| Author |
Message |
Guest
|
 Posted:
Thu Jan 27, 2005 3:14 am Post subject:
Benefit or microprocessor when change from 32-bit to 64-bit |
|
|
I want to know what is the real benefit of changing from 32-bit to
64-bit computing. Please help...
Plus, if you can answer it more general, it's better, for example, the
changing from 8-bit to 16-bit, is it have the same benefits???
Thank you. |
|
| Back to top |
|
 |
glen herrmannsfeldt
Guest
|
| Posted:
Thu Jan 27, 2005 1:33 pm Post subject:
Re: Benefit or microprocessor when change from 32-bit to 64- |
|
|
frzzman@gmail.com wrote:
| Quote: | I want to know what is the real benefit of changing from 32-bit to
64-bit computing. Please help...
Plus, if you can answer it more general, it's better, for example, the
changing from 8-bit to 16-bit, is it have the same benefits???
|
An excuse to convince your boss that you need a new computer
when you really don't.
Unless you run programs that need more than 3GB (plus a little
for the OS) of memory, you don't need one.
-- glen |
|
| Back to top |
|
|
Stefan Monnier
Guest
|
| Posted:
Thu Jan 27, 2005 6:36 pm Post subject:
Re: Benefit or microprocessor when change from 32-bit to 64- |
|
|
| Quote: | I want to know what is the real benefit of changing from 32-bit to
64-bit computing. Please help...
|
Not much, really. That made sense way back when we had so little memory
that bits counted, but nowadays, who'd even notice an extra 32 bits?
When I add memory, I add it at least a GB at a time.
Stefan "puzzled as well" |
|
| Back to top |
|
|
Nicholas O. Lindan
Guest
|
| Posted:
Thu Jan 27, 2005 9:16 pm Post subject:
Re: Benefit or microprocessor when change from 32-bit to 64- |
|
|
frzzman@gmail.com wrote:
| Quote: | I want to know what is the real benefit of changing from 32-bit to
64-bit computing. Please help...
|
The _real_ benefit? Oh, what's reality got to do with it?
There isn't any benefit. A scholarly paper on the lack of benefit would
be far more instructive rather than yet another piece of 'group think'.
It's a serial Vs parallel thing: At any given stage in technology and
for a given performance goal there will be a range of optimum architectures.
64+ bits were popular in the 60's. Univac and CDC are a bit passe these
days. It's bandwidth not busswidth that counts.
If you are making a systolic super-computer then 64 bits may be worth
something. If the application is a box to sit on your desk
or to run the dishwasher then 64 bits is a boondoggle.
| Quote: | it's better ... the changing from 8-bit to 16-bit [?]
|
No. See above.
--
Nicholas O. Lindan, Cleveland, Ohio
Consulting Engineer: Electronics; Informatics; Photonics.
To reply, remove spaces: n o lindan at ix . netcom . com
psst.. want to buy an f-stop timer? nolindan.com/da/fstop/ |
|
| Back to top |
|
|
Luke
Guest
|
| Posted:
Fri Jan 28, 2005 12:50 am Post subject:
Re: Benefit or microprocessor when change from 32-bit to 64- |
|
|
frzzman@gmail.com wrote:
| Quote: | I want to know what is the real benefit of changing from 32-bit to
64-bit computing. Please help...
Plus, if you can answer it more general, it's better, for example,
the
changing from 8-bit to 16-bit, is it have the same benefits???
Thank you. |
|
|
| Back to top |
|
|
Luke
Guest
|
| Posted:
Fri Jan 28, 2005 12:57 am Post subject:
Re: Benefit or microprocessor when change from 32-bit to 64- |
|
|
frzzman@gmail.com wrote:
| Quote: | I want to know what is the real benefit of changing from 32-bit to
64-bit computing. Please help...
Plus, if you can answer it more general, it's better, for example,
the
changing from 8-bit to 16-bit, is it have the same benefits???
Thank you.
|
A 64bit processor is twice as fast a a 32bit processor because it
handles twice the data "at once". But Operating Systems and programs
need to be made specifically for a 64bit processor to get maximum
benefits. Yes 16bit is twice 8bit so its twice as fast.
Lucas J Bennett |
|
| Back to top |
|
|
HP
Guest
|
| Posted:
Fri Jan 28, 2005 5:57 am Post subject:
Re: Benefit or microprocessor when change from 32-bit to 64- |
|
|
<frzzman@gmail.com> wrote in message news:1106777667.297511.232110@c13g2000cwb.googlegroups.com...
| Quote: | I want to know what is the real benefit of changing from 32-bit to
64-bit computing. Please help...
|
You can address more memory 2^64 bytes instead of 2^32
HP. |
|
| Back to top |
|
|
Bill Todd
Guest
|
| Posted:
Fri Jan 28, 2005 6:28 am Post subject:
Re: Benefit or microprocessor when change from 32-bit to 64- |
|
|
glen herrmannsfeldt wrote:
| Quote: | frzzman@gmail.com wrote:
I want to know what is the real benefit of changing from 32-bit to
64-bit computing. Please help...
Plus, if you can answer it more general, it's better, for example, the
changing from 8-bit to 16-bit, is it have the same benefits???
An excuse to convince your boss that you need a new computer
when you really don't.
Unless you run programs that need more than 3GB (plus a little
for the OS) of memory, you don't need one.
|
Well, with a couple of exceptions.
1. You appear to be referring to virtual memory above (having lumped
the OS and the application together in the total). While the Intel/AMD
'physical address extension' (PAE) facilities allow the OS to manage
more than 4 GB of physical memory reasonably effectively (at least for
applications that don't individually want access outside their normal
32-bit limits), across-the-board support for 64-bit operation (e.g., in
mapping component space such as PCI's to physical memory) is cleaner.
2. The 32-bit Windows NT/2K/XP variants have a somewhat brain-dead file
cache implementation that restricts the amount of *virtual* address
space available to the system file cache while simultaneously requiring
far more of it than the actual cached data consumes (at least when the
data is cached in small chunks). This restriction effectively
disappears in 64-bit Windows, allowing its file cache to be considerably
more effective in some cases.
And of course there are those programs which seem to feel they must
promiscuously map entire files directly into their address space to
operate on them rather than use conventional read/write file-access
semantics. But that falls under virtual address space issues, though
not necessarily the ones you may have had in mind.
One might argue that 16-bit environments *could* satisfy virtually every
real computing need, but that would ignore the very significant issue of
ease in creating applications (or for that matter system software) and,
to some degree, performance. Similar arguments apply to the 'need' for
64-bit computing: we just haven't (yet) pushed the limits of 32-bit
systems sufficiently for them to be anywhere nearly as persuasive
outside somewhat specialized cases.
- bill |
|
| Back to top |
|
|
glen herrmannsfeldt
Guest
|
| Posted:
Fri Jan 28, 2005 7:56 am Post subject:
Re: Benefit or microprocessor when change from 32-bit to 64- |
|
|
Bill Todd wrote:
| Quote: | glen herrmannsfeldt wrote:
|
(snip)
| Quote: | Unless you run programs that need more than 3GB (plus a little
for the OS) of memory, you don't need one.
Well, with a couple of exceptions.
|
It was meant to start a discussion, but I think it is
pretty close. The question sounds a little like homework, too.
| Quote: | 1. You appear to be referring to virtual memory above (having lumped
the OS and the application together in the total). While the Intel/AMD
'physical address extension' (PAE) facilities allow the OS to manage
more than 4 GB of physical memory reasonably effectively (at least for
applications that don't individually want access outside their normal
32-bit limits), across-the-board support for 64-bit operation (e.g., in
mapping component space such as PCI's to physical memory) is cleaner.
|
If they hadn't put a 32 bit MMU in IA32, there would be even less
need for 64 bits. Well, that and a good segment selector cache.
Multiple 4GB segments aren't that hard to deal with. Trying to
write code to keep large arrays in PAE is, though.
(snip)
| Quote: | One might argue that 16-bit environments *could* satisfy virtually every
real computing need, but that would ignore the very significant issue of
ease in creating applications (or for that matter system software) and,
to some degree, performance. Similar arguments apply to the 'need' for
64-bit computing: we just haven't (yet) pushed the limits of 32-bit
systems sufficiently for them to be anywhere nearly as persuasive
outside somewhat specialized cases.
|
Loading segment selectors is pretty slow on many machines. A fair
number of programs don't need to access more than 64K for a single data
object, and very few need more than 64K for a single subroutine or
function. If 32 bit processors were a lot more expensive, 16 bits
would have lasted longer.
Now, in Fry's ad today, an Athlon 64 2800+ with motherboard
for $180. At prices like that, it is hard to argue against it.
On the other hand, I have an W2K system with 3GB of swap, and recently
had a program fail with "This program needs more than 3MB of memory."
The problems with signed arithmetic and programmers that don't take that
into consideration when doing comparisons.
-- glen |
|
| Back to top |
|
|
Thomas Womack
Guest
|
| Posted:
Fri Jan 28, 2005 6:06 pm Post subject:
Re: Benefit or microprocessor when change from 32-bit to 64- |
|
|
In article <1106777667.297511.232110@c13g2000cwb.googlegroups.com>,
<frzzman@gmail.com> wrote:
| Quote: | I want to know what is the real benefit of changing from 32-bit to
64-bit computing. Please help...
|
mov rax, rcx
xor rdx, rdx
imul rbx
idiv rsi
does a 64 x 64 -> 64 modular multiply in 75 clock cycles on Opteron.
[please could I ask someone to time it on a Nocona; it's a figure I'd
be interested in and have not seen anywhere]
I've not coded up an efficient 128%64 using x86 registers; a
64*64->128 takes tens of cycles (damned slow shifter on P4), needs
more integer registers than IA32 provides (though PMULUDQ in SSE
registers helps somewhat), and writing it takes a fair amount of
hair. I'd be a bit surprised if I could get (64*64)%64 in under ten
times the number of cycles the Opteron takes.
Opteron/2400 GMPbench score is 5400 or so, Northwood/2800 is about
2200.
So if your work looks like multiplies with medium-sized moduli, or
like multi-precision arithmetic, going to 64-bit GPRs is a substantial
win because the arithmetic instructions do at least four times the
work. The expanded memory access is completely irrelevant; I'd not
bother with more than 512MB in an Athlon64 math server.
OK, I'd like PMULLUQQ and PMULHUQQ (top and bottom halves of
128*128->256 in SSE registers), but they're not compatible with
current SSE implementations. PMODUQQ and PDIVUQQ would also be
very pleasant.
I _think_ those instructions could fit better into Altivec
implementations, which can't have the two-64-bit-halves restrictions
because they'd interfere so badly with vperm. But Altivec is at
least one generation back from SSE for my uses (largest multiply
is 16x16->32 and there are no DP floating-point instructions), so
it'll be a while to wait.
My MiniMac should arrive Monday; I really hope to find some kind of
interesting use for the altivec.
Tom |
|
| Back to top |
|
|
Guest
|
| Posted:
Sat Jan 29, 2005 12:29 am Post subject:
Re: Benefit or microprocessor when change from 32-bit to 64- |
|
|
"I want to know what is the real benefit of changing from 32-bit to
64-bit computing."
The real benefit, is that you can write applications that deal with
arbitrarily large memory structures and never have to worry that 32-bit
indexes into array overflow/wrap, or that 32-bit pointers cannot
address. 64-bits does not, in and of itself, cause the computations to
go faster (since all modern machines use at least 64-bit data paths
throughout the memory hierarchy--sometimes wider)
One should/must not forget that 64-bit data items come with a
penalty--that is less data fits in a cache of a given size. So, one
should use 64-bit data items carefully for best application
performance.
The other real benefit, is that the operating system can vacate the
2GB-4GB space so that a 32-bit application can use all of its linearly
addressable address space. The OS can still reach down into the 32-bit
application by using 64-bit pointers/addressing.
The switch from 8-bit to 16-bit was necessary since so many things
simply need more than 8-bits of area to represent all the thinkgs
software wants to represent. The switch from 16-bits to 32-bits is
similar but of lesser necessity. The switch from 32-bits to 64-bits is
being driven by the larger scale multiprocessors (4P to 64P) in server
environments. These can often be characterized as: given twice as much
main memory, 15%-30% more performance can drop right to the bottom line
simply by being able to address all the memory one can afford to
install. The main memory is serving as a disk cache with similar
properties and disk-scale latencies and disk transfer line-sizes.
Not having 64-bits of linear or physical address space causes the
operating system to have to do a lot of work that is not required with
a larger flatter model. So, this work vanishes (or is signifacantly
reduced), leaving more processor cycles to applications.
Mitch |
|
| Back to top |
|
|
Del Cecchi
Guest
|
| Posted:
Sat Jan 29, 2005 12:50 am Post subject:
Re: Benefit or microprocessor when change from 32-bit to 64- |
|
|
frzzman@gmail.com wrote:
| Quote: | I want to know what is the real benefit of changing from 32-bit to
64-bit computing. Please help...
Plus, if you can answer it more general, it's better, for example, the
changing from 8-bit to 16-bit, is it have the same benefits???
Thank you.
A large advantage is that you can use a very large virtual address space |
and never reuse a block of space in virtual storage. And you can
allocate all disk storage by assigning a virtual address, rather than
goofy heirarchial names.
del cecchi |
|
| Back to top |
|
|
Terje Mathisen
Guest
|
| Posted:
Sat Jan 29, 2005 1:17 am Post subject:
Re: Benefit or microprocessor when change from 32-bit to 64- |
|
|
Thomas Womack wrote:
| Quote: | In article <1106777667.297511.232110@c13g2000cwb.googlegroups.com>,
frzzman@gmail.com> wrote:
I want to know what is the real benefit of changing from 32-bit to
64-bit computing. Please help...
mov rax, rcx
xor rdx, rdx
imul rbx
idiv rsi
does a 64 x 64 -> 64 modular multiply in 75 clock cycles on Opteron.
[please could I ask someone to time it on a Nocona; it's a figure I'd
be interested in and have not seen anywhere]
I've not coded up an efficient 128%64 using x86 registers; a
64*64->128 takes tens of cycles (damned slow shifter on P4), needs
more integer registers than IA32 provides (though PMULUDQ in SSE
registers helps somewhat), and writing it takes a fair amount of
hair. I'd be a bit surprised if I could get (64*64)%64 in under ten
times the number of cycles the Opteron takes.
|
One of the AES contestants was Decorrelated Fast Cipher (DFC), which
used a 64x64 mod (13+ 2^64) modular multiplication as the key method to
gain provable resistance to at least some forms of correlation attacks.
With handcoded PentiumPro assembler code we got it within a factor of
two (afair) of the speed on an Alpha, and up from being one fo the 3-4
slowest (of 15 candidates) ciphers, to one of the 3-4 fastest.
It did take 'some hair' though. :-)
Terje
--
- <Terje.Mathisen@hda.hydro.com>
"almost all programming can be viewed as an exercise in caching" |
|
| Back to top |
|
|
Anton Ertl
Guest
|
| Posted:
Sat Jan 29, 2005 2:09 am Post subject:
Re: Benefit or microprocessor when change from 32-bit to 64- |
|
|
MitchAlsup@aol.com writes:
| Quote: | "I want to know what is the real benefit of changing from 32-bit to
64-bit computing."
The real benefit, is that you can write applications that deal with
arbitrarily large memory structures and never have to worry that 32-bit
indexes into array overflow/wrap,
|
Well, of you are using Java or the dominant "64-bit" variant of C
(I32LP64), you certainly have to worry about that. For most people
these troubles are still a few years down the road, though.
- anton
--
M. Anton Ertl Some things have to be seen to be believed
anton@mips.complang.tuwien.ac.at Most things have to be believed to be seen
http://www.complang.tuwien.ac.at/anton/home.html |
|
| Back to top |
|
|
Nicholas O. Lindan
Guest
|
| Posted:
Sat Jan 29, 2005 6:06 am Post subject:
Re: Benefit or microprocessor when change from 32-bit to 64- |
|
|
<frzzman@gmail.com> wrote:
| Quote: | I want to know what is the real benefit of changing from 32-bit to
64-bit computing. Please help...
|
Calculating the Mandelbrot set at high magnifications.
--
Nicholas O. Lindan, Cleveland, Ohio
Consulting Engineer: Electronics; Informatics; Photonics.
To reply, remove spaces: n o lindan at ix . netcom . com
psst.. want to buy an f-stop timer? nolindan.com/da/fstop/ |
|
| Back to top |
|
|
|
|
|
|
FAQ
Memberlist
Register
Profile
Log in to check your private messages
Log in