AMD Navi GPU

Discussion in 'Video Cards & Monitors' started by Court Jester, Dec 20, 2017.

  1. Court Jester

    Court Jester Member

    Joined:
    Jun 30, 2001
    Messages:
    3,634
    Location:
    Gold Coast
    well there is no Navi thread so its time to start one as Vega is out and gone and is meh Navi leak are starting to appear where it has been spotted in the latest AMD drivers revealing some info about its architecture as this is Raja's last design before he left RTG to join Intel this looks like it may be out sooner than first thought however

    https://wccftech.com/amd-navi-gpu-spotted-in-linux-drivers/

     
    Last edited: Dec 20, 2017
  2. asho444

    asho444 Member

    Joined:
    Feb 7, 2012
    Messages:
    1,870
    Location:
    Sunny Coast
    so looking at a similar concept to infinity fabric as with Ryzen? Could see potential as GPU's have much better memory speeds than CPU's will be interesting to see if its HBM 2/3 or GDDR6
     
  3. OP
    OP
    Court Jester

    Court Jester Member

    Joined:
    Jun 30, 2001
    Messages:
    3,634
    Location:
    Gold Coast
    honestly I think AMD would be foolish to continue down the HBM memory parth
     
  4. Sledge

    Sledge Member

    Joined:
    Aug 22, 2002
    Messages:
    8,587
    Location:
    Adelaide
    Then all that investment would be for nothing
    It will only get better/cheaper as time goes on
     
  5. tunksy

    tunksy Member

    Joined:
    Dec 22, 2010
    Messages:
    380
    Location:
    SA
    They need something big. Vega has failed. Too little too late.

    If they are indeed fusing several smaller gpu dies then they need to be on top of drivers and optimisation.

    And after the investment in hbm they need to stick with it. Still heaps of potential there.
     
  6. darkbastard

    darkbastard Member

    Joined:
    Sep 17, 2004
    Messages:
    862
    Location:
    Duranbah - near Kingsclif
    I can only think that if HBM was not the best way forward it wouldnt be on Nvidia's premier products
     
  7. OP
    OP
    Court Jester

    Court Jester Member

    Joined:
    Jun 30, 2001
    Messages:
    3,634
    Location:
    Gold Coast
    HBM is good but nvidia are using it on the right products not their consumer cards that are meant to be a cheap and mass market. AMD cant supply enough vegas even though they are totally fail at everything but mining and are ridiculously over priced LOL $1200+++ 2 for an asus vega 64 LOL

    using HBM on a mass market gaming card at this point in time was product suicide
     
  8. dr_deathy

    dr_deathy Member

    Joined:
    Apr 27, 2007
    Messages:
    2,790
    Game developers fail not vega.

    https://wccftech.com/amd-rx-vega-gets-22-performance-uplift-wolfenstein-2-latest-patch/

    there is a developer who knows how to make a good engine and make use of modern tech, the 1080Ti is not ahead far enough for the price.

    Funny times when all my mates with 1080Ti's crash all the the in pubg but all us with vegas are all cool, like its a fucking DX11 game, exactly what the 1080Ti was built for HAHAHAHA

    Some of us are into tech, if you want the same old boring shit get nvidia.
     
    Last edited: Dec 21, 2017
  9. OP
    OP
    Court Jester

    Court Jester Member

    Joined:
    Jun 30, 2001
    Messages:
    3,634
    Location:
    Gold Coast
    so what your saying is it is SLOWER than a 1080ti and COSTS MORE than a 1080ti ( custom vega64 vs custom 1080ti )

    clap clap well done amd at failing super hard and choosing hbm
     
  10. Hive

    Hive Member

    Joined:
    Jul 8, 2010
    Messages:
    6,343
    Location:
    AvE
    oh lawdy ayy em dee. GDDR6 is surely going to make HBM irrelevant anyway.
     
  11. ChickenSausage

    ChickenSausage Member

    Joined:
    Feb 3, 2016
    Messages:
    210
    It will be interesting to see if Navi is a steaming pile of failure like Vega is.
     
    Court Jester and Hive like this.
  12. dr_deathy

    dr_deathy Member

    Joined:
    Apr 27, 2007
    Messages:
    2,790
    you have one example, yes that asus is a joke. Vega usually costs less then $800.

    But if you love the cherry pick then go for it.
     
  13. SnooP-WiggleS

    SnooP-WiggleS Member

    Joined:
    Sep 20, 2004
    Messages:
    3,168
    Location:
    Ballarat, Victoria
    Yep, wait for actual stock to arrive, and xfx custom vega to come before judging prices. Either way that Asus Vega price is stupid. But hey they've often tried to get $1500 for a 1080 ti air cooled.

    Unlikely. GDDR has higher write latency than HBM, high end compute will always favour HBM2/3 over GDDR6, lots of random writes/reads at once HBM does better. Also space and power considerations, plenty of applications (eg datacentre) where going smaller means more cards per rack. A nvidia GV100 rack version for example would be much larger with GDDR. While I'm sure it's convenient to diss HBM just because AMD uses it, it's not going anywhere.
    https://www.anandtech.com/show/1136...v100-gpu-and-tesla-v100-accelerator-announced
     
    Hive likes this.
  14. Dragonfly

    Dragonfly Member

    Joined:
    Jan 3, 2003
    Messages:
    4,400
    Location:
    Brisbane
    Yeah that's why Nvidia used HBM for Titan V.
     
  15. OP
    OP
    Court Jester

    Court Jester Member

    Joined:
    Jun 30, 2001
    Messages:
    3,634
    Location:
    Gold Coast
    its not cherry picking when ther is ONLY ONE custom vega 64 card to coose from currently in this country and that is the price for it currently

    yes but that is a compute card not a gaming card like vega. and on compute cards that HBM makes sense but not on a mass market gaming product.
     
    Last edited: Dec 21, 2017
  16. ipv6ready

    ipv6ready Member

    Joined:
    Feb 10, 2014
    Messages:
    2,763
    Location:
    North Sydney
    Currently HBMx is looking like a red herring much like more expensive SLC nand.

    People remember RDram

    Basically cheaper MLC and GDDRx seems more than enough for consumers and it is making SLC/HBM irrelevant vs price/performance

    HBMx and SLC still have its place but seems like more than couple of generation away for the intended consumer SDD or GPU
     
    Last edited: Dec 22, 2017
    Hive likes this.
  17. SnooP-WiggleS

    SnooP-WiggleS Member

    Joined:
    Sep 20, 2004
    Messages:
    3,168
    Location:
    Ballarat, Victoria
    Ahh yes rdram, the rambus based ram that was propriety with the aim of making rambus as much money as possible. Funnily enough being a patent troll and sueing everyone doesn't go so well for core pc technology. Not a relevent comparison IMO. Also RDRAM while having higher bandwidth was much higher latency, also used more power. HBM/HBM2 are JEDEC standards like GDDR/DDR. The only disadvantage on HBM is speculated cost which even now I can't find any actual costs beyond rumours from 6 months ago. I don't see SLC as being a good comparison either, you don't lose half your memory capacity by going HBM2 over GDDR, that alone has made SLC harder and harder to justify as endurance handling got better of MLC and now 3D TLC vnand. SLC is disappearing even from enterprise whereas as everyone is pushing for HBM2 in the higher end compute/AI etc, from google to intel to NEC (and of course nvidia volta) etc. I'd still expect the next AMD midrange to be using GDDR6, but still HBM2 and later HBM3 for Navi high end and beyond.

    Vega is easily a compute card, same chip in the radeon pro line, AMD don't have a separate gaming architecture at the high end unlike nvidia with 1080ti/titanxp (GP102) vs GP100 and now the Volta GV100 (tesla and titan V) which funnily enough happens to spec and perform a lot closer to Vega and prior GCN architectures with Volta typically having minimal or no gains in DX11 vs titan xp but large gains in DX12/vulkan titles where async compute etc can come into play. I point this out partly to show that once you're using vulkan and well coded DX12 the lines blur and a compute focused card works well for gaming as well, but the market is slow to move forward as long as nvidia has DX11 focused cards like pascal.
    https://www.gamersnexus.net/guides/...marks-async-future-is-bright-for-volta/page-2

    On another note I'll be interested to see whether Navi comes out with an AI focused variant special chip full of specialized 'tensor cores' or similar or leaves that to google and nvidia.
     
  18. OP
    OP
    Court Jester

    Court Jester Member

    Joined:
    Jun 30, 2001
    Messages:
    3,634
    Location:
    Gold Coast
    well you should have heard the bleating in the vega thread about how the FE card is not a gaming card and that the RX VEGA will be the gaming card and perform better.

    ----- welll


    --- welll

    -- wellll

    so by that logic rx vega is a gaming card and it is a steaming pile of failure.

    Fe card is the compute card which is also a bigger steaming pile of failure

    navi is yet to be seen but if it is using HMB on the consumer gaming ;roduct it will also be a massive hot pile of steaming failure.

    Raja has jumped ship so he is not smeared with this pile of failure.
     
    Hive likes this.
  19. 426

    426 Member

    Joined:
    May 15, 2008
    Messages:
    233
    Location:
    Melbourne
    The Asus may be pricey, but I don't think its going to end up being a whole lot more than the next AIB equivalent.
    Scorptec had the MSI "airboost" vega 56 in stock for a few minutes. Its just a reference card with more holes in the bracket.
    All sold out at a bargain $870ish.
    Sapphire Nitro 56 on some NZ site for preorder. Bit under a grand. For the 56.

    Pretty sure that when the rest of the customs hit, they'll still be around the 1K mark.

    They sure mine awesomely though...
     
  20. OP
    OP
    Court Jester

    Court Jester Member

    Joined:
    Jun 30, 2001
    Messages:
    3,634
    Location:
    Gold Coast
    lol @ 1k price tag considering you can get a 1080ti for less than 1k AUS currently from Australian sppliers.
     

Share This Page

Advertisement: