More details on AMD graphics refresh for 2015 and new tech in 2016.

Discussion in 'AMD Flavor' started by cageymaru, Dec 10, 2014.

  1. PRIME1

    PRIME1 2[H]4U

    Messages:
    3,942
    Joined:
    Feb 4, 2004
    It won't be twice as fast as a 290x on 28nm. My guess is it will be around the 980, otherwise there would be no reason to push 300w. This will be an expensive chip for them to make. Hopefully it's a lot faster than the 980, otherwise NVIDIA will just release "Big Maxwell" as Titan II for $1000.

    I just don't see AMD getting a lot more from their architecture on 28nm.
     
  2. Dayaks

    Dayaks [H]ardness Supreme

    Messages:
    7,206
    Joined:
    Feb 22, 2012
    This, I'd love a 1 kW card if I can stay away from multi-GPU!!
     
  3. ebduncan

    ebduncan [H]ard|Gawd

    Messages:
    2,005
    Joined:
    Feb 1, 2008

    They are using a new process 28nm from global foundries. So its not the same, I posted a side showing the differences earlier in the thread.

    HBM memory alone triples the memory bandwidth. Twice as fast is very possible compared to the 290x, may be even more. Think of it has the next big leap in performance. Amd is going to wow some people this go around, I think. As always competition is always a good thing for the buyer.
     
  4. The Mac

    The Mac [H]ardness Supreme

    Messages:
    4,492
    Joined:
    Feb 14, 2011
    Its supposed to be on GloFo's High Performance 28nm process.
     
  5. Dayaks

    Dayaks [H]ardness Supreme

    Messages:
    7,206
    Joined:
    Feb 22, 2012
    Except current bandwidth at 4GB isn't a bottle neck?
     
  6. The Mac

    The Mac [H]ardness Supreme

    Messages:
    4,492
    Joined:
    Feb 14, 2011
    Texture pop-in would like a word with you...
     
  7. ebduncan

    ebduncan [H]ard|Gawd

    Messages:
    2,005
    Joined:
    Feb 1, 2008
    It is bottle necked. Memory bandwidth is very important to graphics cards. Which is why every major leap in performance came from using a new memory type or much larger memory bus. The reason you think it is not a bottle neck is because we have been stuck using lower resolutions until 4k and higher are the norm.
     
  8. Dayaks

    Dayaks [H]ardness Supreme

    Messages:
    7,206
    Joined:
    Feb 22, 2012
    The reason I say this is because I increase memory bandwidth by 10% I gain basically nothing (vsync unlocked). I'm locked at 60 FPS though, I supposed for the 144 Hz folk it will matter when you're utilizing above 2GB VRAM.i calculate 144 FPS at 2GB and linearly 72 FPS at 4GB for a 980. My mistake, forgot about those higher Hz monitors.

    8GB VRAM it will def come in handy!
     
  9. notarat

    notarat [H]ard|Gawd

    Messages:
    1,703
    Joined:
    Mar 28, 2010
    Because it's okay for GT980 to be 28 but not AMD. :rolleyes:
     
  10. Creig

    Creig Gawd

    Messages:
    786
    Joined:
    Sep 24, 2004
    Yes, well, you also said "no one outside ATI fanatics...see AMD continuing the high end fight beyond the R600" and "AMD does not giving a flying fig about the high end video card market."

    So I think we'd all rather just wait for the benchmarks rather than relying on your predictions. Nostradamus you're not.
     
  11. Rizen

    Rizen [H]ardForum Junkie

    Messages:
    9,220
    Joined:
    Jul 16, 2000
    They switched to GlobalFoundries 28nm process which is reportedly the best 28nm process available, better than TSMC. http://hexus.net/tech/news/industry/78757-amd-switching-28nm-process-globalfoundries-2015/

    I am not 100% positive - and this is my own conjecture, nothing confirmed - but I am pretty confident this will be for their GPUs.

    With GloFo 28nm they should get higher frequencies, combined with the updated GCN architecture which should be at least on par with the 1.2 revision used in Tonga, and it should be a notably faster chip at the very least.
     
  12. Liger88

    Liger88 2[H]4U

    Messages:
    2,657
    Joined:
    Feb 14, 2012

    Speaking of TSMC and all the hype that everything was getting skipped until 16nm in 2015, it looks like once more TSMC proves to fail at being able to keep up.

    http://wccftech.com/tsmcs-16nm-finfet-faces-delays-qualcomm-jumps-ship-samsung/

    A already 4-6 month delay is now compounded by yet another 3-6 months further. AMD/Nvidia are either going to have to look at 20nm as an option (given 20nm is still considered a rumor by some) or we're going to face nearly another 2 years on 28nm with stagnation. I'm anxious to see how this plays out.
     
  13. LordEC911

    LordEC911 [H]ard|Gawd

    Messages:
    1,461
    Joined:
    Jul 7, 2013
    That is crazy, since the explicitly stated that they were 95% of the way to volume production back in November and since it uses much of the same tools as 20SOC.
     
  14. Rizen

    Rizen [H]ardForum Junkie

    Messages:
    9,220
    Joined:
    Jul 16, 2000
    Well if AMD is transitioning to GlobalFoundries we may see a 14nm FinFET GPU from them before TSMC is up and running, since Samsung has apparently achieved yield breakthroughs and is ramping up production. GlobalFoundries licensed 14nm FinFET tech from Samsung.

    No idea what NVIDIA will do. AMD has been building APUs at GF so I imagine that they have already done some of the work getting GCN production running on GF fabs. NVIDIA could presumably go with Samsung or GF too, but who knows how long it would take to switch.
     
  15. Warstopper

    Warstopper [H]Lite

    Messages:
    67
    Joined:
    Nov 14, 2013
    Nvidia will never use GF, BUT, they could potentially work with Intel. I do not think Samsung is an option for them either. AMD is in a far better position when it comes to using different foundries.
     
  16. Shlunky11

    Shlunky11 n00b

    Messages:
    27
    Joined:
    Dec 13, 2014


    While that is technically possible, I would be SHOCKED to see that happen. There would be a good bit of work to get an Intel fab to run the process for nVidia chips.

    Have you read something somewhere that states this is likely to take place?
    Just curious, as I haven't been all over this as much as I usually am.

    ยง
     
  17. fixedmy7970

    fixedmy7970 Gawd

    Messages:
    524
    Joined:
    Jul 20, 2013
    why would intel let nvvidia, a competitor use their fab? (lol @ forum filters, NVidia must be caps but amd doesn't?)
     
  18. mesyn191

    mesyn191 2[H]4U

    Messages:
    2,983
    Joined:
    Jun 28, 2004
    After 90nm node the difficulty in shrinking processes started to go through the roof and ever since then TSMC has been over promising and under delivering while also being late on their tech.

    Don't expect it to get better, expect it to get worse as time goes on.
     
  19. LordEC911

    LordEC911 [H]ard|Gawd

    Messages:
    1,461
    Joined:
    Jul 7, 2013
    Most definitely.

    I'm so surprised because either a big issue came up in the last 30-45days with 16FinFet or TSMC was openly lying to analysts and shareholders for the last 6 months saying they were ahead of their roadmap.

    EUV is going to be a big issue for everyone.
     
  20. ccityinstaller

    ccityinstaller [H]ardness Supreme

    Messages:
    4,229
    Joined:
    Feb 23, 2007
    I am in the camp that believes it is the latter, as in they were most likely told by engineering that there were issues and marketing replied "Well is there is ANY chance we can get it out in the announced time frame?

    To which engineering replied "Well miracles are possible /s"

    Then marketing said "YES!!! I KNEW IT!! We have everything under control, if there is a chance to get back on track it will be done so no need to worry!":rolleyes::rolleyes::rolleyes:
     
  21. Flopper

    Flopper [H]ard|Gawd

    Messages:
    1,642
    Joined:
    Nov 15, 2007
    No pressure to fix it either :)