970 memory allocation issue revisited

Discussion in 'Videocards - NVIDIA GeForce' started by alanm, Jan 23, 2015.

Thread Status:
Not open for further replies.
  1. bo3bber

    bo3bber Member

    Messages:
    23
    Likes Received:
    0
    GPU:
    GTX 760 SLI
    Just an observation from a relative newbie to the forum.

    There is a lot of hostility from the old timers here, and I don't understand it.

    I read the last 30 pages of this thread, and there is a lot of belittlement that I would not expect from "Ancient Gurus" or "Master Gurus". Maybe reflect a moment on whether that is how you actually want to be seen.
     
  2. SoloCreep

    SoloCreep Master Guru

    Messages:
    685
    Likes Received:
    12
    GPU:
    RTX 2080 TI
  3. flexy

    flexy Member Guru

    Messages:
    198
    Likes Received:
    3
    GPU:
    Riva 128
    It needed some nerds with self-written tools to "magically" make Nvidia admit that the specs of the card are actually not the specs AS WE WERE TOLD.

    Absolutely...mind blowing.

    I wonder how many hardware manufacturers claim fantasy specs which in reality are not even true.

    I also don't buy that major NV reps didn't know about their OWN FRICKING PRODUCT SPECS....so basically they knowingly released specs/data that were wrong. Lame, lame and even more lame. "Communication mistake" yeah right. Seriously...
     
  4. flexy

    flexy Member Guru

    Messages:
    198
    Likes Received:
    3
    GPU:
    Riva 128
    Unless it hits 3GB....right....
     

  5. HeavyHemi

    HeavyHemi Ancient Guru

    Messages:
    6,954
    Likes Received:
    959
    GPU:
    GTX1080Ti
    Yeah because now that you know it is slightly different on the specs...all of a sudden every benchmark done showing performance results up to and including 4K are fraudulent.
     
  6. Fox2232

    Fox2232 Ancient Guru

    Messages:
    11,809
    Likes Received:
    3,366
    GPU:
    6900XT+AW@240Hz
    Rainbows & Unicorns...
    They always disable same part. If it fits 980 TDP profile, it lives, if it does not, it gets cut down.
    If they have need more 970s in stores, they cut more.

    Only reason they need is profit. They would have it lower if only 980 was at stores.
    And making 2 separate chips 980 + 970 would cost more than making 980 and cutting half of them.
     
  7. HeavyHemi

    HeavyHemi Ancient Guru

    Messages:
    6,954
    Likes Received:
    959
    GPU:
    GTX1080Ti
    No they don't. Different SM's and/or ROP can be disabled. It is not the same on every die. I used to do this for a living.
     
  8. The Phoenix

    The Phoenix Maha Guru

    Messages:
    1,247
    Likes Received:
    53
    GPU:
    Gigabyte RTX 3070
    this thread cracks me up its like saying i was banging my girl an she couldnt feel the last half inch lol omg the worlds gonna end lol,if nvidia can fix this cool if not cool 970 still does well for me and im not easily pleased
     
  9. haz_mat

    haz_mat Master Guru

    Messages:
    243
    Likes Received:
    1
    GPU:
    1070 FE
    You're only giving her 3.5 inches? No wonder....
     
  10. HeavyHemi

    HeavyHemi Ancient Guru

    Messages:
    6,954
    Likes Received:
    959
    GPU:
    GTX1080Ti
    I understand why there is some people being peeved but the outrage is way out of proportion to the actual effects.
     

  11. Fox2232

    Fox2232 Ancient Guru

    Messages:
    11,809
    Likes Received:
    3,366
    GPU:
    6900XT+AW@240Hz
    Draw it for us, even very simply scheme. How you make connections to memory where different cutting of SM/ROPs will have effect to one part of vram, but it has to affect always same part.

    And btw. This I agree with:
     
  12. HeavyHemi

    HeavyHemi Ancient Guru

    Messages:
    6,954
    Likes Received:
    959
    GPU:
    GTX1080Ti
    You should have read the article from Anandtech

    [​IMG]

    In any case, while the crossbar shows how the SMMs and ROP/MC partitions are connected via the crossbar, memory segmentation is really a story about the ROP/MC partitions. The SMMs do not play a role in how GTX 970’s memory is configured – the crossbar abstracts all of that – and the crossbar itself matters only in as much as the ROP/MC partitions are connected to it.

    http://www.anandtech.com/show/8935/...cting-the-specs-exploring-memory-allocation/2

    Edit..sorry for the image size, that is what is available.
     
    Last edited: Jan 27, 2015
  13. The Phoenix

    The Phoenix Maha Guru

    Messages:
    1,247
    Likes Received:
    53
    GPU:
    Gigabyte RTX 3070
    totally agreed
     
  14. Fox2232

    Fox2232 Ancient Guru

    Messages:
    11,809
    Likes Received:
    3,366
    GPU:
    6900XT+AW@240Hz
    Image is beautiful, it tells:
    - We disable 2SM
    - then there is crossbar between them and L2 and memory buses, Which in normal situation delivers full performance
    - then we go extra mile and disable one L2 and have it overridden by another L2
    - and for purpose of this memory crippling to be applicable anywhere we have little redundancy channels between L2-MC
    - and we cut them everywhere where they are not needed

    So, there is only one sane question:
    Why is that L2 disabled?
    a) Because we intentionally cripple memory access for that part.
    b) because we allow 970 to have one bad L2 which we disable.

    If b) is your presumption then my question stays:
    How do you cut/disable different parts while affecting same memory chip?
     
  15. HeavyHemi

    HeavyHemi Ancient Guru

    Messages:
    6,954
    Likes Received:
    959
    GPU:
    GTX1080Ti
    Read the article. Your original question was answered and you were mistaken. Do your own research. You've already got one free lesson.
     

  16. Fox2232

    Fox2232 Ancient Guru

    Messages:
    11,809
    Likes Received:
    3,366
    GPU:
    6900XT+AW@240Hz
    It said nothing about the question I wrote. still disabling different L2 results in crippling different memory chip access.
    But it tells that to make 970, you have to cripple one for sure, otherwise 970 would have 8more ROPs.
    So it states it is architectural flaw and bad side effect. Should have been separated.
    (nothing new here, no reason to rant - just if anyone haven't had enough of it)
     
  17. khanmein

    khanmein Ancient Guru

    Messages:
    1,647
    Likes Received:
    72
    GPU:
    EVGA GTX 1070 SC
    can we sue nvidia regarding this issue?
     
  18. Fox2232

    Fox2232 Ancient Guru

    Messages:
    11,809
    Likes Received:
    3,366
    GPU:
    6900XT+AW@240Hz
    no, but ask them to give you option to have only one address space 4GB and ability to make it normal 3.5+0.5.

    Just for laughs of people having random fps drops by getting what they want.
    But to me it is design flaw and 3.5+0.5 is override to hide it.
     
  19. HeavyHemi

    HeavyHemi Ancient Guru

    Messages:
    6,954
    Likes Received:
    959
    GPU:
    GTX1080Ti
    Your original post was:

    Rainbows & Unicorns...
    They always disable same part. If it fits 980 TDP profile, it lives, if it does not, it gets cut down.
    If they have need more 970s in stores, they cut more.

    Only reason they need is profit. They would have it lower if only 980 was at stores.
    And making 2 separate chips 980 + 970 would cost more than making 980 and cutting half of them.


    You now know that is incorrect. When that didn't work, you tried to change to a different claim. And when that didn't work you just kinda repeated your same mistake. Apparently you still don't quite understand the architecture and the reasons behind the tradeoffs for the GTX 970. If reading that detailed article didn't help, I certainly can't be bother to waste pages of text trying to explain the same thing. Hasta.
     
  20. Spets

    Spets Ancient Guru

    Messages:
    3,141
    Likes Received:
    267
    GPU:
    RTX 3090
    http://www.guru3d.com/news-story/does-the-geforce-gtx-970-have-a-memory-allocation-bug.html

    Same thing was said on numerous sites.

    @khanmein
    It'd be a law suit that won't really go anywhere let alone be worth the costs. Realistically no.
     
Thread Status:
Not open for further replies.

Share This Page