AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

Discussion in 'Architecture and Products' started by Deleted member 13524, Sep 20, 2016.

  1. entity279

    entity279 Veteran Subscriber

    Love is primitive, so makes sense ;)
     
    Lightman, Despoiler and Digidi like this.
  2. orangpelupa

    orangpelupa Elite Bug Hunter Legend

    what is this "love" word all of you keeps throwing around. it sounds so foreign
     
  3. Grall

    Grall Invisible Member Legend

    Maybe one of the bleeding-edge gaming luggables. ASUS, Razer or such.
     
  4. Unfortunately, that single-channel Lenovo is the only thin-and-light model of the trio.
    The other two are simply low-cost 15"/2Kg solutions.

    So in the end, none of the solutions is high-end.
    Ryzen Mobile's strengths should play best in an Acer Switch Alpha or HP Spectre X2, yet what they bring to the table are just regular 15 inchers that could perfectly take 25W CPUs with discrete GPUs. And within that TDP range, any Core i5 with discrete geforce MX150 will best the Ryzen Mobile.


    Perhaps it has to do with AMD not implementing LPDDR in Ryzen Mobile.
    If AMD fails to get 2-in-1 designs because they only support regular DDR4, it'll be a huge misstep..

    Here's hoping the next Ryzen Mobile comes with LPDDR4E support so they can just double the bandwidth.


    It probably will if the cooling is built for 25W and the TDP is set accordingly. To be honest, if the Surface Pro 4 can implement TDP-up to 25W on their U processors, I fail to see how any laptop at 13.3" and up is incapable of doing the same.

    AMD claims 23% higher performance with mXFR implementations, so I'd say the nominal 15W implementations will get around 1GHz sustained clocks in the iGPU.
    Still, XBone (theoretical) performance at 15W makes one wonder what will be achievable at 7nm. Portable PS4Bone here we go!
     
    el etro and Grall like this.
  5. Bondrewd

    Bondrewd Veteran

    Oh yes, whether good or bad, the shitstorms all over the internet will be worth it.
     
  6. Grall

    Grall Invisible Member Legend

    Eh, remind me again, what does "mXFR" stand for...? :p (Old, senile; pick any 2 as the explanation...)
     
  7. Bondrewd

    Bondrewd Veteran

    mobile eXtended Frequency Range.
     
    Grall likes this.
  8. MDolenc

    MDolenc Regular

    Sorry to be a bit provocative but:
    Are you expecting games to start rendering a bazillion offscreen/backfacing/sub pixel sized triangles?
    I agree this is potentially a very nice feature. But we know way to little about it. And if it's not going to be accessible through APIs (D3D12/Vulkan) it's going to be limited to above mentioned culling cases and to "single pass stereo multi res" rendering hacks for VR (though as I understand there is push to get these standardized).
     
  9. Infinisearch

    Infinisearch Veteran

    IIRC according to Rys its going to be automatically enabled by drivers. i.e. Your PSO's will automatically use primitive shaders instead of the standard pipeline if they can be used. (same for dx11 as well, I think)
    As for what I'm expecting, I'm want them to deliver on there promise of (IIRC)17 primitives per clock instead of 4. As to what this implies, I'm hoping for at least a performance boost when high degrees of tessellation is employed. But in truth I'm hoping for a nice performance boost overall since supposedly a bunch of bottlenecks are avoided and also better resource utilization. (attribute storage according to AMD)

    That was the technical side of me. The consumer/fanboy side of me wants AMD to have a truly competetive product on their hands. They already trade blows with the competition by Nvidia, if primitive shaders and potentially HBCC's non-over allocated memory performance improvements can make AMD's products clear winners I'd be delighted and not mind that Vega came out a year later than Nvidia.
     
    Heinrich4 and Grall like this.
  10. MDolenc

    MDolenc Regular

    The 17 primitives per clock came with a caveat: it should be able to reject that many. AMD was never promising it can render 17 primitives per clock. That's still 4. Hence my question above: are you expecting games to start rendering bazillion offscreen/backfacing/sub pixel sized triangles? Abuse of tesselation Crysis 2 style could certainly produce a bunch of primitives to cull this way. But if primitive covers a sample (singular) it will have to be rendered and that is 4 primitives per clock.
    I know what Rys said and I agree that's probably all we're going to see, but to me the real interesting use cases would be to get direct access to this.
     
  11. Grall

    Grall Invisible Member Legend

    Would it not also be able to discard opaque overdraw surfaces?
     
  12. Infinisearch

    Infinisearch Veteran

    When you say render you mean rasterize, right? I didn't think it would increase that number but I did think it would increase the number of primitives per clock that can be buffered for use. This I think has value from a performance perspective.
    No just hoping AMD's geometry performance improves somewhat.
    It isn't just Crysis 2 though is it? IIRC games have held back on tessellation due to AMD's performance with regards to it.
    I agree there are interesting things to try with explicit primitive shaders but if primitive shaders were designed with AMD's weaker geometry performance in mind I'm sure AMD also had the automatic use of them on their mind from the start.
     
  13. Infinisearch

    Infinisearch Veteran

    No primitive shaders are before rasterization.

    edit - If DSBR were advanced enough it could.
     
    Last edited: Oct 27, 2017
  14. monstercameron

    monstercameron Newcomer

    [​IMG]

    Though not the same chip/specs, banded kestrel was supposed to have a lpddr4 controller.
    This is the SOC I'm more interested in, hopefully for a low cost win tablet or a GPD WIN style product.
     
    el etro likes this.
  15. el etro

    el etro Newcomer

    And Smach-Z, the handheld PC. Smach-Z matters more than it looks like.
     
  16. DuckThor Evil

    DuckThor Evil Legend

    Smach-Z appears to be just a hoax at this point.
     
  17. Frenetic Pony

    Frenetic Pony Regular

    Hoping for DDR4x/LPDDR4x support for the H1 14nm+ refresh. DDR4x support would be fantastic for gaming purposes. Maybe we'll see it alongside the rumored Kestrel SOC mentioned above. Aside, I've been looking for a replacement for my SP3 for over a year now and so far everything disappoints.

    And yeah, I can see a new PS4/Switch hybrid type model coming out say, 2019. Full backwards compatibility in Switch style body/controller combo could be a killer system. If anyone from/related to Sony is reading this I nominate the Playstation Edge as the name.
     
  18. Grall

    Grall Invisible Member Legend

    Well, yes, I must have been confuzzling primitive shader with DSBR. :p Getting old and silly it seems...
     
  19. Digidi

    Digidi Regular

    Vegas geometry performance is not week. You can see here: (must select cards and settings in the drop down menue)
    http://www.pcgameshardware.de/Radeo...3/Tests/Benchmark-Preis-Release-1235445/3/#a2

    You see on Polygoneoutput Vega beats up Titan X Pascal. And Titan X Pascal have 6 Rasterizer, AMD only 4.

    Only when 100% culling is used, than Vega falls behind. That's the key where Primitive shader can help.
     
  20. Infinisearch

    Infinisearch Veteran

    Just to be clear I'm not sure if DSBR does any form of occlusion related tricks.
     
    Grall likes this.
Loading...

Share This Page

Loading...