64 bit os

Discussion in 'OT Technology' started by CaseGreen, Aug 21, 2007.

  1. CaseGreen

    CaseGreen ...watching people die can make you feel so alive. OT Supporter

    Joined:
    Jan 18, 2003
    Messages:
    3,213
    Likes Received:
    0
    Location:
    Houston, TX
    What are the advantages of using 64 bit xp/vista over 32 bit version? is there an improvement in preformance. What i understand is it lets you use alot more ram. :dunno:
     
  2. dorkultra

    dorkultra OT's resident crohns dude OT Supporter

    Joined:
    Oct 14, 2005
    Messages:
    22,753
    Likes Received:
    29
    Location:
    yinzer / nilbog, trollhio
    you won't notice an improvement unless the software you use is meant to utilize 64bit
     
  3. DAN513

    DAN513 OT Supporter

    Joined:
    Mar 10, 2003
    Messages:
    10,089
    Likes Received:
    2
    Location:
    204
    it will break the 3.5gb barrier, but the actual benefit for most people is limited.
     
  4. lowfat

    lowfat 24/Mac/SciFi/PC Crew OT Supporter

    Joined:
    Oct 27, 2003
    Messages:
    63,949
    Likes Received:
    0
    Location:
    Grande Prairie, AB, Can
    :werd: 64-bit does nothing for performance.
     
  5. P07r0457

    P07r0457 New Member

    Joined:
    Sep 20, 2004
    Messages:
    28,491
    Likes Received:
    0
    Location:
    Southern Oregon
    That's not true. There have been some pretty cool demos showing the same product limited to 32-bit and then 64-bit, running on the same machine, same ram, within the 3GB limit, etc.... Showing the 64-bit version blowing through tasks much faster -- specifically memory-intensive rendering and image manipulation.
     
  6. P07r0457

    P07r0457 New Member

    Joined:
    Sep 20, 2004
    Messages:
    28,491
    Likes Received:
    0
    Location:
    Southern Oregon
    Actually it was the Apple WWDC this year.

     
  7. Corp

    Corp OT Supporter

    Joined:
    Sep 11, 2003
    Messages:
    28,201
    Likes Received:
    0
    Location:
    Providence, RI
    i wanna see that picture, even though my computer would probably catch on fire trying to load it :o
     
  8. P07r0457

    P07r0457 New Member

    Joined:
    Sep 20, 2004
    Messages:
    28,491
    Likes Received:
    0
    Location:
    Southern Oregon
    yea, check out the podcast of the event. The pic was crazy. It was HUGE.
     
  9. retorq

    retorq What up bitch??

    Joined:
    Dec 14, 2006
    Messages:
    6,061
    Likes Received:
    0
    Location:
    Mohave Desert
    The 64bit machines had the same clock speed, cache speed, sizes, etc.?
     
  10. Ebtromba

    Ebtromba Active Member

    Joined:
    Nov 22, 2005
    Messages:
    6,050
    Likes Received:
    0
    Location:
    alexandria, VA
  11. P07r0457

    P07r0457 New Member

    Joined:
    Sep 20, 2004
    Messages:
    28,491
    Likes Received:
    0
    Location:
    Southern Oregon
    it was the EXACT same machine. Not a clone, not two machines of the same config.... But the SAME exact machine.... once running 32-bit and then 64-bit. SAME machine.
     
  12. wxyz

    wxyz New Member

    Joined:
    Jul 28, 2007
    Messages:
    22
    Likes Received:
    0
    I don't doubt the fact that the demo showed a difference, but really i cannot understand why such a gap could be caused by the different addressing technique.

    Maybe could the software piece be badly written (I mean not optimized in the 32 bit version) or could it be an operating system issue?

    Or could the demo be built exactly to show a huge difference in certain rare operations? (you can program an ad-hoc example showing such a thing)

    there are two things that i can think of:

    1. performance of memory access above the 4gb limit*; if you have to resort to paging, you have to tell the chipset to switch the page for you, thus you could have to slow down to perform this step

    I don't expect it to be 3 times the time required for a memory access in an _usual_scenario; to program something that have such an impact you should do something like following a *very* long list made pointer-by-pointer where the nodes are accurately scattered among various memory blocks (say: interleaved, the wrong way)

    2. you are doing something which doesn't really have to do with the addressing but with the other use of long registers: fixed point arithmetics.

    You could be in one of the cases when you were using floating point (usually slower) operations in place of fixed point/integer (usually faster) because 32 bits weren't enough to store your numbers without loosing too much precision.


    anyway, still i can't see why a 4gb 64 bit machine should be better at multi-threading than a 4gp 32 bit machine with the same overall architecture; on the opposite - if we are unlucky - the use of 64bit long pointers could fill the on-processor caches more easily, ultimately slowing down overall memory access bandwidth.

    if mp/mt is what i want to achieve, i would rather have a huge number of cores with relatively short cache hierarchy (to lower the latency, especially the one coming from context switching when it is necessary) and a nice crossbar or the like - instead of a bus - to access my memory modules. 32 bit per core would be ok ;)

    what do you think?



    *for physical or logical addressing limitations under win32, take a peek on wikipedia about pae and 3gb.


    --
    I'm in the process of learning English, feel free to correct me!
     
    Last edited: Aug 22, 2007
  13. P07r0457

    P07r0457 New Member

    Joined:
    Sep 20, 2004
    Messages:
    28,491
    Likes Received:
    0
    Location:
    Southern Oregon
    Damn you're dumb. 62-bit is MORE than just an answer to the 4GB address limit. And it's MORE than just an answer to the 2038 date problem.


    64-bit apps can access more data per cycle than 32-bit apps. A properly optimized, memory-intensive 64-bit app will out-perform a properly optimized, memory-intensive 32-bit app.
     
  14. wxyz

    wxyz New Member

    Joined:
    Jul 28, 2007
    Messages:
    22
    Likes Received:
    0
    64 bit-ness is basically all about memory access and use of integers / fixed point, because it's defined by the length of the general purpose / indexing processor registers.

    _actual_ memory access (on cheap architectures with conventional memory access schemes) is already (32 bit) nearly as fast as 64 bit because what you really fetch/write back through the slow paths (outside the processor) is done in block/in successive addresses.

    the thing you say about applications is true or wrong depending on what the application does. it's easy to build a benchmark that seems faster in 64 bits but *SCREAMING HORROR* even the contrary: 32 bits faster than 64.

    it basically depends on how much screwed up (or "how layed out") is the memory structure you want to access, how the relative memory speeds (ram, caches) are, and how small/big the smallest data bit you want do process is.


    sad but: as for a lot (everything?) else people like to think as an absolute truth, it depends on the observation perspective.


    for average use, at the end of the day, the performance is really really close.
     
  15. P07r0457

    P07r0457 New Member

    Joined:
    Sep 20, 2004
    Messages:
    28,491
    Likes Received:
    0
    Location:
    Southern Oregon
    depends on what you consider "average" use. If that involves editing photos, for example, I think 64-bit is better.
     
  16. skinjob

    skinjob Active Member

    Joined:
    Jan 6, 2001
    Messages:
    2,337
    Likes Received:
    0
    Location:
    Aztlán
    I'd be suspicious of apple since they've been known to massage these type of comparo's before.

    The explanation could be as simple as writing an app that does all its math in 64-bits, then compiliing it using 32-bit and 64-bit compilers. Doing 64-bit math, 32-bits at a time will of course take longer. The question is, did the app really need 64-bit math to accomplish its task?
     
  17. wxyz

    wxyz New Member

    Joined:
    Jul 28, 2007
    Messages:
    22
    Likes Received:
    0
    certainly it doesn't hurt ;)

    memory-access-wise i don't sincerely think it can give a huge improvement over _performance _ (large continuous blocks => relatively small page switching overhead)

    certainly ease things on what you can actually load in your editor (provided it's 64bit native): let's say you use 16 layers of 16mpixel 32 bit depth, 4 pictures at the same time - if i'm not mistaken (...it's late in europe :) ) these are exactly 4gigs, taking away all the 32 bits of addressing (again, pae can help to achieve the same result on a 32 bit machine, if both the hardware and the software support it).

    hmmmm.... just speculating ... imo performance can indeed have a huge boost when applying filters if
    - they can be implemented in an acceptable way (with enough output quality) using the wider registers to scale fixed point numbers instead of floats.
    - taking advantage of the doubled size of the registers if the mmx/sse instructions have been improved to do so (two "3d operations" instead of one with a single instruction; here you could effectively take the advantages you were mentioning above)

    the same could hold for games geometry or physics and other matrix-crunching applications.

    on the other hand, i suspect that office/internet applications and software that relies heavily on symbolic computation and/or linked list data and/or multithreading could slightly slow down

    i would be curious to benchmark something like a web browser or web server or news server implemented in java or .net on a dual or quad-core, 32bit vs 64 bit; maybe massively-connecting p2p apps too....


    (going to sleep, good night everybody)
     
  18. P07r0457

    P07r0457 New Member

    Joined:
    Sep 20, 2004
    Messages:
    28,491
    Likes Received:
    0
    Location:
    Southern Oregon
    nothing at this point NEEDS 64-bit math unless dealing with >4GB; however, a NEED for 64-bit is different from being about to BENEFIT from 64-bit.
     
  19. skinjob

    skinjob Active Member

    Joined:
    Jan 6, 2001
    Messages:
    2,337
    Likes Received:
    0
    Location:
    Aztlán
    Crunching 64-bit numbers and addressing >4GB of memory are two unrelated things. Any app that crunches double-precision floating-point has already been doing math in 64-bits.

    I'm just saying that unless Job's reveals what's going on under the hood of his "home-brew" app, I'd be skeptical about whether that benefit in this case is real or artificial.
     
  20. P07r0457

    P07r0457 New Member

    Joined:
    Sep 20, 2004
    Messages:
    28,491
    Likes Received:
    0
    Location:
    Southern Oregon
    Jobs is not the only one to demonstrate the benefit. However his presentation had a positive impact on me -- it was quite memorable.

    There's nothing wrong with effective marketing bonded with proper technology.

    If you don't believe 64-bit is the future then you can believe whatever you want... I choose to believe the truth.
     
  21. wxyz

    wxyz New Member

    Joined:
    Jul 28, 2007
    Messages:
    22
    Likes Received:
    0
    number crunching: at least me was referring to the opportunity to use fixp instead of fp if the switch from 32 to 64 bit int registers enables such a change (depends on what we are doing).

    truth: i don't know if there is 0,1,or more.

    future: certainly i would like to see more than one available, thus appropriate resources for you that could enjoy benefit from wider int/idx registers, or appropriate for me that could prefer multithreading (and thus simplier cores, many of them, maybe sacrificing cache sizes, better interconnection structures).

    believe: this word doesn't apply to me if it means "have faith in" a specific (commercial?) strategy.

    speaking of truth: beware of the Jobs reality warping field; it's not that he lies, he just seems incredibly gifted to make you see things from the perspective he prefers*

    ... gahhh, it's morning too soon :(, going to work. have a nice day folks.


    * i'm NOT implying that people is defenseless against him, it's just that he's quick and natural in making attendees lower their defense. nobody knows why, but he succeeds.


    -- feel free to correct my english (too)
     
  22. Tom93R1

    Tom93R1 Member

    Joined:
    Aug 16, 2004
    Messages:
    411
    Likes Received:
    0
    Location:
    Tempe, AZ
    What processors are truely 64-bit? I know Intel Core2 and Athalon 64 are just 64-bit memory addressing but the core is still really 32 bit. Are Xeons the same? I believe Itanium is 64 bit wide processing, but otherwise you need something like a Sun to get it, right?
     
  23. wxyz

    wxyz New Member

    Joined:
    Jul 28, 2007
    Messages:
    22
    Likes Received:
    0
    i sincerely don't know.

    iirc once upon a time being n-bit wasn't even only based on the register _maximum_ size, but on the size of the smallest chunk of data you could address.

    thus, for an architecture to be 64 bit, it was mandatory to have all accesses with that word length. if you look at riscs there must be a large number of them that require data padding (if you want to use a single octect, you have to waste 7 octects in padding).

    e.g.: i'm currently working on a 16-bit processor (it can address and move around only 16 bit words), which has 32 bit int registers.

    the documentation officially (and imo correctly) states that the byte is 16 bit (an "hardware byte" it's not the same as an octet of course).

    in contrast, iirc intel-like arcs (64 bit too) have always been able to address single octects without padding/alignment restrictions, am i right?
     
  24. Ebtromba

    Ebtromba Active Member

    Joined:
    Nov 22, 2005
    Messages:
    6,050
    Likes Received:
    0
    Location:
    alexandria, VA
    quoted since I don't think anyone saw it.
     

Share This Page