New Upcoming ATI/AMD GPU's Thread: Leaks, Hopes & Aftermarket GPU's

Discussion in 'Videocards - AMD Radeon' started by OnnA, Jul 9, 2016.

  1. OnnA

    OnnA Ancient Guru

    Messages:
    15,814
    Likes Received:
    4,954
    GPU:
    3080Ti VISION OC
    Read ahead :nerd: and no visual penalty.

    Great news :banana:
     
    Last edited: May 15, 2017
  2. Ryu5uzaku

    Ryu5uzaku Ancient Guru

    Messages:
    7,412
    Likes Received:
    512
    GPU:
    6800 XT
    Might actually be the professional card but still! 1600mhz!
     
  3. OnnA

    OnnA Ancient Guru

    Messages:
    15,814
    Likes Received:
    4,954
    GPU:
    3080Ti VISION OC
    And what is this:


    {0x1002, 0x6860, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},

    {0x1002, 0x6861, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},

    {0x1002, 0x6862, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},

    {0x1002, 0x6863, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},

    {0x1002, 0x6864, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},

    {0x1002, 0x6867, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},

    {0x1002, 0x6868, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},

    {0x1002, 0x686c, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},

    {0x1002, 0x687f, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10}


    :banana:

    UPD. Some industry leaks
    "Friend that works at Amazon germany claims MSI is lowering their price of the flagship MSI 1080Ti Gaming X just to lower their overall stock of 1080Ti ahead of VEGA release just because they know the performance of VEGA and that it will disrupt the high end market.
    (Price at Amazon.de is 780 Eur incuding shipping) "
     
    Last edited: May 15, 2017
  4. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,093
    Likes Received:
    941
    GPU:
    Inno3D RTX 3090
    If it's the professional card that would mean even higher clocks for consumer cards. If we use the simple OC rule of 10% (that even Fiji adhered to), then that card will go up to 1760MHz at least. Which is insane for an AMD GPU.

    The insane GV100 Volta which will never sold to normal consumers unless it's at 7nm, has 15Tflop of compute power at 810mm2. NVIDIA's CEO admitted that they have to go through multiple wafers just to get one properly working.

    IF AMD has managed at least Maxwell-level graphics/triangle performance with Vega, and gotten those frequencies, a theoretically overclocked Vega will reach 14.5Tflop of compute power. That's insane, it's like a year earlier and in a consumer product that is 40% smaller (Vega is 530mm2).
     

  5. Ryu5uzaku

    Ryu5uzaku Ancient Guru

    Messages:
    7,412
    Likes Received:
    512
    GPU:
    6800 XT
    It would be rather insane also I think it is plausible. Fiji while it clocked like ****, some of the chips did go to 1200-1250 which in the end is already 9.8-10.2Tflops. While the only little bit bigger 980 ti didn't get that high in Tflops even with their rather great overclocks. And then again 1080 matches Fiji @ 1200 in Tflops with it's 2ghz clocks.
     
  6. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,093
    Likes Received:
    941
    GPU:
    Inno3D RTX 3090
    The only "hole" in GCN's design is the triangle and graphics memory performance really. Polaris was a great leap towards fixing both, if Vega hit at least moderately some of these targets, things will be interesting.

    We should not though that Volta (GV100) also includes the Tensor cores and some extra hardware to reach that 810mm2. In theory, if that is removed I can see it being around 700mm2, which is still insane. I don't believe we'll see it in a consumer-grade card until 7nm hits. For people forgetting their numbers, TSMC's "12nm" is basically an optimized 16nm node just for NVIDIA, hence there aren't any really great node size reductions at play here. A 3840 shader core Titan is something like 610mm2, so for a 5000+ shader Volta GPU 700mm2 is actually an optimistic target.
     
  7. Denial

    Denial Ancient Guru

    Messages:
    14,013
    Likes Received:
    3,805
    GPU:
    EVGA RTX 3080

    The Titan XP is only 471mm2, not 610. The GP100 is 610 but that comes with FP64. My 1080Ti right now hits 14.5Tflops (2050Mhz OC) in a 471mm2 package and it doesn't have a year of architectural enhancements like Vega does, nor the denser 14nm process.

    5000+ shader GV chip is definitely doable at ~600mm2 - the problem was TDP. At 3840 Nvidia is already at 250w. But Nvidia is claiming Volta has 50% perf/w improvements with FP32 - thus higher core counts and/or clocks should be doable at 250w.
     
    Last edited: May 15, 2017
  8. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,093
    Likes Received:
    941
    GPU:
    Inno3D RTX 3090
    Aaaaand... You're completely right. Now we have to wonder about the actual architectural improvements, and if they all went to improve the clocks or to other parts as well. I have also learned to get NVIDIA's performance claims from presentations with a grain of salt. I'm sure they've managed great strides but I don't believe that the 50% will hold under all situations.
     
  9. OnnA

    OnnA Ancient Guru

    Messages:
    15,814
    Likes Received:
    4,954
    GPU:
    3080Ti VISION OC
    :grad:
    Little Off one ;) or not?

    I just can't stand those Pseudo News about (not literally) HBM_2 is slower than GDDR5x or GDDR6

    OMG

    They show up with some small numbers like ~512GB/s

    HBM_1 was 128GB/s on Stack with 1.2v (Specs / 1.3v on GPU) That gives us ~512BG/s from 4x Stacks
    Basically my Fury have ~700GB/s with OC + V-mod + Timing-Mod
    Also HBM_1 has 4096Bit Bus with 1GT/s (transfers per second)
    HBM_2 is ~200-256GB/s on Stack so, if VEGa will have 16GB then it will pack 4x256GB/s
    and HBM_2 has 4096Bit Bus with 2GT/s ! (transfers per second)
    So it will get us close to ~1.4TB/s !!!! Yes twice more than HBM_1 (with OC,V and Mod it can go even Higher)
    And of course HBM has Next-Gen'ish Response Times (unmatched to date) !!
    And it's High Bandwidth ~4096Bit !!
    And it will "eat" only 15tW !!!
    And it's small

    Why Fiji is so Great at 1440p or 4k???? with "only" 4.2GB gRAM HBMs?

    Of course it's only IMHO :idea:
     
    Last edited: May 17, 2017
  10. Tuga

    Tuga Member

    Messages:
    48
    Likes Received:
    0
    GPU:
    nvidia 670 gtx
    Things are turning very interesting now. I just hope the price isn't 999 or something..
     

  11. Loophole35

    Loophole35 Ancient Guru

    Messages:
    9,797
    Likes Received:
    1,161
    GPU:
    EVGA 1080ti SC
    We are supposed to be getting some official news on Vega and possible "Threadripper" today from AMD.

    source: W C C F
     
  12. Maddness

    Maddness Ancient Guru

    Messages:
    2,198
    Likes Received:
    1,430
    GPU:
    3080 Aorus Xtreme
    That would be great. I thought it was more along the lines of investor news though.
     
  13. OnnA

    OnnA Ancient Guru

    Messages:
    15,814
    Likes Received:
    4,954
    GPU:
    3080Ti VISION OC
    Hmm some new VEGA "Hype News" Hits my inBOX :)
    Grain of salt as always.

    So:

    $399 -> Better than 1080
    $499 -> Better than 1080Ti/Titan
    $599 -> Best until Big Volta GDDR6

    VI.5 ....

    Today is AMD Financial Analyst Day 2017.
    -> http://webcastevents.com/events/amd/analystday/live/player.htm

    UPD. 1
    Confirmed by Mr. Papermaster
    Raja K. will give us runing VEGA Demo !!!!

    UPD. 2
    Sniper Elite 4 PC 1440p

    Vega will pack min. 40% more perf. than Fury-X (Base Vega)
    So:
    Fury-X 71FPS Average
    and
    VEGA ~100FPS Average (IMO it will have ~125FPS)

    and 4K
    VEGA ~71-80FPS

    UPD. 3

    57-80FPS in RoT Tomb Raider

    The final configuration of Vega was finalized some two years ago, and AMD's vision for it was to have a GPU that could plow through 4K resolutions at over 60 frames per second. And Vega has achieved it. Sniper Elite 4 at over 60 FPS on 4K. Afterwards, Raja talked about AMD's High Bandwidth Cache Controller, running Rise of the Tomb Raider, giving the system only 2 GB of system memory, with the HBCC-enabled system delivering more than 3x the minimum frame-rates than the non-HBCC enabled system, something we've seen in the past, though on Deus Ex: mankind Divided. So now we know that wasn't just a single-shot trick.

    [​IMG]

    [​IMG]

    [​IMG]


    ! So we have almost 3x Fury-X performance (in some areas "only" 1.5x) !

    UPD. 4

    Some Tests of VEGA Frontier Edition 8Pin+6Pin PCIe :deflate: so = 225tW <- lol that PowaHHH

    [​IMG]

    [​IMG]

    [​IMG]

    [​IMG]

    VEGA Frontier Edition Retail 8Pin+6Pin PCIe (My note: AMD/ATI shared older renders with Vega Pro that has 2x8PCie ;) )

    [​IMG]
     
    Last edited: May 17, 2017
  14. Loophole35

    Loophole35 Ancient Guru

    Messages:
    9,797
    Likes Received:
    1,161
    GPU:
    EVGA 1080ti SC
    Same source that said Vega would launch in October?
     
  15. OnnA

    OnnA Ancient Guru

    Messages:
    15,814
    Likes Received:
    4,954
    GPU:
    3080Ti VISION OC
    AMD Financial Analyst Day Raja Koduri

    Raja was very forthcoming today on AMD's stage, as he usually is. Raja introduced AMD's new Radeon Vega Frontier

    Edition product that is built to supply deep machine learning as well as being an excellent workstation card as well.

    Designed to handle the most demanding design, rendering, and machine intelligence workloads, this powerful new graphics card excels in:

    Machine learning. Together with AMD’s ROCm open software platform, Radeon Vega Frontier Edition enables developers to tap into the power of Vega for machine learning algorithm development. Frontier Edition delivers more than 50 percent more performance than today’s most powerful machine learning GPUs.

    Advanced visualization. Radon Vega Frontier Edition provides the performance required to drive increasingly large and complex models for real-time visualization, physically-based rendering and virtual reality through the design phase as well as rendering phase of product development.

    VR workloads. Radeon Vega Frontier Edition is ideal for VR content creation supporting AMD’s LiquidVR technology to deliver the gripping content, advanced visual comfort and compatibility needed for next-generation VR experiences

    Revolutionized game design workflows. Radeon Vega Frontier Edition simplifies and accelerates game creation by providing a single GPU optimized for every stage of a game developer’s workflow, from asset production to playtesting and performance optimization.

    Based on the new Vega GPU architecture, Radeon Vega Frontier Edition has been built from the ground up and features Vega’s High Bandwidth Cache Controller, the cornerstone of the world’s most advanced GPU memory architecture. Combined with HBM2, Radeon Vega Frontier Edition expands the capacity of traditional GPU memory to 256TB, allowing users to tackle massive datasets with ease.

    The official Radeon Vega Frontier Edition page is now live

    -> http://radeon.com/en-us/vega-architecture/
     
    Last edited: May 17, 2017

  16. Lane

    Lane Ancient Guru

    Messages:
    6,361
    Likes Received:
    3
    GPU:
    2x HD7970 - EK Waterblock
    Well, memory "bottleneck, was mainly due to the 4GB limitation of HBM1 ( Fiji )... Triangle output ( for tesselation ), i will say it depend the game ( we will see what this new geometry engine do on Vega ). Benchmark with extreme tesselation dont say all. ( if games us extreme level, this will not help ).

    The real advantage has Nvidia ( and who was a big secret ), used since maxwell is the tiled-based rasterization.. Vega will got it too.

    This technic is used a lot on mobile GPUs.

    http://www.anandtech.com/show/10536/nvidia-maxwell-tile-rasterization-analysis

    http://www.realworldtech.com/tile-based-rasterization-nvidia-gpus/

    It accelerate a lot the process of rendering frames, triangle output ( by not computing what is not needed) and have an extreme impact on Power usage. Basically rendering only what is needed ( without having to cull anything ).

    The most funny is nobody was know it was used in Maxwell allready.
     
    Last edited: May 18, 2017
  17. Fox2232

    Fox2232 Ancient Guru

    Messages:
    11,808
    Likes Received:
    3,370
    GPU:
    6900XT+AW@240Hz
    That sound pretty dumb. Did you meant something else than you wrote?
    Because culling IS process which excludes geometry from being further processed.

    "Basically rendering only what is needed" means that all unnecessary objects been culled... "( without having to cull anything )" then sounds pretty stupid.

    Btw, guy in that video has theory. Theories often happen to be wrong. From what I have seen there, I have much better theory. But I'll keep it to myself as I have no need to incite flame.
     
  18. OnnA

    OnnA Ancient Guru

    Messages:
    15,814
    Likes Received:
    4,954
    GPU:
    3080Ti VISION OC
    I have some theories + some leaks too ;)

    So my Leak about SniperElite IV was correct
    (published 30min before R.K. showed this)
    So if we stick only with those 2 Demos: RoT TombRaider & S. Elite 4
    We can clearly see that this Vega has 1080Ti Performance in both games in 4k !

    That's the facts.

    Q. What we don't know?
    A. Is this a Big Vega or Medium/Small one? (It's possible tho)
    A. Is Big Vega for Gaming wil have 16GB HBM_2 ? (2GT/s Vs 1GT/s of HBM1)
    A. It will have 2x8PCie or 1x8+1x6?

    Do You think is possible for us (Fiji owners) to Have HBCC in Relive to turn ON :D (I believe so)

    :nerd: Please give us some nerd decisions
     
    Last edited: May 18, 2017
  19. PrMinisterGR

    PrMinisterGR Ancient Guru

    Messages:
    8,093
    Likes Received:
    941
    GPU:
    Inno3D RTX 3090
    Please say, that's why we have this place, for conversation and flaming each other over indifferent tech companies :D
     
  20. Lane

    Lane Ancient Guru

    Messages:
    6,361
    Likes Received:
    3
    GPU:
    2x HD7970 - EK Waterblock
    I was a bit fast to write, i dont want enter technical explanation, but read just the Anandtech article about it.
     

Share This Page