Page 28 of 49 FirstFirst ... 18262728293038 ... LastLast
Results 271 to 280 of 486

Thread: 28nm APU speculation.

  1. #271
    Senior Member
    Join Date
    Jun 2010
    Quote Originally Posted by integrated View Post
    That would be tragic!

    Another generation wasted on single-channel memory...
    They can improve the memory controller and use faster RAM.
    And it could be enough for an APU below 4w or 9w, the performance with this power is not going to be much better than the one we have with Brazos.

  2. #272
    Senior Member
    Join Date
    Jul 2011
    Nobody said anything about socket.

    But yeah, although nobody said "APUs are going to be manufactured only on GF", only GF was referenced, so GF it is.

  3. #273
    Quote Originally Posted by kalelovil View Post
    Admittedly Brazos gets a lot less than theoretical performance from its single memory channel, but doubling the width of the interface is not a feasible solution for chip meant for small, low-cost systems.
    Exactly, that would be a lot of extra cost to put two memory sticks in the pc and some room to route the traces. Not to mention that dual channel would force some minimum die size on the chip, for not that much in return.

    Given how the Fusion memory controller progressed since Brazos launch, it makes far more sense to add the circuits to make best use of available bandwidth than add more. I'd expect a DDR1600, maybe DDR1866 stick should be a good match for a 2CU GCN based GPU.

  4. #274
    Senior Member
    Join Date
    Sep 2011
    We pretty much know Geforce 640m is performing nicely in notebooks so I was kinda interested how should Kaveri IGP perform against it. I think comparing 3dmark vantage GPU score between them to have some clue about performance isn't a bad idea. My estimate is 500-600Mhz for the best mobile Kaveri.
    3dmark vatage GPU score
    Geforce 640m: 6450
    desktop HD 7750: 10012
    mobile Kaveri IGP(500-600Mhz): 6258-7509(-3% or +16% vs 640m)
    This of course means Kaveri would need 45-54GB/s to feed it if HD 7750 has 72GB/s not to mention the cpu needs some bandwidth too.
    So next year(Q1-Q2) you can get ~Geforce 640m performance depending on bandwidth and clock limitations, nice.

  5. #275
    I'm starting to think that Kaveri might need it's own memory on-die to really shine. It's time for AMD to make this move I believe - it will catapult them miles ahead of intel and basically kill Nvidia's low range dead.

  6. #276
    Quote Originally Posted by TESKATLIPOKA View Post
    My estimate is 500-600Mhz for the best mobile Kaveri.
    If you look at the AFDS presentations, they have leaked estimates for Kaveri - I don't remember the exact figure but IIRC it was in the ~900 MHz range. Of course i.e. 8 GCN CU. Could be 600+ for mobile then.

  7. #277
    8-bit overflow
    Join Date
    Apr 2012
    Kaveri will probably be clocked pretty high. 28nm gnc has proven to be able to clock high without the need for much power.

  8. #278
    Quote Originally Posted by James View Post
    I'm starting to think that Kaveri might need it's own memory on-die to really shine.
    Why don't you consider off-chip and off-package?

  9. #279
    Quote Originally Posted by James View Post
    So CES next year could be Kabini or Temash launch. It also appears to have socket compatibility with Brazos/Brazos 2.
    As AMD integrates FCH into Kabini and Temash APUs, no socket compatibility will be maintained.

  10. #280
    Quote Originally Posted by Guild View Post
    Kabini and Temash will come first (I am assuming they are two variants of the same die, ala Zacate/Ontario) at CES 2013 and Kaveri should come in Q2.
    They are unlikely the same in my POV. They are both SoC chips, but the Temash chip is optimized for tablets. Remember that AMD got a streamlined FCH in Hondo.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
WordPress Appliance - Powered by TurnKey Linux