1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

AMD Radeon VII memory speed benefits?

Discussion in 'Videocards - AMD Radeon' started by kevmb1, Feb 3, 2019.

Tags:
  1. kevmb1

    kevmb1 New Member

    Messages:
    2
    Likes Received:
    0
    GPU:
    Both
    Hi all,

    Sorry if this is posted in the wrong section.
    I am by no means an AMD or Nvidia fan. I love them both.
    I just am looking for an answer to this specific question.

    AMD Radeon VII coming with 4096 memory bus, 1Tb/s memory bandwidth. See link below..
    https://www.powercolor.com/product?id=1548302770

    My question: What specific applications or uses would take advantage of that huge memory speed advantage?
    Is it deep learning or cryptomining? or something else?

    My personal situation is more 3D CG, which makes me lean more towards Nvidia.. But I don't want to ignore that huge difference in memory speeds and what possible advantages it may have. (or not have)

    Thanks in advance.
     
    Last edited: Feb 3, 2019
  2. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    15,621
    Likes Received:
    1,617
    GPU:
    Sapphire 5700XT P.
    Compute from my understanding will really love the high-speed low-latency afforded by HBM, going up from 2 stack 2048-bit total to 4 stack 4096-bit total bus and improving the speed and along with that also the bandwidth even further plus an additional 8 GB is going to make this a really useful card although from reading up on it the FP64 performance is not going to be quite the same as the MI50 workstation Radeon GPU this is almost identical to.
    (There's also the Frontier GPU already from when Vega first launched but it's clocked lower.)

    Gaming will see some gains too maybe much thanks to the memory now being less of a bottleneck but compute and work related primarily even if it's somewhat marketed towards gaming, there is Deep Learning and other server tasks too but from my understanding CUDA has a dominance here over OpenCL and is well established so AMD is going to have a bit of a rough time trying to compete against that.

    EDIT: Hmm I do wonder how it could do with rendering related tasks thinking of it, memory should be helpful here too but I don't know how it would stack up against something like the Turing and immediate competitor the what would it be, 2070?

    EDIT: Let's see here Vega 64 competes with the 1080 sometimes a bit faster sometimes less so, bit of a gain from drivers and also stability and other fixes but overall performance well it's not going to be astronomically higher unless the title compared against was utterly broken either driver or app side.

    And from leaks so far VII is around 20 - 25% faster which isn't bad for what it is really, something like a tweaked Vega arch not quite the full 4096 cores but not cut down entirely to Vega 56 levels either and with some other tweaks and improvements.

    So that ends up surpassing the 1080 but probably not the 1080Ti though it might be close or even match it in a few benchmarks and then Turing as the current competing class of cards from NVIDIA and the 2060 variants coming soon but yeah 2070 or around there maybe a bit faster though pricing will also play a role and at 700$ the Vega VII is a bit on the higher end of things though HBM and other bits also drive up production costs.

    Remains to be seen if the hardware in Vega that wasn't entirely utilized or saw further development from AMD will be a thing now though it seems it will be in a different form such as via extensions for Vulkan for things like primitive discard.


    Probably going to clear up more once reviews land starting early next week and more info from AMD too hopefully plus we'll see what the driver situation looks like both how it performs and any potential issues and what can be done over the next couple of months as the driver matures further.
    (2070 should be on a similar performance level I think as the 1080, 2080 pulling ahead a bit closer to 1080Ti levels and then the 2080Ti is in a league of it's own but also with a price tag to match at 1000$ and then around and above that in Euro.)


    EDIT: But I could be wrong, three days and we will have those reviews if I'm not fumbling the date too, that's going to clear a lot of questions up for performance, features and possibly any problems and the software side of things. :)



    EDIT: But the above is more about real-time performance and 3D, it's going to be different with applications and various programs ranging from compute and workstation tasks of all kinds to encoding and media work or rendering and 3D on a professional level via various tools and other programs from modeling to full on rendering of scenes with all kinds of effects and so on.

    I do recall AMD targeting and marketing it towards gamers somewhat but overall it is still more of a work card and workstation with compute tasks as a major focus though then it's running against NVIDIA and their CUDA and established software and industries well involved in using that. Not going to be easy to compete with I'm thinking.

    CUDA also allows NVIDIA's GPU power to be flexible in a very broad number of subjects from AI to graphics such as rendering and then including media and more for leveraging their power in a varied number of tasks.

    Or how to say, making it more complex than it is but competition will be tough and AMD has it's work cut out trying to break into and establish themselves against NVIDIA both for gaming and the professional user segment and applications of all kinds. :)
     
    Last edited: Feb 3, 2019
    Embra, kevmb1, Undying and 1 other person like this.
  3. user1

    user1 Maha Guru

    Messages:
    1,385
    Likes Received:
    445
    GPU:
    hd 6870
    Apart from compute as mentioned above,

    It will improve performance at higher resolutions(particularly above 4k),and large amounts of texture data, if you use huge amounts of memory, the 16gb of hbm + high bandwidth caching ability of vega that allows you to use system memory more effectively, it will probably have an advantage over an equivalent NVIDIA card with less memory,
     
    Embra likes this.
  4. JonasBeckman

    JonasBeckman Ancient Guru

    Messages:
    15,621
    Likes Received:
    1,617
    GPU:
    Sapphire 5700XT P.
    For this in particular though I wonder if NVIDIA might not have the lead here especially with RTX as a early way into ray-tracing acceleration. It might take time to fully establish for real-time use but for work and rendering and offline it's not a concern but can still increase productivity by a lot having hardware acceleration ready as long as the program has support for leveraging the tech NVIDIA provides. That's a big factor.
    (I don't think AMD has DX12 DXR or the Vulkan extensions for ray-tracing support outside of NVIDIA's RTX specific extensions so they can't yet compete at all against that if it's a major point of interest or deciding factor.)

    EDIT: Ah good to have additional posts and more answers on this subject of the strengths and possibly weaknesses of Vega VII especially if they're competing against NVIDIA and not just gaming but also for professional work in various fields. :)
     

  5. GlennB

    GlennB Master Guru

    Messages:
    206
    Likes Received:
    51
    GPU:
    Sapphire Vega 56 EK
    https://www.reddit.com/r/Amd/comments/9xnkwa/dxr_fallback_on_vega_raytracing/

    https://forums.guru3d.com/threads/r...-mods-bios-tweaks.416287/page-48#post-5607107

    These 2 might provide some interesting information on the subject. It seems Vega supports DXR on a hardware level.
     
  6. kevmb1

    kevmb1 New Member

    Messages:
    2
    Likes Received:
    0
    GPU:
    Both
    Wow thanks for your extensive answer. I appreciate it.

    Looking forward to the benchmarks.
     
  7. OnnA

    OnnA Ancient Guru

    Messages:
    9,430
    Likes Received:
    1,916
    GPU:
    Vega 64 XTX LiQuiD
    As i said, BFV has a lot better Reflection Maps than BF1 -> that's why i was confused (and Happy) :D
    But still, truth is needs to be said -> ATI don't have MS DirectX DXR implemented Yet !

    It will be done in time, and it will be done by Compute Units (Vega nCU) in DX12/VLK by Asynchronous compute.
    So -> It will be Done right, just Wait™
     
  8. Chastity

    Chastity Ancient Guru

    Messages:
    1,946
    Likes Received:
    479
    GPU:
    Nitro 390/GTX1070M
    Coming to Windows 10 in April update
     
    Flanker35M, JonasBeckman and OnnA like this.
  9. MaxBlade

    MaxBlade Master Guru

    Messages:
    888
    Likes Received:
    10
    GPU:
    980Ti/970/390/+ more
    It will be nice but.. it WILL come at a cost. What you and everyone else does not have a clue in is how LONG this will be. From everything I read and understand its not happening soon.
     
  10. devastator

    devastator Member Guru

    Messages:
    115
    Likes Received:
    24
    GPU:
    rtx 2070
    all that memory b/width does nothing unless u got a gpu that can use it all but good for future gpus that are more faster then today gpus thats for sure
     

  11. doublee

    doublee Master Guru

    Messages:
    316
    Likes Received:
    0
    GPU:
    GTX 690 with 100/100/500
    It benefits Ethereum mining. The VII is the best mining GPU out there (for ETH) at the moment.
    My Gigabyte 2080 Ti isn't even close.
     
    Last edited: Jun 12, 2019

Share This Page