AMD RX480-X and RX490

Discussion in 'Desktop Hardware' started by Support.2@XOTIC PC, Aug 17, 2016.

Thread Status:
Not open for further replies.
  1. Support.2@XOTIC PC

    Support.2@XOTIC PC Company Representative

    Reputations:
    486
    Messages:
    3,148
    Likes Received:
    3,490
    Trophy Points:
    331
    I rarely see discussion on the AMD cards, so I thought I'd just make one :) I've always been an Nvidia guy, but this generation of cards are weird. What does everyone think about the RX480?

    What are your expectations regarding a possible RX480-X or RX490 release?

    Which cards will age better? Does AMD have the upper hand in DX12?
     
  2. Mr.Koala

    Mr.Koala Notebook Virtuoso

    Reputations:
    568
    Messages:
    2,307
    Likes Received:
    566
    Trophy Points:
    131
    The 480 is already fully enabled, there's no chance for a RX480X unless they give that name to a different chip, or if clocks of current samples are heavily limited by fab and they can offer a significant clock bump after fab maturing. Both are unlikely.

    When Vega lands and how well they will go against GP102/104 are anyone's guess now. Technically NVidia is not fully utilizing the available space with GP102 (471mm2), leaving plenty of head room to AMD, but giving the financial situation the red company has been through, I won't hold my breath waiting for them to catch any opportunity.


    For the last a few years we do see AMD aging better, though this could be largely blamed on their own software team (at least looking from the eyes of an outsider). With newer APIs getting somewhat lower access to hardware more software dev load in placed on the game engine devs, so the aging progress probably won't repeat itself (at least not necessarily in favor of AMD, since 3rd party devs have to learn about NV at the same time).
     
    Last edited: Sep 20, 2016
  3. Support.2@XOTIC PC

    Support.2@XOTIC PC Company Representative

    Reputations:
    486
    Messages:
    3,148
    Likes Received:
    3,490
    Trophy Points:
    331
    Well said.

    Looks like I'll be sticking with Nvidia for a few more years!
     
  4. tgipier

    tgipier Notebook Deity

    Reputations:
    197
    Messages:
    1,596
    Likes Received:
    1,572
    Trophy Points:
    181
    GCN isnt particularly optimized for clock like pascal and it shows on the polaris as AMD probably tried to clock it far past its efficiency point. The stock rx480 is essentially on par with maxwell I believe. The architecture is old and it shows.

    Nvidia have very little room on GP102. The full shader unit would give it at best 4-5% performance if even considering its only 7% more shaders. With AIB coolers, a 3328/3072 CUDA core 1080 TI with 320bit on 10gb gddr5x at 750-900USD would be enough to take back the crown and allowing some price drop on 1080/1070 if AMD launches Vega above 1070, below 1080.(Thats would my guess for a 4096 shader Vega part) If Vega is delayed or noncompetitive, said 1080 TI might become 1180, with a full 1180 TI to follow if needed.

    AMD ages better since they essentially uses the same architecture. I saw somewhere that polaris have a 18% IPC increase compared GCN 1.0, hardly enough to compared with NVIDIA's efficiency leap on Maxwell and Pascal.
     
  5. 1nstance

    1nstance Notebook Evangelist

    Reputations:
    517
    Messages:
    633
    Likes Received:
    221
    Trophy Points:
    56
    Friend of mine upgraded to an RX480. He has been very happy so far. Works like a charm.
     
    TomJGX likes this.
  6. Mr.Koala

    Mr.Koala Notebook Virtuoso

    Reputations:
    568
    Messages:
    2,307
    Likes Received:
    566
    Trophy Points:
    131
    Arch name is a marketing term. They could call all of their following products GCN for the next 30 years if they want, assuming they survive for that long.
    They don't have much space left on GP102, they do have plenty of space left in fab. A full die is at least 600mm2, especially for TSMC.
    IPC numbers on GPUs don't make much sense out of context. GPUs are all about throughput, and they don't have a continually increasing IPC curve through history like CPUs do. Going by official shader counts Fermi is still the IPC peak in the green lineup. Doesn't mean it's much better than what we have now.

    GCN 1.0, especially Tahiti, pushed a bit too far on shader density, leaving a lot of them scratching their heads waiting for work. IPC increase from that is normal.
     
    Last edited: Sep 20, 2016
    Jarhead likes this.
Thread Status:
Not open for further replies.

Share This Page