AMD Polaris 10 GPU To Offer Near 980 Ti Performance For 299 USD?

Discussion in 'Frontpage news' started by Hilbert Hagedoorn, Apr 26, 2016.

  1. Ieldra

    Ieldra Banned

    Messages:
    3,490
    Likes Received:
    0
    GPU:
    GTX 980Ti G1 1500/8000
    Do you realize Pascal, unlike Polaris, is an architecture? Saying 390x compete with Pascal makes no sense,you need to specify gpu. I expect 390x to compete with gp106. Not that it's much of a competition when power consumption is like 1/3

    As it stands now the high end will be totally uncontested until amd releases Vega, which sucks. For everyone.
     
    Last edited: Apr 26, 2016
  2. Fox2232

    Fox2232 Ancient Guru

    Messages:
    11,808
    Likes Received:
    3,370
    GPU:
    6900XT+AW@240Hz
    GCN has different ROPs than Kepler/Maxwell, so different performance per ROP per clock.
    But think about R9-290x/390x and its performance per clock. Reason why it is in some scenarios very close to Fury X is that bot have 64 ROPs.

    And that is why I think that 2560SP+160TMUs + 64ROPs will reach similar performance as Fury X if clocked around 1150~1250MHz.

    If AMD really wanted 'Titan Killer', all they needed was to give Fiji 16 more ROPs instead they stayed with 290X count.
     
  3. Ieldra

    Ieldra Banned

    Messages:
    3,490
    Likes Received:
    0
    GPU:
    GTX 980Ti G1 1500/8000

    Wait man, there's no such thing as different performance per rop per clock. Rop performance is in pixels per clock, if you have more pixels per clock you have more rops.

    If anything I think high memory bandwidth helps fury X vs hawaii

    If amd had wanted more rops on Fiji they would have been able to fit them :p
    [​IMG][​IMG]

    2560 at 1300 is still WAY less powerful than Fiji

    The real advantage gcn 2.0 brings vs previous is improved front-end, meaning you hopefully will never see **** like this again

    [​IMG]
     
    Last edited: Apr 26, 2016
  4. vase

    vase Ancient Guru

    Messages:
    1,652
    Likes Received:
    2
    GPU:
    -
    No.

    Fox is correct. Because simplified: Fillrate = ROP * Clock

    so if you have different ROPs that means different performance out of the ROP per Clock calculation.

    You are clear that a ROP is a hardware component right?
     

  5. Ieldra

    Ieldra Banned

    Messages:
    3,490
    Likes Received:
    0
    GPU:
    GTX 980Ti G1 1500/8000
    Let me put it this way.

    The GTX 970 has 56 ROPs that can process one pixel each, per clock, or you can says it has 7 ROPs that can process 8 pixels each, per clock.

    The 970 is limited to 52 pixels per clock by it's SM count. Which is why having 56 ROPs vs 64 doesn't matter because it won't help peak fillrate.

    To be frank with you vase, I think i understand the hardware much better than you do.

    This is from before the 970 fiasco so that's why it still says 64 rops
    http://techreport.com/blog/27143/here-another-reason-the-geforce-gtx-970-is-slower-than-the-gtx-980
     
    Last edited: Apr 26, 2016
  6. Fox2232

    Fox2232 Ancient Guru

    Messages:
    11,808
    Likes Received:
    3,370
    GPU:
    6900XT+AW@240Hz
    What I mean is simple thing, Fury X has 45% more TMUs/SP over R9-290x/390x.
    But very same ROP count. In scenarios where TMUs/SP are main players Fury X can pull away, but Fury X is mostly very small improvement over R9-390x in ROP tense situations.
     
  7. Ieldra

    Ieldra Banned

    Messages:
    3,490
    Likes Received:
    0
    GPU:
    GTX 980Ti G1 1500/8000
    Yeah, which is why I've been arguing that the Fury X isn't rop limited most of the time, because it outperforms 390x despite higher clocks on the latter, even at 4k

    Fury X has same TMU/SP as 390x btw, or did you mean it has 45% more TMUs and SPs ?

    Cause that makes sense :p I thought you meant 45% TMU:SP ratio so I was like wwaaaaaaaat is he smoking
     
    Last edited: Apr 26, 2016
  8. Mugsy

    Mugsy Master Guru

    Messages:
    287
    Likes Received:
    15
    GPU:
    RX 5700 XT Pulse
    Say what now ? Polaris is an architecture.
     
  9. Ieldra

    Ieldra Banned

    Messages:
    3,490
    Likes Received:
    0
    GPU:
    GTX 980Ti G1 1500/8000
    So many people get this confused. When you say Polaris I know you're talking about Polaris 10 and 11. When you say PASCAL I don't know what the **** you're talking about, because Pascal isn't the name of any gpus.

    Polaris 10, 11 and Vega 10, 11 are all GCN 2.0 architecture.

    Edit: wait apparently they're 1.3
     
    Last edited: Apr 26, 2016
  10. zer0_c0ol

    zer0_c0ol Ancient Guru

    Messages:
    2,976
    Likes Received:
    0
    GPU:
    FuryX cf
    gcn 4.0
     

  11. Denial

    Denial Ancient Guru

    Messages:
    13,973
    Likes Received:
    3,732
    GPU:
    EVGA RTX 3080
    No it's not, GCN is the architecture. Polaris is just a name for a subset of GCN chips.

    What happened to 2 and 3?
     
    Last edited: Apr 26, 2016
  12. Ieldra

    Ieldra Banned

    Messages:
    3,490
    Likes Received:
    0
    GPU:
    GTX 980Ti G1 1500/8000
    Such a common mistaken and it's why I get so many people bitching and whining at me saying "WHY DO YOU THINK POLARIS IS WORSE THAN PASCAL" when I'm actually talking about GP104.

    I may as well be trying to talk to my cat



    They were divided by 10 and summed with 1 :p
     
  13. Denial

    Denial Ancient Guru

    Messages:
    13,973
    Likes Received:
    3,732
    GPU:
    EVGA RTX 3080
    This whole forum is like that, especially lately. Unless you literally specify/clarify everything, someone will quote you and pretend like you have no understanding of anything at all.
     
  14. zer0_c0ol

    zer0_c0ol Ancient Guru

    Messages:
    2,976
    Likes Received:
    0
    GPU:
    FuryX cf
    amd does not count them like that for some reason

    2011 — 28nm GCN 1.0 — Tahiti / Cape Verde
    2013 — 28nm GCN 2.0 — Hawaii / Bonaire
    2015 — 28nm GCN 3.0 — Fiji / Tonga
    2016 — 14/16nm FinFET Polaris (GCN 4.0)
     
  15. Denial

    Denial Ancient Guru

    Messages:
    13,973
    Likes Received:
    3,732
    GPU:
    EVGA RTX 3080

  16. zer0_c0ol

    zer0_c0ol Ancient Guru

    Messages:
    2,976
    Likes Received:
    0
    GPU:
    FuryX cf
  17. Ieldra

    Ieldra Banned

    Messages:
    3,490
    Likes Received:
    0
    GPU:
    GTX 980Ti G1 1500/8000
    it's 1.x
     
  18. zer0_c0ol

    zer0_c0ol Ancient Guru

    Messages:
    2,976
    Likes Received:
    0
    GPU:
    FuryX cf
    No it is not.. amd does not count them as such
     
  19. Fox2232

    Fox2232 Ancient Guru

    Messages:
    11,808
    Likes Received:
    3,370
    GPU:
    6900XT+AW@240Hz
    That's right. But there is optimal ROP to TMU to SP count. And there are so many wild combination on AMD side.

    There is similar situation with HD 7850 and HD 7970.
    http://www.gpureview.com/show_cards.php?card1=678&card2=663
    As HD 7850 has bit lower clock, HD 7970 has double performance of everything inside with Exception of ROP work as both have 32 ROPs.

    Yet HD 7970 is only 50% faster in average than HD 7850.

    I wonder what's perfect ratio for GCN, but I think that TMU+SP/ROP ratio of HD7970 = Fury X is not best, it needs more ROPs.
    That HD 7970 is only 0~6% faster than HD 7950 @ HD 7970 clock.
    http://hexus.net/tech/reviews/graphics/34761-amd-hd-7950-vs-hd-7970-clocks/?page=2
    But HD 7970 has 14% more TMUs.
     
  20. Denial

    Denial Ancient Guru

    Messages:
    13,973
    Likes Received:
    3,732
    GPU:
    EVGA RTX 3080
    I can't find a single slide where it specifically says "4.0" just "4th Generation"

    Regardless it's semantics, so i don't even know why your making a big deal about it. Which is like my entire point to that one post. No matter what you say on this forum, your quoted and corrected for pointless stuff.

    It's still GCN. Polaris is still a subset of GCN products. Polaris is still not an architecture.
     

Share This Page