Page 11 of 429 FirstFirst ... 9101112132161111 ... LastLast
Results 101 to 110 of 4288

Thread: Polaris 10 size / performance estimation

  1. #101
    Senior Member
    Join Date
    Dec 2010
    Location
    Switzerland Geneva Cern
    Posts
    2,937
    Quote Originally Posted by integrated View Post
    Supposedly clocks could be a good deal higher on 14nm...
    Even a 20% increase wouldnt make a ~5billion gpu faster than hawaii. If you compare 390 to 380x it is like 30% - 40% faster.
    Current rig: i5 3570 @3.8GHz, Asus 7750, Dell 24'' 2560X1440, Corsair Vengence 16GB Ram DDR3 @1600MHz, Chieftec 550W PSU, 480GB OCZ SSD + 500GB Seagate + 1TB WD hd, Windows 10 64-bit

  2. #102
    Some people are talking about GPUs hitting 1.4Ghz on 14nm - are they just being hopeful?

  3. #103
    Senior Member
    Join Date
    Jan 2010
    Posts
    3,548
    Quote Originally Posted by Nerdmaster View Post
    Even a 20% increase wouldnt make a ~5billion gpu faster than hawaii. If you compare 390 to 380x it is like 30% - 40% faster.
    But it's clearly going to have more than 5 billion transistors. IMO your scaling estimate is way off, more like a half-node shrink.

    And then on top of that, 14nm allows for far higher frequencies at the same TDP, or far lower TDPs at the same frequency.

    You get 20% clocks or 35% lower power from going from 20nm to 14nm: http://images.dailytech.com/nimage/S...ide_2_Wide.jpg We're talking about 28nm to 14nm.

    Quote Originally Posted by integrated View Post
    Some people are talking about GPUs hitting 1.4Ghz on 14nm - are they just being hopeful?
    Depends on other limiting factors in the AMD GPU design. The process may be willing, but ...

  4. #104
    Senior Member
    Join Date
    Dec 2010
    Location
    Switzerland Geneva Cern
    Posts
    2,937
    Quote Originally Posted by JeeBee View Post
    But it's clearly going to have more than 5 billion transistors. IMO your scaling estimate is way off, more like a half-node shrink.

    And then on top of that, 14nm allows for far higher frequencies at the same TDP, or far lower TDPs at the same frequency.

    You get 20% clocks or 35% lower power from going from 20nm to 14nm: http://images.dailytech.com/nimage/S...ide_2_Wide.jpg We're talking about 28nm to 14nm.
    More like a half node shrink??? I am estimating a 2x density increase over tahiti which was also the first 28nm gpu. Do you think we should expect a 2.5x increase?

    I already accounted the 20% improvement in clocks.
    Current rig: i5 3570 @3.8GHz, Asus 7750, Dell 24'' 2560X1440, Corsair Vengence 16GB Ram DDR3 @1600MHz, Chieftec 550W PSU, 480GB OCZ SSD + 500GB Seagate + 1TB WD hd, Windows 10 64-bit

  5. #105
    Senior Member
    Join Date
    Jan 2010
    Posts
    3,548
    Okay, let's do this again.

    TSMC 28nm SRAM Cell: 0.16 ?m^2 in 2012 (so this may have got smaller over time)
    Samsung 14nm SRAM Cell: 0.064 ?m^2 in 2014 (LPE)

    LPP is smaller, btw.

    0.160 / 0.064 = 2.5

    Oh man, I didn't expect it to come out so precisely.

    Clocks can increase more than 20%, that's my point, it's merely 20% from 20nm, not 28nm. 28-20nm gives a 25-25% clock speed increase on top of that. I think I read 1.65x clocks /possible/ on 14nm versus 28nm at the same TDP, but cannot find the darned image now.

  6. #106
    Senior Member
    Join Date
    Feb 2010
    Location
    Planet Earth, Solar System, Milky Way
    Posts
    3,521
    Quote Originally Posted by JeeBee View Post
    But it's clearly going to have more than 5 billion transistors. IMO your scaling estimate is way off, more like a half-node shrink
    I expect the 232 sq mm 14nm chip to be able to fight for the performance crown against a Fury X. Even if it is slightly slower, it would still be called a success.

    This is if everything goes to plan and they achieve all needed optimisations, which are currently being in the works.

  7. #107
    Senior Member
    Join Date
    Jan 2010
    Posts
    3,548
    ************ right. I'm hedging my bets on a 3072 core GPU, clocked at 1.3GHz, in around 150W.

    Although I note today AMD said Polaris is 2x better Perf/W than 28nm...

  8. #108
    Senior Member
    Join Date
    Jun 2010
    Posts
    4,237
    Quote Originally Posted by integrated View Post
    Some people are talking about GPUs hitting 1.4Ghz on 14nm - are they just being hopeful?
    For Polaris? I would say yes, but who knows.

  9. #109
    Senior Member
    Join Date
    May 2012
    Posts
    4,745
    Quote Originally Posted by integrated View Post
    Some people are talking about GPUs hitting 1.4Ghz on 14nm - are they just being hopeful?
    Depends ?

    Intels 'real' 14nm iGPUs can run at 1150. But GloFlos 14nm is more like 20nm, so in theory it should have slightly lower power density ? And so possibly be able to run a bit faster ? Not sure about 1.4, but 1.2-1.3 might be possible ? As a WAG.
    To find the right answers you must ask the right questions.

  10. #110
    Senior Member
    Join Date
    Dec 2012
    Posts
    7,432
    Quote Originally Posted by Nerdmaster View Post
    More like a half node shrink??? I am estimating a 2x density increase over tahiti which was also the first 28nm gpu. Do you think we should expect a 2.5x increase?

    I already accounted the 20% improvement in clocks.
    Use pitcairn. for 232mm^2 double that density you get about 6.1b transistors.

    Pitcairn launched a few months after Tahiti with higher density. Polaris 11 will likely launch a few months after Polaris 10.

    Polaris 10:, 2.5-3.5 billion transistors. Tahiti*100*2 ranging to Tahiti*130*2*1.08
    Polaris 11: 6.1-6.6 billion transistors. Pitcairn*232*2 ranging to Pitcairn*232*2*1.08


    [1.08 comes from density increase that Apple A9 from Samsung over TSMC]
    -Q

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