Slashdot Mirror
Introduction to 64-bit Computing and x86-64
James writes "Ars Technica has an article up explaining 64-bit computing from the x86 angle, specifically from the angle of AMD's Hammer. The article explains the details in that usual Ars style, and I found it very useful for thinking about what kinds of applications I may want to put to the test on one of these when we get a box in the office. Even non-x86 freaks may appreciate this, since it breaks down some of the basic advantages of 64-bit computing, and just who can expect to see gains in the near future."
While IA-64 isn't as bad as many here like to paint it, I'm looking forward to x86-64, if for no other reason than the continued pressure and competition for Intel.
We've all heard the rumors that Intel has a 'secret' project to produce a P4 that executes x86-64. I have a feeling that they might have to unveil it, if only for marketing reasons. Wouldn't it be ironic, Intel adopting AMD's extensions to the x86 instruction set!
After reading the topic article (of which I saw on HardOCP yesterday), I was left with one question. Can you mix 32 bit and 64 bit code inside of ONE process? I know the OS can support 32 bit and 64 bit processes under Long mode, but can you mix code in a single process? It would be nice to know if you could build an application, query the CPUID feature flags, and then produce optimized code paths for both 32 bit and 64 bit. This would be nice because you can just supply one set of binaries to your customers instead of two seperate binaries. Anyone have any experience with the Opteron who can answer this?
The main perk of popularization of 64 bits is, if we think of the future, that more source code will be written with 64 bit capability in mind. Long ints will be used where they seem to be needed, etc. Open source movement is probably the most obvious benefactor, considering how harnessing that extra mile is/will be a trivial matter of recompiling. If we get a fast, cheap, non-x86 64 bit system in the future, all the better. But 64 bit capability should definitely be available on the desktops of the masses also, whether they needed 10 gig databases or not! More power for AMD!
I'm looking forward to industrial strength, rock solid Opteron servers to finally put to rest all the speculation how Linux on x86 machines is not up to all the tasks of Solaris/HP-SUX systems...
Save your wrists today - switch to Dvorak
But what will be really interesting is when operating systems will be specifically designed for this. With 64bit, a lot of the cruft in 32bit operating systems can finally go away. For example, memory mapped I/O finally becomes useful again, shared libraries don't clash as easily, etc.
a small point, but the author is wrong about underflow. from the review:
you get a situation called either overflow (i.e. the result was greater than the highest positive integer) or underflow (i.e. the result was less than the largest negative integer)
underflow is when you're trying to represent a fractional number smaller than the smallest floating point number available. ie: you went too close to zero.
Do you know what you're talking about? By math speak most addresses are just integers, but in computers a (void *) can be cast to a unsigned long without any loss of data, but thats not the point. x86-64 can only access 40bits of physical/virtual address space, even though internal datapaths can handle 64bit. He says that the biggest improvement will be large math and double the registers. As for x86, it's always been a hack. Im not a 16bit x86 buff, but it used switchable banks of memory. At no point could you address more than 64k in a segment. In the pentium, Intel added PSE (page size extensions) with allow 36bits total memory, but only a 32bit address space. Thats why you can configure a x86 linux kernel to handle 64gigs of High Memory. This is useless for any kind of IO address space, and still only 2^32 can be seen by any process. In the Pentium Pro they added PAE (physical address extention) which again lead to 36-bit addressing, but if your EIP register is only 32 bits long, how do you represent code in 0xfffffffff? Extended addressing on the x86 has always been a hack. Finally with x86-64 programs will be able to see more than 4gigs (minus what the OS steals. Win2kPro takes 2gigs, but Win2kServer and Advanced Server can take 1). Even alphas only have a 43bit phy/vir address space. When the time comes when it's cheap to have a terabyte of ram, Im sure they will increase the address space. This time it's not a hack though
If you'd actually READ the article (and understood it) you'd see the author was differentiating between the adressable range and the number of bits that could be stored in an address. Obviously the old 8 bit CPUs used to combine registers (eg, the Z80 had the IX and IY if I remember correctly - probably not as it was so long ago!) - if they didn't they'd have 256 bytes of RAM! The old segmented address space of the x86 was just bizarre, due to the way the segment registers were used. The segment register was 16 bit, even though you only needed the top 4 bits, so you could access the same address space in thousands of ways, but were confined to a "window" of 64K in any segment ;-)
Code, Hardware, stuff like that.
They forget to mention that even if you can have more than 4gb on xeon machine you cannot address any single block of more than 4gb.
True
Forget about putting your oracle db into memory.
Not true
On a Xeon machine, Oracle will let you put up to 62GB in memory. The trick is an operating system call that fiddles the page table and "swaps" pages from areas you can't see to areas inside your address space. It works well for applications that aer conscious of 4K pages and not page thrashing. Databases are the prime example, executing a few dozen privileged instructions and changing the page tables is much faster than going to disk.
Here's a description of how to do it on Linux:
For Windows, it's called Address Windowing Extensions (AWE).
I think MySQL supports these tricks too
Alan.
To summarize, Microsoft might run into a chicken-and-egg problem on 64-Bits: Nobody runs Windows on 64-Bits because it's not faster -> Nobody makes Win64 software -> Nobody runs Windows on 64-Bits.
Add into that the fact that Microsoft is traditionally very incompetent and slow when it comes to adopting new architectures and you get the idea.
I think that Microsoft will lose the majority of their server marketshare and a large chunk of their desktop marketshare during that transition. Simple market inertia will prevent Microsoft to be thrown out of the desktop market, but because of the 64-Bit transition, the Linux desktop market might finally gain critical mass and endanger the Windows domination in the long term.
I'm not an engineer.
What I got from this article (well written, BTW) is that 64-bit processors provide more processing power--a concept different from speed. For instance, a Porsche can fly down the highway, but its engine has insufficent power to tow a 5th wheel RV. A Mack truck isn't speedy, but has a strong engine that can tow the heaviest loads up the same speed--more work is performed.
That would be why Apple's PowerPCs are still in the running, despite their clock slowness. They are falling behind fast as, after a point, speed does matter, in combination with improved processor power in the latest 32-bit Pentiums, and certainly the Xeon. Vector processing, such as Altivec, helps keep Apple competitive with their current chips--just barely. Many in the PC community await AMD's offering while IBM works on blending its POWER chip technology into a 64-bit PowerPC, with Altivec.
Imagine making the Porsche's engine more powerful but maintaining its speed advantage so it can haul ass as well as tow. Add nitro for extra ooompf (vector processing) and you have a dream machine. That's why it seems that 64-bit apps and processors seem to be such a holy grail.
That's my take on it. Clarification is always appreciated.
Vos teneo officium eram periculosus ut vos recipero is.
"Should Apple move from 32-bit PPC to 64-bit PPC, Mac users should not expect the same kinds of performance gains that x86 software sees from the jump to x86-64. 64-bit PPC gives you larger integers and more memory, and that's about it. There are no extra registers, no cleaned up addressing scheme, etc."
Well yea, they already have more then 0 general purpose registers, and a flat memory space, like every other chip besides x86 has had for the last 20 years now. The embedded chips I've used on projects that cost $7 a piece even have more registers then x86 does.
64bit is about the memory, and using that memory, plain and simple. x86 just happens to be using this as a chance to catch up with the cutting edge concepts of the 1980s.
As for x86 needs to die once and for all, it's hacked, hacked again, and hacked yet again. x86 was and is a 16bit system. And now AMD wants to hack it yet again. Can anyone doubt that 80% of the silicon is for supporting legacy apps at this point? Are people that damn lazy they can't type 'make' on a new system? It's not like anyone uses "int" anymore and assumes it's N bits long.
- Adam L. Beberg - The Cosm Project - http://www.mithral.com/