AMD: Navi Speculation, Rumours and Discussion [2017-2018]

Discussion in 'Architecture and Products' started by Jawed, Mar 23, 2016.

Tags:
Thread Status:
Not open for further replies.
  1. sonen

    Newcomer

    Joined:
    Jul 13, 2012
    Messages:
    53
    Likes Received:
    33
    8700k/R9 -> Ryzen APU sounds like a pretty dumb idea TBH.

    Certain Vulkan games (DOOM, Wolfenstein II) used to default to Integrated GPU, with no way to change this short of disabling Integrated. This has been fixed in the latest WII patch.
     
  2. ToTTenTranz

    Legend Veteran Subscriber

    Joined:
    Jul 7, 2008
    Messages:
    9,853
    Likes Received:
    4,463
    JEDEC updated the HBM standard to support 2.4 Gbps (from Samsung's Aquabolt production) and 12-hi stacks.

    Not only did they increase the number of stacks but the maximum density per stack seems to have doubled to 16Gbit.

    So 4-stack solutions will now be able to get up to 96 GBytes of total RAM at a whopping 1.23 TB/s.


    If there's a Navi card using HBM2, they won't have to worry about VRAM size for different markets even if it only gets 2 stacks.
     
  3. RecessionCone

    Regular Subscriber

    Joined:
    Feb 27, 2010
    Messages:
    499
    Likes Received:
    177
    They didn’t say anything about cost, though, did they? A lot of product decisions hinge on cost, not technology.
     
  4. ToTTenTranz

    Legend Veteran Subscriber

    Joined:
    Jul 7, 2008
    Messages:
    9,853
    Likes Received:
    4,463
    You don't say...
     
    swaaye likes this.
  5. yuri

    Newcomer

    Joined:
    Jun 2, 2010
    Messages:
    178
    Likes Received:
    147
    Let's hope there is none.

    AMD's attempts to produce a flagship based on an "exotic memory" have pretty much failed: Tiran - canceled, Fiji - a lab sample full of limitations, VEGA10/20 - well... being a Vega. I wish Navi would be more successful than its predecessors.
     
    Heinrich4 and del42sa like this.
  6. Frenetic Pony

    Regular Newcomer

    Joined:
    Nov 12, 2011
    Messages:
    324
    Likes Received:
    84
    64gb+ AMD MI... whatever. 70, 80? Point is 7nm Vega with up to 96gb of memory would could sell for a lot to the right customers. Goodbye being memory bound!

    Can't see it as useful in consumer oriented Navi though. But heck, maybe there'll be a corporate oriented Navi down the line.
     
  7. del42sa

    Newcomer

    Joined:
    Jun 29, 2017
    Messages:
    166
    Likes Received:
    82
    I hope they will learn from Nvidia and will use more common GDDR6 memory for NAVI and leave HBM to HPC and datacenter only
     
    Dangerman likes this.
  8. w0lfram

    Newcomer

    Joined:
    Aug 7, 2017
    Messages:
    156
    Likes Received:
    32
    Why?

    High-end, is high end. And the cost of HBM is going down, not up. Not to mention.. HBM chips are getting faster (ie: 1,500 MHz vs 2,400 MHz), not slower. So I do see a market for it, just not on the cheap GDDR6 level. It all depends on what works best with Infinity Fabric 2.0 and Navi uarch.

    I honestly do believe that AMD achieved unified memory over the IF2.0 and will use Vega2 design using HBM2 @ 2GHz. With the HBMC at the hart of it all.

    *shrugs*
     
    ToTTenTranz likes this.
  9. Rootax

    Veteran Newcomer

    Joined:
    Jan 2, 2006
    Messages:
    1,151
    Likes Received:
    571
    Location:
    France
    Source about HBM price ?
     
  10. Anarchist4000

    Veteran Regular

    Joined:
    May 8, 2004
    Messages:
    1,439
    Likes Received:
    359
    Simple maturity of the tech will bring prices down.

    GDDR6 for mid range cards is fine, but any high end product really should be using HBM to save energy. Not to mention the added bandwidth.
     
    iroboto likes this.
  11. pharma

    Veteran Regular

    Joined:
    Mar 29, 2004
    Messages:
    2,911
    Likes Received:
    1,608
    I only see price drops for the cheaper HBM versions (no die buffer/less TSV's) offering 200 GB/sec. Not likely to find these on high end products. Lower price would also depend on HBM adaption rate and competitive alternative products offering similar benefits.
     
    DavidGraham likes this.
  12. Urian

    Regular

    Joined:
    Aug 23, 2003
    Messages:
    621
    Likes Received:
    55
    But it really exist or is it vaporware?
     
  13. del42sa

    Newcomer

    Joined:
    Jun 29, 2017
    Messages:
    166
    Likes Received:
    82
    it's just a wishful thinking....

    To save energy ? look at the power consumption of Nvidia Pascal/Turing and compare them with Vega. Nvidia doesn't need to use HBM to keep power consumption of their GPUs down and they don't need them to maintain high bandwidth either.

    Don't get me wrong, I have nothing against HBM technology at all, but the technology didn't helped AMD sell more GPUs nor it gave them any advantage over Nvidia. Nvidia quickly realized HBM is more suitable for high margin products while keeping less exotic memory in consumer GPUs maintain their margin high.

    I hope AMD have learned the lesson.
     
  14. itsmydamnation

    Veteran Regular

    Joined:
    Apr 29, 2007
    Messages:
    1,296
    Likes Received:
    395
    Location:
    Australia
    This post is nothing but hand waving and terrible logic/false eqivalances.
     
  15. del42sa

    Newcomer

    Joined:
    Jun 29, 2017
    Messages:
    166
    Likes Received:
    82
    Sure, whatever you wish :)
     
  16. ToTTenTranz

    Legend Veteran Subscriber

    Joined:
    Jul 7, 2008
    Messages:
    9,853
    Likes Received:
    4,463
    So much hate for HBM in this thread.

    HBM1 allowed AMD to make a GPU with a very competitive performance/power ratio compared to Maxwell cards despite using a less efficient chip. HBM2 allowed AMD to make what seems to be the most power-efficient mobile GPU of this generation.

    Vega 10 apparently sold very badly according to some.. I guess I live in an alternate universe where, until recently, all Vega 10 cards were constantly sold out despite their price being about twice their MSRP.
     
  17. del42sa

    Newcomer

    Joined:
    Jun 29, 2017
    Messages:
    166
    Likes Received:
    82
    Re-read seeking alpha transcript again :

    https://forum.beyond3d.com/posts/2046857
     
  18. ToTTenTranz

    Legend Veteran Subscriber

    Joined:
    Jul 7, 2008
    Messages:
    9,853
    Likes Received:
    4,463
    Are you somehow trying to pretend the Vega 64/56 wasn't sold out at twice the MSRP for most of its lifetime because a couple of months ago there was overstock due to the mining crash?

    Are we also allowed to say the GTX1070 was a failure because nvidia is sitting on a huge pile of those cards?

    Convenient post-truths could go either way..
     
    Anarchist4000 and no-X like this.
  19. del42sa

    Newcomer

    Joined:
    Jun 29, 2017
    Messages:
    166
    Likes Received:
    82
    Yes Vega was sold out. But not because it was most successfull GPU in AMD history nor because it uses HBM. Actually it was mining craze who save Vegas and RTG ass...

    GTX1070 was failure ? Well then I agree you trully must live in an alternative universe :)
     
  20. ToTTenTranz

    Legend Veteran Subscriber

    Joined:
    Jul 7, 2008
    Messages:
    9,853
    Likes Received:
    4,463
    You don't really know what made Vega 10 so good at hashing, do you?

    Whoooosh
     
Loading...
Thread Status:
Not open for further replies.

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...