My analysis of the exploding 3090's

Discussion in 'Frontpage news' started by David Lake, Jul 24, 2021.

  1. David Lake

    David Lake Master Guru

    Messages:
    739
    Likes Received:
    37
    GPU:
    Titan V watercooled
    I cant say how this is happening but I can say what is happening with my experience in electronics...
    When a MOSFET fails it fails closed circuit and can explode VIOLENTLY as some of you may know so on the card in this image there are surface mount fuses designed to pop to prevent a more catastrophic failure and to do so it must get hot as you can see, the reference designator or refdes for a fuse on a PCB is F followed by a number as you can see by the identical component F6503 circled in the top of the image with the blown fuse circled below it.

    3090 PCB.jpg

    Keep in mind these fuses are only designed to prevent a far more catastrophic failure and not to be replaced.
     
  2. Astyanax

    Astyanax Ancient Guru

    Messages:
    11,380
    Likes Received:
    4,272
    GPU:
    GTX 1080ti
  3. MonstroMart

    MonstroMart Master Guru

    Messages:
    999
    Likes Received:
    466
    GPU:
    RX 6800 Red Dragon
    Can't speak for the hardware since i've been out of touch for many years. I do have my computer engineer degree but i got it in 2001 (started in 97) and i've been programming, building PC and running personal servers since then. But i see lot of people blaming Amazon for this. I guess it's popular to blame Amazon. But i've been a dev for 24 years. Did assembly, C, C++ and dozen of other languages. Did some "legal" wdasm work back in windows 95 days. I did crash workstations often in the past. Was quite common doing assembly and even working with pointers and memory allocation in C. I never ever killed a machine despite running some more than questionable in development pieces of software and wdasm work. Honestly in 2021 a piece of software should not in any case be able to brick a card unless it's doing something intentionally malicious. As an experienced dev myself i find it very misinformed to actively blame the devs of the game for this. Since it looks like it's happening to this card only it's more than extremely likely to be a bios or hardware/design problem. It's a game in beta maybe it's doing some weird thing not happening in other games. Beta software doing weird thing is not uncommon at all. Anything in beta will have bugs and coding mistakes.
     
    Brasky and Airbud like this.
  4. tsunami231

    tsunami231 Ancient Guru

    Messages:
    12,120
    Likes Received:
    929
    GPU:
    EVGA 1070Ti Black
    I like other I fail to see how a "uncaped" FPS can kill card even if it doing "2000" fps, EVGA is replace there 3090 card that killed by the beta, which odd in self, make me thing it more of something hardware design in them. then again I also think we will never know the truth either
     

  5. David Lake

    David Lake Master Guru

    Messages:
    739
    Likes Received:
    37
    GPU:
    Titan V watercooled
    There is no way the extreme framerate itself would cause the issue, it has to be a fault with the power delivery, if windows has a security hole you don't blame the people who exploit it you blame the security hole, its the same thing with hardware failures such as this.
     
  6. Neo Cyrus

    Neo Cyrus Ancient Guru

    Messages:
    9,929
    Likes Received:
    724
    GPU:
    Asus TUF 3080 OC
    Software should never be able to cause direct damage like that. It's obviously a faulty hardware design.
     
    FatBoyNL likes this.
  7. Airbud

    Airbud Master Guru

    Messages:
    777
    Likes Received:
    1,264
    GPU:
    PNY GTX 1060 XLR8
    @David Lake .Thanks!....something to think about...
     
  8. Reddoguk

    Reddoguk Ancient Guru

    Messages:
    2,192
    Likes Received:
    287
    GPU:
    RTX3090 GB GamingOC
    When i look at that picture as someone relatively knowledgeable about PCBs and electronics i see a short between the PCB copper layers but i have no idea if that was the cause or if that was in the aftermath of something else going first.

    I don't know the exact number of copper layers in this gfx card but it's probably high number and these must be forced very close to one another and sometimes heat plus high power draw = high amps then electrons start leaking through the layers and an arc is formed and then it's game over for that card because that arc can detonate components.

    Last time i seen a card like that it was mine and it was struck by a massive lightning bolt which killed every single thing in my pc. My HDDs and gfx card looked simular to this.
     
    Last edited: Jul 25, 2021
  9. The Laughing Ma

    The Laughing Ma Ancient Guru

    Messages:
    4,427
    Likes Received:
    880
    GPU:
    Gigabyte 2070 Super
    Perhaps but it still begs the question why is it this piece of software that is exposing this flaw, what's unique about it?
     
  10. sykozis

    sykozis Ancient Guru

    Messages:
    21,934
    Likes Received:
    1,124
    GPU:
    MSI RX5700
    Unless eVGA can prove that the game devs or users are solely to blame, they're required by law to honor their product warranty.

    When cards were damaged while FurMark was running, FurMark was blamed..... So, I guess FurMark wasn't the problem at all? Was just a lot of faulty cards that people didn't want to acknowledge....

    Funny how years ago, FurMark was blamed for cards failing.....but now it's deemed entirely impossible for software to do anything that can result in damage....

    Not disagreeing that it's a hardware problem, just making a point. I fully expect the issue is with the design of the cards themselves. Especially since it seems to largely be an issue with eVGA cards. There's still no reason to have uncapped framerates on menus.
     
    Airbud and FatBoyNL like this.

  11. Astyanax

    Astyanax Ancient Guru

    Messages:
    11,380
    Likes Received:
    4,272
    GPU:
    GTX 1080ti
    Furmark was only exposing the terrible power delivery on Fermi gpu's at the time.
     
  12. David Lake

    David Lake Master Guru

    Messages:
    739
    Likes Received:
    37
    GPU:
    Titan V watercooled
    I think the problem with furmark is nVIDIA never anticipated the current draw from such an application, this seems like a similar but more complex issue.
     
  13. sykozis

    sykozis Ancient Guru

    Messages:
    21,934
    Likes Received:
    1,124
    GPU:
    MSI RX5700
    It would seem, in that regard, that this new game from Amazon is essentially doing the same for the eVGA 3090..... A weak power delivery system will be subject to failure at the hand of any application capable of stressing it sufficiently.
    Personally, I subjected 2 Fermi cards to FurMark with no issues. Granted, they were GTX460 and GTX560Ti..... I still miss that 560Ti.....thing overclocked like crazy. Was also the last graphics card I actually overclocked. But I digress.... If eVGA skimped out on the power delivery, and this new game is coded in such a way that it stresses the power delivery system similar to what FurMark did, it would seem eVGA is solely at fault here.

    That didn't stop FurMark from being blamed for graphics card failures....
     
  14. MonstroMart

    MonstroMart Master Guru

    Messages:
    999
    Likes Received:
    466
    GPU:
    RX 6800 Red Dragon
    It's a popular beta and the RTX 3090 has not been around for that long? It's currently the 3rd more streamed game on Twitch despite being a beta opened to people who pre-ordered the game only. Before a couple of weeks ago it was almost impossible to buy a RTX 3090 they are just starting to show up online. Beta have bugs and coding mistakes. Nothing wrong about it. Before something goes into production there's always bugs it's inevitable that's why you do an alpha and beta. It's very well possible that this game has a coding mistake that creates unusual spike loads rarely seen. There's nothing special about that since it's a beta. Looking at bullzoid video it really looks like it is a power delivery problem. It's also been reported that the card goes over the set power limit by a significant margin. The card should never allow an unsafe operating state to happen while simply playing a game specially since it's a 2 thousands $ card.
     
    Last edited: Jul 25, 2021
    Airbud likes this.
  15. tsunami231

    tsunami231 Ancient Guru

    Messages:
    12,120
    Likes Received:
    929
    GPU:
    EVGA 1070Ti Black
    I know it just, then again I also never seen Game being blamed for killing GPU either special when it uncapped, People have been runing uncapped FPS for games decades, what make this new MMO from Amazon any diffrent then countless other games ran uncapped. the amount power drawn should be the same,, cause gpu is just gona draw as much FPS as possible with its own limits

    Software should general never able to kill card but that dont mean much wee see drivers kill cards which consider software but that extreme rare
     

  16. Neo Cyrus

    Neo Cyrus Ancient Guru

    Messages:
    9,929
    Likes Received:
    724
    GPU:
    Asus TUF 3080 OC
    I don't have any clear or verifiable info but now I'm hearing that it might be due to a bug in nVidia's drivers that causes the hardware protections to not work properly, and supposedly this bug has gone unfixed for quite some time. If anyone has a good source or can verify any that, let us know.

    I don't know why I forgot nVidia's drivers exist and assumed it 100% had to a hardware problem, I should never underestimate nVidia's drivers' power to destroy video cards. But to be clear, again, I don't know if that's actually true or just a rumour.
     
  17. tsunami231

    tsunami231 Ancient Guru

    Messages:
    12,120
    Likes Received:
    929
    GPU:
    EVGA 1070Ti Black
    which consider software, if nvidia is to blame it will get interesting, but not first time Nvidia drivers cause the death of cards, I remember right 10+ years ago there was driver bug that kept fans from spinning that killed cards, and it was one the FEW drivers Nvidia pulled so it could not be download, atlest from them.

    How drivers can by pass hardware protections is beyond me though, I also not surprised by anything that happen in computers anymore.
     
  18. Astyanax

    Astyanax Ancient Guru

    Messages:
    11,380
    Likes Received:
    4,272
    GPU:
    GTX 1080ti
    it has nothing to do with nvidia's driver.
     
    Valken and DanielRX like this.
  19. alanm

    alanm Ancient Guru

    Messages:
    10,496
    Likes Received:
    2,614
    GPU:
    Asus 2080 Dual OC
    Maybe not in this case, but drivers without adequate protection measures have resulted in dead cards in the past (GTX 590). Although user-caused by over-volting, should have had in-built protection in drivers or bios from the start. Nvidia released driver update to address that problem.
     
  20. metagamer

    metagamer Ancient Guru

    Messages:
    2,231
    Likes Received:
    905
    GPU:
    Palit GameRock 2080
    Was it drivers or was it graphics cards that lacked OCP measures that were at fault?
     

Share This Page