Slashdot Mirror
Are We Not Ready For 64-Bit?
Q3vi1 writes "The Inquirer posted an intriguing article about how Intel doesn't think that we'll be ready for mainstream 64-bit computing until 2007. Coupled with the fact that MS isn't supporting the Opteron yet for their Windows 2003 Server, we may see a delay in consumer applications for 64-bit computing. However, as this article states, some people don't really care and will just go for Linux and AMD as a nice marriage."
I'd be interested to know how many operations on today's computers actually even use up all 32 bits available to them. I'd expect those situations to be rare: Matrix math operations, some addressing.
64 bit computing might speed up your data processing if you are a scientist, but it would probably slow down business applications.
In general, scientists that need high processing speeds can buy supercomputer time, or extra 32 bit machines. Why would we want to move to 64 bit on the desktop?
I just hope AMD realises that the platform should belong to the owner and keep Opteron/Athlon64 free of TCPA. This together with a Palladium free Linux would be the major reason for me to leave the comfort of the Wintel platform.
But I fear if AMD state they are remaining TCPA free they've got no chance of seeing a Palladium enabled Windows 200x on Opteron/Athlon64 - goodbye mass(ive) market.
$2B OR NOT $2B = $FF
As I have been MS free since 2000, I really couldn't care less what they do or don't do. As for Intel, here's some news for them, they DO NOT have a monopoly like their special friend. I'll gladly purchase an AMD Opteron to run my shiny new Linux 2.6 kernel sometime this fall while the WinTel boys play their reindeer games. In fact, dare I say it, I'm GLAD this is happening. Hopefully, this will finally show Intel that their future is not tied to MS as it was in the past.
Good people do not need laws to tell them to act responsibly, while bad people will find a way around the laws-Plato
Do I need 64 bits? No.
Do I want competition to the Xeon in 4 way systems? (price and spec them, it is insane! 1.6 Ghz and 1200 a pop). Hell^yeah.
Opteron is not about 64 bits, it is just a nice addition. Opteron is about competition in the low end server/high end desktop market (which is intel dominated btw). The reaosn intel is naysaying 64 bits is because they have no competing thec in this area other than the Xeon which has terrible price/performance numbers.
There is nothing wrong with being gay. It's getting caught where the trouble lies.
If they're cheap enough, 64bit will help gaming in a big way. The counter-strike team reported a ~30% increase in performance just by recompiling. Granted CS doesn't need a cray to run, but Battlefield 1942 has had some 64 player servers which I believe needed dual Athlons. 64 people is fun, but how bout 128?
Not only that, but with an (relatively)inexpensive 64bit chip out there I could see more servers popping up to play on. More servers hosting large games would be great! Feed my addiction please.
-- taking over the world, we are.
Your view of 'when are we ready for 64-bits' largely depends A) on how much money you are willing to spend on RAM and B) how soon your OS supports more than 4 GB of RAM on potential 64-bit hardware (PAE hacks notwithstanding).
If you're willing to spend $200 for RAM in your system, then when 4 GB of RAM is cheaper than $200, you'll basically be wanting a 64-bit system (PAE hacks notwithstanding).
With pricewatch.com showing 1 GB of PC133 SDRAM going for as little as $120, I'd guess that another 4x drop in RAM prices would lead to substantial consumer demand for 64-bit hardware.** And that doesn't even include the demand for 4+GB RAM now in database applications. Whatever the case, this would seem to be earlier than 2007. Unless Microsoft doesn't get its act together (they were pretty late with 32-bit 386 support, IIRC)... which wouldn't be such a bad thing, for Linux at least. But I wouldn't count on that.
--LP
** Yes yes, technically you probably need to spend a bit more to get higher density RAM so that you can fill or exceed 4 GB given the limited number of memory slots available in your system.
2GB, and that's shared across all the apps.
:-)
Linux can do better than that. In Linux you get 3GB per process.
Do you care about the security of your wireless mouse?
Who cares about 64 bits if the mainstream applications are written in a way that keeps wasting CPU power left and right without much care about peformance or efficient design ? Twice as large addressing space. Ok, so what ?
:)
The one of the only few areas where 64 bits will make an actual tangible difference is a crypto and OS themselves, but these would not probably be a factor enough to speed the introduction of 64 bits CPU.
I mean if the money Intel spent on R&D would've gone on to the (re-)eduction of applicaiton designers, we would've still be doing just fine with old'n'trusty Pentiums. Slightly exagerrated, of course, but you got the point
3.243F6A8885A308D313
Correct me if I'm wrong, but isn't the PowerPC/G4 based off of the POWER4 architecture... but with all the 64-bit (as well as several other "high-end" features) stripped out?
Anyhoo, I know 64-bit sounds nice and all... But my question is: will moving from 64-bit integers, floats, and memory addresses actually improve performance? To my understanding of processor architecture, the answer is definitely 'maybe.'
If the 'new' processor just adds 64-bit extensions, and doesn't actually optimize further than the previous generation, then your 'new' 64-bit chip is now handling data that is twice as large... and probably doesn't need to be twice as large. I mean, really... how often does any of our software make use of 64-bit integers or floats? How often do you need to break the 4GB memory barrier? Sure, years down the road these limitations (especially the memory) will be problems... but for right now, why get all hyped over 64-bit when it will do nothing but double the size of everything?
Personally, for the time being, I'd much rather see highly optimized and blazing fast 32-bit processors on the market. Leave the 64-bit for when I actually need over 4 gigs of memory.
Come on guys.... you know I'm right. Unless you're in the scientific fields that use huge numbers and insane gobs of memory, 64-bit data paths will be wasted cpu-real-estate.
/dev/random
It's not so much if we're ready, it's which 64-bit chip are we ready for.
I believe AMD's chip is the best for the home user, Itanium will be too power hungry. You'll never see an Itanium notebook in my opinion as the design isn't a real world solution.
The masses using a PC as an entertainment hub in the living room will only happen when PCs are nearly silent, But the way we're going they never will be.
Ummm.... the very fact that we're talking about 64-bit computing means that a hardware update is necessary, and Intel's attempts at 64-bit processors have so far not been backwards compatible, or do so only at a steep performance price. Meanwhile, Apple's best bet at 64-bit goodness(the PPC 970), will run 32-bit code natively.
As 64-bit becomes more of an issue, Apple's investment in the far-sighted PowerPC will really begin to yield significant benefits.
Front Side bus mainly applies to memory hog applications (which of course the 64bit issue can applies to this as well).. But when we look at speedups for games such as counter-strike, what we see is that the speedup really comes from adding an extra set of registers.
What the x86 line needs is not more memory, cleaner instruction set, or better memory management, but more registers.
I'm somewhat upset that AMD didn't make it's x86-64 use 32 registers (the defacto standard these days). Morever, given that Intel added a ne w operating mode for use with SSE, I don't see why AMD couldn't have added one that merely provided the 32 registers for x86-32. They would have a HUGE jump on Intel with hardly any additional cost. Most compilers implement algorithms that are register-set-size nuetral, so it should not be a dramatic rewrite for anybody's compiler. Granted, you muck-up most of the op-codes, requiring more bits per register, but that's what the new mode is for.
Going to 64bit, otoh is an enormous change.. Not to mention, I don't believe AMD gives you the option of using the extra register without requiring 8 bytes per pointer (thereby losing the space-efficiency of 32bit coding).
While I'm happy to see any progress, I'm disappointed that the root-canal type changes that will be necessary to support x86-64 aren't going to bring the PC up to even 1990's technology standards.
-Michael
The thing AMD needs to do is put 16 or 32 DIMM slots in the motherboard for their 64-bit processor. As many others have pointed out, RAM is dirt cheap for up to 1GB DIMMs. I could buy a 64-bit processor and motherboard plus 32GB of RAM for a reasonable sum.
That's 32 gigabytes. Just the disk caching speedups alone would be worthwhile. My firm belief is the only reason these huge RAM sizes aren't common is the 4GB physical / 3GB per process limits of current 32 bit OSs.
Garbage Collection
"Real Languages" use garbage collection (ha, just trolling).
Seriously, the ability to use an address space that is gignormous is really worth a lot for garbage collection algorithms. For example, you can allocate into reserved portions of the address space and then the type of an object can be determined by its location. You can also use copying collectors without a big hit. Reserving half your address space for copying sucks at 2GB, it doesn't matter much for 17179869184 GB.
Also the "single address space" operating system concept needs more research. However, to get that research going now would require low cost plentiful hardware.
The fact is, there are tons of useful reasons to have 64 bits, we just don't know what they are because we haven't had 64 bits on a commodity platform.
If you have 64 bit addresses and about 1GB of flash RAM, you can completely avoid all the trouble of traditional filesystems. Have your OS use the disk like one big area of RAM, buffer into the NV RAM, keep all the metadata in NV ram, and use a journaled approach for metadata. Speed and simplicity instead of B-trees and inodes and such.
There are all kinds of reasons for 64 bit.
((lambda (x) (x x)) (lambda (x) (x x))) http://www.endpointcomputing.com a scientific approach to custom computing.
Convenient that Intel thinks widespread demand for 64-bit won't occur until 2007 since they don't have a 64-bit desktop processor and Itanium tanked. AMD knows better. Check out their Studio64 which has quotes from tech leaders, analysts, press, etc. on when 64-bit will hit big and what this means for you and me: http://www.amd.com/us-en/Weblets/0,,7832_8366_7823 ,00.html
While most people don't need over 4GB of RAM, having 64 bits can make life much easier for programmers and provide significant performance advantages. For example, no more relocation will need to occur for shared libraries. Every library could be mapped to a unique address without worry of address clashes so no relocation is necessary (although one of the benefits of the Opteron is better support for relocatable code in 64-bit mode).
Memory mapped files could be the norm. Handling large files becomes much simpler, especially random access.
-Aaron
This post is encrypted twice with ROT-13. Documenting or attempting to crack this encryption is illegal.
Well, if we really want to pay attention to history, there is NOT a lot of room for RAM to grow. There is only a 4 - 16X growth capacity here. Just a few short years ago, 32M was a lot of memory, and was the typical amount sold on a new PC. I'd hate to see some funky expanded memory crap like we had back in the 640K barrier days. The days of 4G machines are NOT far off.
Note that most standard PC's can't handle the full 4G anyway due to video and other expansion cards snarfing larger and larger chunks of the address space.
We can learn a lot from the IDE folks about how to NOT anticipate the future as year after year we kept slamming into the limits of the spec-of-the-day causing all sorts of problems.
A larger problem than memory is PCI bus bandwidth. Before 64 bit processors can really shine, we need a better bus. Hell, the current generation of PCI can't even handle today's 32 bit processors well, especially in SMP boxen.
Until the end of 1995.
To be fair, Intel was already there with 32-bit chips, just M$ had to change all its 16-bit code and update its OS. This time it looks like the positions are reversed (unless Itanium ever takes off).