AMD Athlon 64 bit CPU preview on Xbit labs

Discussion in 'Beyond3D News' started by marco, Apr 20, 2003.

  1. marco

    Regular

    Joined:
    Jan 30, 2002
    Messages:
    303
    Likes Received:
    0
    Location:
    The netherlands
    Xbit labs previews an AMD Hammer based system. :

    "We managed to get our hands on an engineering sample of the AMD Athlon 64 2800+ processor. We couldn’t help testing it, of course! In this article we are going to discuss the major features and benchmarks results for this processor."

    Read more here
     
  2. Tagrineth

    Tagrineth SNAKES... ON A PLANE
    Veteran

    Joined:
    Feb 14, 2002
    Messages:
    2,512
    Likes Received:
    9
    Location:
    Sunny (boring) Florida
    Hmm, looks more or less as I expected.

    Kinda disappointing, in the end.
     
  3. Saem

    Veteran

    Joined:
    Feb 7, 2002
    Messages:
    1,532
    Likes Received:
    6
    The only thing disappointing is the clock rate.
     
  4. mczak

    Veteran

    Joined:
    Oct 24, 2002
    Messages:
    3,015
    Likes Received:
    112
    I'd rather say the performance rating is disappointing. IF (and that's IMHO a quite big if) the performance stays the same until it's released, a 2800 rating doesn't seem to be justified overall for the 1.6Ghz Athlon 64. 2400 or maybe 2600 would be more honest, since people will be comparing this to P4 C versions (yeah I know AMD says the model rating is supposed to compare to a thunderbird). If you also consider in FPU-intensive (not SSE2 optimized) applications it won't even beat a Athlon XP 2000+ (since its FPU seems to be 100% identical to the one in the Athlon XP) a 2800 rating seems a bit high.
    This is, of course, assuming the model rating is based on 32bit OS/applications - it should be a bit faster (not the fpu scores though) in 64bit OS/applications (how much is anyone's guess, but probably not too much - AMD themselves said it won't be much, I'd guess around 5% or so for "typical tasks"). So even if you'd consider the 64bit mode the performance rating probably won't look good.

    mczak
     
  5. vrecan

    Regular

    Joined:
    Feb 17, 2003
    Messages:
    284
    Likes Received:
    0
    I've read that when ported to 64bit you can usually see between 15-25% performance improvement in server situations. And this is without really optimizing for 64bit code so who knows what is really possible. The only thing I know is this is kinda dissapointing I wonder how much better the opterons fair?
     
  6. Saem

    Veteran

    Joined:
    Feb 7, 2002
    Messages:
    1,532
    Likes Received:
    6
    What are you talking about? Much of the Hammer is the same interms of execution resources when compared to the Palimino. There is very little difference and that's mainly in the BPU. Now integer code will obviously run faster, since latency and the improved BPU will help there. The FPU will be helped since the on board memory controller and cache should feed the units much better than the Palimino's memory sub system could ever hope to do.

    In 64 bit mode you're forgetting the extra registers, you do know they'll help improve performance across the board, considering how register starved x86 is.

    As for the rating, I think the entire rating system is bunk. Use MHz, as simple as that. And if Intel has the lead, well then oh well. Education the consumer, get a decent marketing/education campaign going. It might help them in other areas, like demanding higher prices.
     
  7. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    from the X-bit labs article
    Not a very professional statement, IMHO. One should atleast expect that X-bit labs would do a little research, before doing a 64bit processor preview. There is a handful of 64bit-ready Linux distros around presently, and UT2k3-64 was shown running on SuSE (I think) 64bit-edition a while ago. nVidia also has their linux drivers available for both IA64 and AMD64. This isn't totally fresh news, X-bit labs... :oops:
     
  8. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    I would expect 0% improvement with 64 bit codes other than if new registers come into play. Maybe 64 bit windows or linux has different page sizes which may make addressing faster but I doubt it.

    You may get 25% improvement only if youre addressing large amuonts of memory since you dont have to do software hacks to get around addressing > 4GB.
     
  9. Saem

    Veteran

    Joined:
    Feb 7, 2002
    Messages:
    1,532
    Likes Received:
    6
    Well the extra registers will almost always come into play. Have you ever done any x86 assembly?
     
  10. Dave H

    Regular

    Joined:
    Jan 21, 2003
    Messages:
    564
    Likes Received:
    0
    The extra registers will always come into play assuming the application is explicitly compilied as x86-64. Unfortunately, the only x86-64 aware compiler at the moment is GCC, and even a 15-25% improvement over plain x86 GCC performance may still be slower than code compiled with Intel's compiler. (Which, quite obviously, is never going to be x86-64 aware, barring revelations on the Yamhill front.) The thing to watch is what compiler is used in the Opteron SPEC submissions on Tuesday...

    Of course in the real world the only compiler that matters is Microsoft's. Once it has full, optimized x86-64 support (assuming it ever does) we should see the extra registers boost in any application that ships with a seperate 64-bit version, which is to say none and Unreal Tournament '03. Well, 64-bit Windows, although who knows how much of the codebase will actually be ported.

    The extra registers should add a significant performance boost (15-25% sounds excessive to me, but we'll see), but even if the tools were in place for significant support, porting to x86-64 requires significant effort to make your source 64-bit clean, and means shipping a seperate binary. Frankly, the performance benefit is smaller and the obstacles larger than with 3DNow!, and we all know how widespread that turned out. (Of course it wasn't negligable and neither will x86-64 be.)

    The transition to 32-bit x86 software took 10 years (counting from the introduction of IA32 with the 386) to get off the ground as far as the average consumer was concerned. x86-64 won't take nearly that long, but it won't make any difference anytime soon either. For the forseeable future, Hammer's 64-bitness is about as useful as Parhelia's support for displacement mapping. (As far as the consumer market is concerned. If you're buying an Opteron to run a database on Linux, it'll matter a lot sooner; then again, the larger register file isn't going to affect database performance much.)
     
  11. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    Intel fanboys.

    Why so freaking negative? Can't you look past your bias Dave H, do you work for Intel???? It will benefit all of us if AMD succeeds with the Athlon64! Gaming performance is absolutely mindblowing, a 1.6Ghz chip beats the 2.8Ghz 800Mhz Fsb p4 in UT2003(very Cpu intensive game). If they could release a 2.2, 2.4 Ghz versions of the Athlon64 I'd be in heaven. I do not want to see Intel as a MS like monopolist in the Cpu market. :(
     
  12. mczak

    Veteran

    Joined:
    Oct 24, 2002
    Messages:
    3,015
    Likes Received:
    112
    No, I'm not forgetting them - I put the above 5% guess there exactly because of the extra registers. Someone else mentioned 15%-25%, but that was server tasks - and let's face it, typical non-server tasks will not profit from the 64bit registers, but only because there are more registers. 5% seems low, since the x86 is (as you mention) really register starved, but afaik register renaming does quite a good job today, so don't expect miracles.
    And FP scores should not improve in 64bit mode, since the fpu isn't changed at all - still the same stack-based mess (IIRC AMD wanted to introduce a completely new fp mode, but abandoned it in favor of SSE2). Though I could be wrong on that, maybe x86-64 compilers will always use SSE2 instead of the old FPU.

    I don't like the rating either, but as long as computers get advertized almost exclusively by their MHz rating, that's hard for AMD to change - intel has a bit more advertizing budget I think...
    So the rating as such isn't that much of a problem IMHO - if you're interested in the real Mhz number for the right reasons, you can still figure this out easily.
     
  13. WaltC

    Veteran

    Joined:
    Jul 22, 2002
    Messages:
    2,710
    Likes Received:
    8
    Location:
    BelleVue Sanatorium, Billary, NY. Patient privile
    Why on earth would you think "MHz" is an accurate barometer of anything other than MHz when you are dealing with differing cpu architectures? MHz tells you absolutely nothing in that regard. If one cpu does more work per clock than another, it can be clocked at a lower MHz--and run just as fast (if not faster) than the cpu which does less work per clock but is clocked higher. Therefore to go strictly by MHz when comparing cpus of differing architectures can be nothing but highly misleading. (Contrast Itanium with P4, or Banias with P4, for instance--don't even have to go outside of Intel to see why MHz comparisons like this are worthless.)

    When comparing identical chip architectures then MHz is a valid performance indicator--but that's as far as it goes.
     
  14. Tagrineth

    Tagrineth SNAKES... ON A PLANE
    Veteran

    Joined:
    Feb 14, 2002
    Messages:
    2,512
    Likes Received:
    9
    Location:
    Sunny (boring) Florida
    Re: Intel fanboys.

    IF. And that's a really REALLY big IF. By the time AMD has a 2.2 or 2.4GHz Athlon64, Intel will probably have a >3.5GHz Prescott...
     
  15. Ilfirin

    Regular

    Joined:
    Jul 29, 2002
    Messages:
    425
    Likes Received:
    0
    Location:
    NC
    Here's a vote for an "average instructions per clock * clockspeed" based metric..
     
  16. Saem

    Veteran

    Joined:
    Feb 7, 2002
    Messages:
    1,532
    Likes Received:
    6
    Server tasks were benifiting from the onboard mem controller and the extra registers. The 64bit ness really didn't have much to do with it. Register renaming is nice, but I don't think it's as nice as the extra registers.

    I'll concede the FPU point, you're right.
     
  17. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    Re: Intel fanboys(girls).

    [/quote]
    IF. And that's a really REALLY big IF. By the time AMD has a 2.2 or 2.4GHz Athlon64, Intel will probably have a >3.5GHz Prescott...[/quote]


    What the hell is it with all the negativity shoveled towards AMD, if it wasnt for them you'd now be paying 1100$ for a 2.0Ghz chip. I don't give a crap if the Athlon64 is faster than Prescott, as long as it stays competitive performance wise I'll get it, just like now I own 3 AthlonXP based systems. :roll:
     
  18. Saem

    Veteran

    Joined:
    Feb 7, 2002
    Messages:
    1,532
    Likes Received:
    6
    I'm all for competition,

    I'm guessing you're either new to the board or haven't figured out how things work here. A statement of fact doesn't imply advocacy, figure this out before you continue posting. AMD looks like it's in trouble, financially and possibly on the technical front with their CPUs.
     
  19. mczak

    Veteran

    Joined:
    Oct 24, 2002
    Messages:
    3,015
    Likes Received:
    112
    Re: Intel fanboys.

    IMHO AMD should be able to get out 2.2Ghz Athlon 64 pretty quickly. This is the frequency some Athlon XP already run at, and since the Athlon 64 is very similar (with even slightly longer pipeline) it should reach at least the same frequency if they'd use the same manufacturing technology. And SOI shouldn't reduce the achievable clock rates afaik, but maybe there are some problems left with that (SOI should have been used already for barton according to some roadmaps, but it apparently wasn't ready yet).
    Of course, if intel sticks to their schedule, we'll see prescotts this year. They should easily reach clockrates above 4Ghz (intel won't release them at that speed intially, but they overclock probably pretty well unless intel locks them like described in their patent), so AMD obviously needs a 90nm process next year if they want to compete at the top-end.

    mczak
     
  20. Anonymous

    Veteran

    Joined:
    May 12, 1978
    Messages:
    3,263
    Likes Received:
    0
    fanboys

    Where the hell do you see facts here pal, all I see is assumptions. I'll give you a fact, as it stands AMD IS competitive with Intel and this had lead to lower prices across the board, a clear benefit for the consumer.
     
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...