Results 1 to 4 of 4
  1. #1
    Joined
    Jan 2004
    Location
    Croatia
    Age
    39
    Posts
    453

    Cool Exciting times ahead for GPUs in 2016... finally some big performance improvements :)

    I'd like to hear your thoughts on what 2016 will bring (skip down to below the line if you're too lazy to read it all)
    It seems that 2016 will finally be the year that we get a nice large performance boost compared to the previous generation. It seems that we are getting EVERYTHING in 2016... and that all aspects of our current GPUs are being upgraded.

    1) For the first time since 2011, we are moving to a new manufacturing process that will bring a number of enhancements.
    Since 2011 we have had the 28nm process, and now we will be moving to the 16/14nm (depending on fab) process. Apart from the simple shrink, another new technology that will improve the process is FinFet, which increases density, amongst other things.
    Straight away, the flagship GPU dies will go from about 8-9 billion transistors to 16-18 billion in just the first generation of 16/14nm FinFet.

    2) We have been using SD RAM (or variants of it) since the 1990s, and now, we have finally found a worth successor to increase performance and reduce latency. HBM2 should give the new flagship GPUs from nVidia and AMD a memory bandwidth of 1TB/s... which is huge compared to most of today's cards. It seems the 4096 bit memory bus is split up into four 1024 bit pieces, each of which will give 256GB/s of bandwidth.

    3) Both nVidia and AMD have promised a whole new GPU architechture, so we should not be seeing the same GPUs as we have now, tweaked with more transistors, but something entirely new.

    4) DirectX 12 should give some nice improvements to what is possible in games.

    5) Sooner or later, it seems that nVidia will release nvlink, a new interconnect that will augment, and maybe eventually replace PCI-express and improve speed for compute-heavy workloads. It seems that for now, the nvlink will only be to connect graphics cards directly to each-other, so that this data doesnt have to travel over the PCI-express interconnect.
    It seems that AMD is working on some sort of new interconnect too, but I have not been able to find much information on it.

    So there we have it... we seem to be getting everything = process shrink, faster new RAM standard, new GPU architechtures, a new directx version with native games coming, and new interconnects.

    Hopefully we should see a giant leap in gaming performance due to all the above advancements, especially for the cheaper cards that will end up getting HBM memory on them eventually. Even the smallest and cheapest implementation of this would be a single 1024 bit bus giving 256GB/s of memory bandwidth, which was 'flagship level' bandwidth until recently.

    ____________________________________________________________________________________________________

    I expect at least a 2x performance improvement across the board, at each price level, from this GPU generation to the next.

    What are your thoughts?
    Last edited by Mr_Speedy; 09-17-2015 at 06:58 PM.

  2. #2
    Joined
    Sep 2015
    Posts
    138

    Re: Exciting times ahead for GPUs in 2016... finally some big performance improvement

    Does nvidia have anything to compete with HBM?

  3. #3
    Joined
    Jan 2004
    Location
    Croatia
    Age
    39
    Posts
    453

    Re: Exciting times ahead for GPUs in 2016... finally some big performance improvement

    Both nVidia and AMD will use HBM 2 in their 2016 flagship graphics cards.
    Currently, AMD is using HBM 1 in their Fury X card.

    Because AMD helped develop HBM memory, they will have 'priority access' to HBM 2 memory, and should therefore be first to market with HBM 2 products... but nVidia will get it too. I have no idea how long this 'priority access' will give AMD an advantage before nVidia ships products with it.

  4. #4
    Joined
    Mar 2002
    Posts
    553

    Re: Exciting times ahead for GPUs in 2016... finally some big performance improvement

    Quote Originally Posted by Mr_Speedy View Post
    Both nVidia and AMD will use HBM 2 in their 2016 flagship graphics cards.
    Currently, AMD is using HBM 1 in their Fury X card.

    Because AMD helped develop HBM memory, they will have 'priority access' to HBM 2 memory, and should therefore be first to market with HBM 2 products... but nVidia will get it too. I have no idea how long this 'priority access' will give AMD an advantage before nVidia ships products with it.
    The question may well be whether "priority access" will even matter by then.

    AMD currently rocking an 18% market share and bound to be headed South of that by 2016.

    Will AMD be able to finish their next design before Pascal with their diminished R&D budget and staff?

    If AMD buys all the HBM2 they need will there be a lot left for NV because AMD just doesn't sell many parts?

    What does "priority access" mean anyway?

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •