Comments Locked

28 Comments

Back to Article

  • psychobriggsy - Saturday, November 5, 2005 - link

    We will be getting the 2.6GHz S939 FX60 dual-core processor soon - I assume it has 2x 1MB L2 cache. This will be a nice performance processor for people who like spending money on their system.

    Sadly I assume there will be no FX59 3GHz single core FX. I imagine that this speed isn't a real goer on the 90nm process for AMD! It's sad because 3GHz would have been a nice number for AMD to reach this year, even if it cost $1200 a processor and they only ever made 1000 or so, lol.

    What is interesting is that by the middle of next year, AMD will have a dual-core 2.8GHz FX62 processor - the same speed as the current FX57, but with two cores! I assume this will cost around $1100 and thus be pointless for most of us, but if AMD can make them, then a Socket M2 4200+ starts to look very interesting for overclocking (wot, no M2 X2 3800? :( ) Also the 2.6GHz X2 will emerge, and that's a good thing too. It'll certainly keep the pressure on Intel through 2006.

    And what's with the M2 Sempron 3500+? Why not make the line-up neat for once - 3000+, 3200+, 3400+, 3600+, 3800+?
  • IntelUser2000 - Friday, November 4, 2005 - link

    Jarred Walton said he thought of IPC on Pentium 4 and Athlon 64 as follows:

    Pentium 4: ~1.5
    Athlon 64: ~2.2

    The relative figures are right but I have to disagree on the numbers. Itanium 2 1.6GHz 9MB L3 cache's IPC on specint is ~1.5(if anyone doesn't agree that spec is a good benchmark for real world apps, IT IS for workstation/HPC apps).

    Considering that Itanium 2 at 1.6GHz it achieves score of 1590 for best score on specint, and Athlon 64 FX gets 1970 at 2.8GHz and Pentium 4 gets 1815 with 3.8GHz and 2MB L2, I have to say I disagree with your numbers.

    Just on specint, the numbers would be:
    Pentium 4: 0.7
    Athlon 64 FX: 1.05

    I would say the figures are also right for other apps since most CPUs don't even get close to ideal IPC.
  • bob661 - Friday, November 4, 2005 - link

    Anyone here planning to buy a M2 socket CPU when it's released? I'm kind of on the fence. I COULD use Pacifica and dual cores though.
  • Googer - Thursday, November 3, 2005 - link

    A few years ago, I took the time to read the DDR2 white papers. They stated that as part of the specification that DDR2 Memory controllers maintain backward support for the original DDR specification, meaning that a DDR2 controller should function with DDR1 installed in it's sockets. Intel seems to have completely ignored this in almost all of their chipsets, Except for the i915 chipset memory controller. I sure hope AMD maintains backward compatability with DDR1 on socket M2.

    But again it has been a while since I read this and somethings probably have changed, all I can do is be hopefull.
  • xsilver - Thursday, November 3, 2005 - link

    has it been discussed as to if M2 will bring cost savings to amd? cost savings and hence cheaper prices for us? (staring straight at the x2-3800 for around $200)

    also would it be a bad time to go from 1gb ram to 2gb because of the immenent move to ddr2?
  • JarredWalton - Thursday, November 3, 2005 - link

    I don't expect M2 chips to be any cheaper to produce (until they move to 65nm process), and I also don't expect them to be significantly faster than their 939 counterparts. I think this will be very much like the change from socket 478 to socket 775 Pentium 4 chips. A few areas will be faster, but if you have a fast 939 system there won't be a major incentive to immediately upgrade to M2.
  • Anton74 - Thursday, November 3, 2005 - link

    The 3700+ and X2 4400+ are not listed, and a Manchester core FX instead of Toledo?

    Is AMD cutting down on cache, or is this just a coincidence?
  • stephenbrooks - Friday, November 4, 2005 - link

    Yes, I noticed that too. Right now they have a sort of 2x2 matrix of products with all combinations from {2x512KB,2x1MB}x{2.2GHz,2.4GHz}. That's actually quite a confusing product line for your average pleb to understand in a way.

    I think what they might be seeing is that people are reluctant to pay for extra cache on a lower-clocked chip - they prefer to go for the higher clocks first (which garauntees a performance increase) and then up the cache (which only gives patchy increases), so we're seeing the large caches only on the top processors.
  • stephenbrooks - Thursday, November 3, 2005 - link

    What do you folks think the frequencies of these things are? I'm guessing:
    FX-60 2x2.6GHz
    FX-62 2x2.8GHz
    ...so you could see that as +5 for dual core.
  • xenon74 - Thursday, November 3, 2005 - link

    Well if X2 5000+ M2 will be 2,667GHz (8x333)
    then X2 5200+ M2 may be 2,833GHz (8,5x333), or even 3GHz (9x333) but highly unlikely
    and:
    X2 4800+ M2 (7,5x333=2,5GHz) full L2
    X2 4600+ M2 (7,5x333) half L2
    X2 4400+ M2 (7x333=2,33GHz) full L2
    X2 4200+ M2 (7x333) half L2
    and:
    FX?? 939 DC(13X200=2,6GHz) full L2
    FX62 M2 DC(8,5x333=2,833GHz) full L2

  • JarredWalton - Thursday, November 3, 2005 - link

    HyperTransport is not related to the RAM speed. There is no reason to assume that the CPU bus/HT bus will change from 200 MHz. The roadmaps are still stating 1000 MHz HyperTransport, and while 3x333 is 1000 MHz, 5x200 will also get you there. The memory controller already runs asynchronously to the HT speed - regardless of what memory speed you run. There is no "1:1 overclocking" on AMD platforms. Consider:

    3800+ = 12X200 = 2400 MHz
    RAM = CPU/6 = DDR800
    RAM = CPU/8 = DDR600
    RAM = CPU/9 = DDR533
    RAM = CPU/12 = DDR400
    RAM = CPU/15 = DDR320
    RAM = CPU/18 = DDR266

    So my bet is that M2 chips remain as 200 MHz increments, HyperTransport remains at 200 MHz base speed, and the only thing that changes is the CPU divider that generates the memory speed. However, as with your figures, this is only a guess. Maybe they will move to a 333 MHz base bus speed. I'd be surprised, but anything is possible.
  • xenon74 - Friday, November 4, 2005 - link

    Hm, I always thought that relation between CPU and Memory was more of a ratio than a divider and CPU base frequency reacts as kind of a baseline.
    DDR400 200MHz 1
    DDR366 183MHz 11/12
    DDR333 166MHz 5/6
    DDR300 150MHz 3/4
    DDR266 133MHz 2/3

    DDRII-667 333MHz 1
    DDRII-600 300MHz 9/10
    DDRII-533 266MHz 4/5
    DDRII-500 250MHz 3/4
    DDRII-466 233MHz 7/10

    In this case ratio stays at every given stock CPU frequency (2GHz, 2.2GHz,...).
    Situation is quite different with divider usage - as it changes according to CPU frequency and frankly I think that's not very likely.

    Nevertheless I speculate of 333 mostly because of PC Welt article.



  • JarredWalton - Friday, November 4, 2005 - link

    Ratios were the way of most previous chipsets and CPUs - socket 775, 478, 462, 423, 370, etc. When AMD moved the memory controller onto the CPU, the ratio system suddenly became meaningless. The memory still runs at a ratio, but it's a ratio of the CPU speed. CPU-Z will report the memory divider on socket 754/939/940. Usually, it's the same as the CPU multiplier, but if you want to run the RAM slower or faster, it can be a lot of other values.
  • Jep4444 - Thursday, November 3, 2005 - link

    i dont think we can guess the Socket M2 Fx so easily because they will have a higher HTT than 200mhz
  • tfranzese - Thursday, November 3, 2005 - link

    quote:

    Other interesting noteables include the lack of any new Socket 939 SKUs. Even though the AMD roadmap clearly states the Socket 939 will stick around into Q1'07, the only new processor we actually see between now and then is the FX-??. Hopefully Socket M2 sticks around long enough to actually get one or two different processors on the same board like in the old days of Socket 463. In fact, even though the roadmap explicitly claims that Socket 939 Semprons will show up before the end of the year, the only new Semprons listed are Socket M2.


    If my memory is correct, it was socket 462.
  • Googer - Thursday, November 3, 2005 - link

    quote:

    Other interesting noteables include the lack of any new Socket 939 SKUs. Even though the AMD roadmap clearly states the Socket 939 will stick around into Q1'07, the only new processor we actually see between now and then is the FX-??. Hopefully Socket M2 sticks around long enough to actually get one or two different processors on the same board like in the old days of Socket 463. In fact, even though the roadmap explicitly claims that Socket 939 Semprons will show up before the end of the year, the only new Semprons listed are Socket M2.



    Motherboards are so cheap I just wonder why any one worries about getting so much life out of them?

    Oh and FYI, it is socket 462 not 463 aka Socket A.
  • phaxmohdem - Thursday, November 3, 2005 - link

    Ahh Socket 463 the good ol' days :) I must have missed that one though, went straight from 462 to socket 940. I've heard those 463 socket chips will give an FX-57 a run for their money when overclocked to 8.6756 GHz though.. Higher or lower than that however and you're running with the PII's :(
  • JarredWalton - Thursday, November 3, 2005 - link

    Typo fixed. Thanks!
  • Pete84 - Thursday, November 3, 2005 - link

    Agreed, I was wondering what they were referencing, then I realized it was Socket A.
  • nserra - Thursday, November 3, 2005 - link

    Isn’t FX64 misleading for AMD64 FX64?

    What was the problem with FX61? FX63?

    And the end of high speed single core is really necessary.
    Only if dual core lags behind 200Mhz clock of single core. But if it’s at a rate of 400Mhz, I think I would keep both.
  • dualblade - Thursday, November 3, 2005 - link

    even numbers for dual core fx chips, odd numbers for single core. since they're all going to be dual core, we shouldn't expect any odd numbers (unless amd decides to get frisky with a dual core fx69 ;) )
  • highlandsun - Thursday, November 3, 2005 - link

    Ah yes, two cores, one inverted and facing the other, and ...
  • Missing Ghost - Thursday, November 3, 2005 - link

    lol! hot cpu!
  • microAmp - Thursday, November 3, 2005 - link

    FX-61 to FX-63 will eventually lead to FX-69; maybe AMD doesn't want that and choose an even number route.
  • Gamingphreek - Thursday, November 3, 2005 - link

    quote:

    he only new processor we actually see between now and then is the FX-60


    You said you just went under NDA and had to delete a certain processor from the list. I think you just overlooked it right here. ;)

    Good article as always though!!

    -Kevin
  • ksherman - Thursday, November 3, 2005 - link

    In fact, even though the roadmap explicitly claims that Socket 939 Semprons will show up before the end of the year, the only new Semprons listed are Socket 939.

    probably should say "...listed are Socket M2"
  • Furen - Thursday, November 3, 2005 - link

    Is the new s939 FX really going to be a manchester core, or is that just a mistake? I'd expect it to be a Toledo, at least...
  • KristopherKubicki - Thursday, November 3, 2005 - link

    The roadmap outlines it as Manchester, although I would expect Toledo as well.

    Kristopher

Log in

Don't have an account? Sign up now