiles to take advantage of it.

In practice, few applications would benefit from being able to do this, mainly those scientific in nature, databases, possibly graphics. With this in mind, most of the open source apps I might like to use with a memory pool larger than 4GB are already coded to deal with 64-bit address spaces because of their historic use on machines such as SGI, Alpha, etc.

Does anyone know why Apple wouldn't provide this type of kernel tweak, even if it was only in the OSX server or a special cluster or grid edition?

[rogerkylin]

I am most interested in the ability to allocate > 4 Gb memory. We (work) are going to be purchasing more computing nodes for our cluster. We currently have dual 2.8 and dual 3.0 GHz Xeons.

One program typically uses 1 GB+ (the sky is the limit) of memory. I've done some testing and for 1.3 GB test cases, the 2 GHz G5 is 30% faster than the 3 GHz Xeon.

If the rumors are true about a G5 XServe, and 2.6 GHz G5's, the memory addressability is the only thing holding the G5 back.

(to be fair, I have done no testing on the Opteron which tends to be more expensive, but for which a 64-bit OS already exists).