64-bit or 32-bit: When will it matter?

[name]

Look here.

http://www.microsoft.com/whdc/system/platform/server/PAE/pae_os.mspx

 

[David Schwartz]

David Schwartz,
Can you answer a few questions please?
Sure.

1. Just so I know, what exactly is PAE? (yes, I know what it stands for)

PAE allows an x86 processor to access more physical memory than it can
address with 32-bits. Specifically, it allows 36-bit *physical* address.

2. What does it do?

It basically adds an extra 4-bits in the page table mappings.

3. Why is it a bad workaround?

Because it doesn't increase the address space seen by a process. Because
it's a workaround rather than a real solution.

4. How much RAM can Win 9x, Win NT, 2K and XP, Mac OS 8, 9 and X, and
other *nix (assuming most popular distros) handle without using this "PAE
workaround"?

I'm asking these because there are quite a few things I'd never heard of
in Brendan's post.

Typically about 3Gb.

If you're looking for more specific information, use a search engine. It
would take many dozens of pages to answer your questions in detail because
they're not very specific.

DS

 

[No One]

It does seem EVERYTIME there is a shift in the CPU for PCs, we have this
same discussion. Only servers need it. Only workstations need it.
Only hard core gamers need it. Only power users need it. Only...oh,
its required. I remember the shift from the 8088 to the 8086..who needs
16 "real" bits??? 8086 to 80286....20 bit memory access?? what do I
need with 16 MB of RAM??? then the 80386...who needs 32 bit registers,
we still run DOS....and so on....

 

[Brendan Trotter]

Hi,

PAE is an ugly workaround. A lot of people specifically purchase the
maximum amount of memory their OS can handle without PAE because they want
to avoid it.

I found PAE to be a fairly elegant extension, except for CR3 (the page
directory pointer table address) being restricted to 32 bits instead of
being extended to 36 bits like the rest of the paging system - a minor
hassle for an OS's memory manager considering that it needs to
differentiate between 20 bit physical address (for ISA DMA), 32 bit
physical addresses (for 32 bit PCI devices) and 36 bit physical addresses
anyway.

The only other relevant issue is the overhead of PAE paging structures,
which consumes roughly twice the amount of memory as an equivelent
32 bit paging system when operating on 4 KB pages, and roughly 4 times
the memory when operating with PSE (4 MB or 2 MB page sizes).

Please note that the lower level paging structures (page directories and
page tables) are identical for both 64 bit paging and PAE. To handle the
higher 32 bits of a 64 bit linear address the PDPT (page directory pointer
table) was extended from 32 entries to 512 entries, and a new top level
table was added (the PML4E). Considering that the only real differences
were needed to support 64 bit linear addressing, and that AMD left the
remainder unchanged, I'd say Intel's PAE "work-around" was quite good.

I can understand people trying to avoid installing 4 GB or more of RAM,
but I'd suggest this has more to do with things like 32 bit PCI devices
using bus mastering in conjunction with buggy/old device drivers rather
than PAE itself (ie. any mechanism that supported physical addresses
larger than 32 bits would've caused similar problems regardless of how
good it is). This wouldn't apply to a 64 bit OS as the device drivers
all need to be updated/rewritten (and hopefully tested on computers
with > 4 GB of memory).


Cheers,

Brendan

 

[Cool_X]

Thanks name,
I wouldn't have had time to fix that on my own.

Cool_X

 

[Cool_X]

I meant, I wouldn't have had time to FIND that on my own.

Sorry for the typo that I didn't catch.

Cool_X

 

[David Schwartz]

Oh there were, but they were painful to use in most cases. To address your
main point, it depends on your definition of commodity software, but by
any definition I don't see that as a "killer app" justifying moving from
32 to 64 bit hardware before the old system is due for replacement.
Actually I would consider that over half of the computers in desktop use
are going to be replaced in 6-7 years, with nothing more than attrition
driving it.

Exactly. People will wind up with 64-bit capable hardware without
specifically intending to have it just through attritition. Once that
happens, software will start to be released either as 64-bit only or with
significant benefits on 64-bit platforms.

I am. Based on two different justifications. The best is that there hasn't
been a 64 bit killer app for the Mac, and that's been 64 bit for a decade.
The other is that there *is* a point when people have enough and are not
willing to make an upgrade because they don't see the need.

I disagree with both points. On the first point, the 64-bitness of Macs
is not comparable to the 64-bitness of PCs for two reasons. One is that
64-bits on PCs is accompanied by other changes such as register size. The
other is that memory has now reached the point where a 32-bit limitation of
virtual memory size is significant.

As for your second point, people have been arguing that for decades and
it has never been proven right. I personally don't believe it -- people will
always want to do more and will always push their tools to the limit to
increase what they themselves can do.

If you can get access to sales info, most 32 bit systems aren't ordered
with max memory, largest disk, or fastest CPU. That certainly suggests
that people don't feel the need.

No, that's not the reason. It's because people buy for the sweet spot.
That is, the buy equipment that gives them the most bang for their buck. The
same goes for software requirements -- you can make better software if you
make the requirements greater, but you can't aim so high that no market is
left. The combination of these two forces makes 64-bit only software in six
years almost inevitable.

I'm still confident that 64 bit hardware will come driven by replacement
rather than upgrade.

I don't understand the difference between replacement and upgrade.
Perhaps you could explain. Aren't these the same things?

I agree that most computers will include 64 bit capability, but only
because it will be standard. Intel and AMD are unlikely to spend any money
in 32 bit only products, when they need more performance and lower power
foar more.

Well that's the point. As soon as the vast majority of power users are
64-bit capable, power user software will start to be released as 64-bit
only.

I predict the big 64 bit software push will be driven by greed, I mean
marketing. When enough people have gotten 64 bit hardware, Microsoft will
suddenly release new versions of all apps, with new features, and in 64
bit only. I predict they will offer *very* cheap upgrade from 32 bit
versions, because they know they will make the money on Windows-64 o/s
upgrades. But until most people have the hardware they won't push 64 bit
only, because it locks them out of a majority of the market.
Whatever.

As for gamers? I define a gamer as someone who spends at least $100 extra
on a computer for memory, faster CPU, or detter display. Oddly, that lets
out a fair percentage of people who do little else with their computer. If
they didn't spend money on hardware at 32 bits, will they jump to 64
intesad of spending the money on more games? For that matter, are the
games on the 64 bit Mac better? (real question, I have no idea)

This brings up the other flaw in your Mac example. Until a large
percentage of systems are 64-bit, there's no reason to develop software that
benefits from 64-bits.

DS

 

[Nate Edel]

I remember the shift from the 8088 to the 8086..who needs 16 "real"
bits???

There was no "switch from the 8088 to the 8086" - very few manufacturers
ever used the 8086.

8086 to 80286....20 bit memory access?? what do I need with 16 MB of
RAM??? then the 80386...who needs 32 bit registers, we still run
DOS....and so on....

And yet both of these were enough faster than the then-available models of
the prior generation processor that people pretty much jumped at buying them
if they could afford it.

The improvements with the x86 64-bit systems aren't quite so dramatic, but
they're quite significant at least on the server side: you'd be daft to buy
a pre-Nocona Xeon-based or an Athlon MP-based server, just because Nocona
and Opteron for reasons entirely unrelated to the 64-bit-ness offer very
siginficant performance advantages over their past generations.

It's not clear to me that the same is true for the Intel 64-bit Pentium 4s,
but it also costs basically nothing to get it.

Of course, you're right... memory needs increase pretty much at a pace with
the increase of memory capacities and the decrease in memory costs... we're
only a drop or two in price away from 2gb+ on the desktop being pretty
usual, at which point 64-bit processors get a lot more attractive.

 

[Bill Davidsen]

Exactly. People will wind up with 64-bit capable hardware without
specifically intending to have it just through attritition. Once that
happens, software will start to be released either as 64-bit only or with
significant benefits on 64-bit platforms.




I disagree with both points. On the first point, the 64-bitness of Macs
is not comparable to the 64-bitness of PCs for two reasons. One is that
64-bits on PCs is accompanied by other changes such as register size. The
other is that memory has now reached the point where a 32-bit limitation of
virtual memory size is significant.

As for your second point, people have been arguing that for decades and
it has never been proven right. I personally don't believe it -- people will
always want to do more and will always push their tools to the limit to
increase what they themselves can do.

Let's see, in the 60's car manufacturers built larger and larger
engines, until around 427-450 cubic inches very few people were
interested. Looks like people didn't buy more than they needed.

And Ford decided that there was a market for an SUV sized between an
Expedition and a school bus. They stopped making it for the model year
after three months or so.

No, that's not the reason. It's because people buy for the sweet spot.

Exactly! They buy what they need and a little more.

That is, the buy equipment that gives them the most bang for their buck. The
same goes for software requirements -- you can make better software if you
make the requirements greater, but you can't aim so high that no market is
left. The combination of these two forces makes 64-bit only software in six
years almost inevitable.

Vendors aim for the sweet spot too, features cost to develop and
maintain, so you don't see an unlimited number of features.

I don't understand the difference between replacement and upgrade.
Perhaps you could explain. Aren't these the same things?

If my computer (car, lawn mower, stove, tires, whatever) is near the
MTBF, is getting unreliable, making funny noices, then I get a new one.
That's replacement. And in business that means the cost is depreciated.

If I see a new computer (car, spouse, camera) which is just so much
*neater* than what I have, then I get a new one before the old one has
been fully utilized. Or depreciated. That's upgrade.

If there's a feature I actually need, it's still upgrade, but has a much
different rationale.

My bet is that most personal computers will be replaced as they get
older. I find it really unlikely that any company which provides less
than the fastest CPU and largest memory will be doing an upgrade, sexy
isn't deductable, and few applications go from small to huge in the
lifetime of a computer.

Well that's the point. As soon as the vast majority of power users are
64-bit capable, power user software will start to be released as 64-bit
only.

That's what I said, eventually. As soon as the market for software
running on Win98 dries up no one will make it... but they do today, so
what does that tell you about residual market. Mass market applications
are going to be out in 32 bits for years to come.

This brings up the other flaw in your Mac example. Until a large
percentage of systems are 64-bit, there's no reason to develop software that
benefits from 64-bits.

All the big game systems have been 64 bit for a while, seems to me I've
seen just one or two (hundred) new title for the older 32 bit versions.

 

[Cool_X]

Bill,
I agree with most of your post, and you definitely seem to know what you're talking about, but
with all due respect, I disagree with two things you said:

1. "> If I see a new computer (car, spouse, camera) which is just so much

*neater* than what I have, then I get a new one before the old one has
been fully utilized. Or depreciated. That's upgrade."

I object b/c I believe that "upgrade" means making any improvement to any existing unit, and
"replacement" means buying another separate unit.

So in your statement, I disagree with you saying upgrade" b/c you didn't make any improvement
to the old PC.

You also said:
"If my computer (car, lawn mower, stove, tires, whatever) is near the MTBF, is getting
unreliable, making funny noises, then I get a new one. That's replacement."

IMHO, getting a new lawnmower and getting a new PC are both replacement, for the reasons stated
above.

2. You also said "All the big game systems have been 64 bit for a while, seems to me I've seen
just one or two (hundred) new title for the older 32 bit versions."

a. Assuming you're not talking about game console machines (like Gamecube), how can "all the
big game systems" have been "64 bit for a while"??? Dell sells a lot of big gaming systems
(eg. their Dimension XPS, which has become famous, AFAIK), and I don't think a single one of
them has had a 64-bit CPU, b/c Dell doesn't use AMD at all.

b. When you said "one or two (hundred) new title", you were meaning that as a small number, right?

c. And where did you get your info from to make the quote that I copied in the beginning of
this question?

Please let me know about this.

Cool_X

 

[David Schwartz]

David Schwartz wrote:

Let's see, in the 60's car manufacturers built larger and larger engines,
until around 427-450 cubic inches very few people were interested. Looks
like people didn't buy more than they needed.

This is not a response. The car market is too different from the
computer market for there to be any reason to expect one to do what the
other has done. I'm sure we're all familiar with the joke about what cars
would be like if the markets were similar.

And Ford decided that there was a market for an SUV sized between an
Expedition and a school bus. They stopped making it for the model year
after three months or so.

There's just no comparison.

Exactly! They buy what they need and a little more.

No, they don't buy what they need. They buy the sweet spot. The sweet
spot is purely a price/performance issue and has nothing to do with what
anyone needs.

Vendors aim for the sweet spot too, features cost to develop and maintain,
so you don't see an unlimited number of features.

Exactly. And as the sweet spot moves up, vendors will aim for higher and
higher targets regardless of what anyone needs.

If my computer (car, lawn mower, stove, tires, whatever) is near the MTBF,
is getting unreliable, making funny noices, then I get a new one. That's
replacement. And in business that means the cost is depreciated.

If I see a new computer (car, spouse, camera) which is just so much
*neater* than what I have, then I get a new one before the old one has
been fully utilized. Or depreciated. That's upgrade.

If there's a feature I actually need, it's still upgrade, but has a much
different rationale.

Or, quite commonly, you need another computer. So you buy the latest and
greatest, and give your computer to the next person down the line in your
family.

My bet is that most personal computers will be replaced as they get older.
I find it really unlikely that any company which provides less than the
fastest CPU and largest memory will be doing an upgrade, sexy isn't
deductable, and few applications go from small to huge in the lifetime of
a computer.

Three years is the typical lifetime. It's often driven by hard drive
failure. Most people have no backups, and once you have to reinstall
everything anyway, you might as well have better performance and more
current applications.

That's what I said, eventually. As soon as the market for software running
on Win98 dries up no one will make it... but they do today, so what does
that tell you about residual market. Mass market applications are going to
be out in 32 bits for years to come.

The difference is that it's not too terribly hard to make software that
runs on Win98 and WinXP and still gets all the key advantages of XP. Try to
use an iPod on 98.

All the big game systems have been 64 bit for a while, seems to me I've
seen just one or two (hundred) new title for the older 32 bit versions.

Not a reasonable analogy for two reasons. First, the increase in 64-bit
machines has not correlated with a decrease in 32-bit machines. Second, I
never said there would be no new 32-bit software, just that there would be
more and more 64-bit only software -- so if the analogy were valid, it would
support my point

DS