AMD Gives Statement on the PCI-Express Overcurrent Problems

Discussion in 'Frontpage news' started by Hilbert Hagedoorn, Jul 2, 2016.

  1. HeavyHemi

    HeavyHemi Guest

    Messages:
    6,952
    Likes Received:
    960
    GPU:
    GTX1080Ti
    Begging your pardon, but if your basic input is you find this thread pointless and annoying, why are you being forced to participate in it? Oh wait...:)

    It's rather easy in a Kepler or Maxwell BIOS to change the power draw limits per power input. I'm not familiar with recent AMD BIOS modding so I'd have to defer to those who are.
     
    Last edited: Jul 4, 2016
  2. sykozis

    sykozis Ancient Guru

    Messages:
    22,492
    Likes Received:
    1,537
    GPU:
    Asus RX6700XT
    The Asus P7P55-LX is a weak board to start with.

    The damage to the 24pin socket supports your thoughts. Based on the image, it's impossible for the graphics cards to have caused the damage. The graphics card could not have possibly caused high resistance at the 24pin socket.

    The image of the Asus P7P55-LX indicates a PSU, motherboard or connector failure. If the graphics card was at fault, the damage would be localized to the PCIe slot or the graphics card itself. I'm not an electrical engineer, but it is my job to diagnose electrical and electronics failures, as well as make repairs. In 10 years, I've yet to see damage occur at a connector, when the component responsible for the damage was located else-where.

    I'm not saying that there is not a problem with the design or functionality of the RX 480 itself, but the image from bitcointalk.org shows a completely different cause of the failure. The user even admits it was his first ever build, which would lead most of those who have experience building to question why the user is immediately condemning the RX 480 for an obvious failure else where. When it comes to connectors, a failure AT a connector, is generally directly caused by the connection itself.
     
  3. PF Prophet

    PF Prophet Master Guru

    Messages:
    276
    Likes Received:
    31
    GPU:
    Acer A770+1070AmpEx
    im a pc/mac/ect tech, most of the time when u see that kinda damage its because the PSU used was a piece of ****(im being kind) like first gen modular psu's that failed at both ends of the 20+4 pin connector so often, we got so we refused to sell them for 3years, we wouldnt carry anything but super flowerpartially modeular builds because the barrel connectors NEVER failed, and the 24+4/24pin was always hared wired.....in years i only saw 2 of those fail one because the guy stopped the fan and it burned out, the other because the idiot tried to run a pelter/tec off the psu's pci-e rail....(180watt pelter+6pin hacked adapter=burned psu)

    and i agree, if this was a pci-e slot issue....it would show damage on the card or board at the slot(or along traces to the slot, those traces are much to fine to not burn out before the connector if thats what caused the issue...

    any asus board with LX in the name is trash in my exp... :/
     
  4. PF Prophet

    PF Prophet Master Guru

    Messages:
    276
    Likes Received:
    31
    GPU:
    Acer A770+1070AmpEx
    a report from a friend who dosnt do forums or social media, hes got 6 of these cards(hes a tester) in systems *SPAM* back to a qx series c2q chip, he only tests with that rig because amd asked him to, running 2 in cf on it has yet to kill the near top end asus board in that system and he let it loop furmark for a full day when he was at work(forgot it running)

    he also has a first gen amd quadcore system that has had 3dmark run for 6hrs straight on loop no fail.

    a 6 core amd system with a 790xt-ud4p that he did 6 hrs on without fail.

    a first gen i7 980x at 4.6 dual cards and playing games for 6+hrs, and a few hours of haven when it was forgotten when he ran to frys and microcenter.

    the 9370@4.6(under volted like mine!!!) same deal though no trip to sore just started watching sports and forgot it for 3-4hrs set to loop 3dmark. its on a crosshair-v f-z

    hes also got a 4k series 6 core(he couldnt remember the model when we talked on the phone a few min ago he wasnt home) with a top end asus board.
    (he really dosnt like this rig, it just pissed him off to much when he was trying to get a stable oc out of it...)

    he also has a few other boxes he keeps around for testing the only one that wouldnt even post with the amd cards was an old nvidia chipset board that he quit using when he discovered his 5770 he just got wouldnt even post with the board...(bios issue nobody bothered to fix, u could flash a special bios for the 5770 but you lost 7% perf so....)

    he even tested with a 480 chipset board from sapphire that couldnt oc for **** but he kept it for testing... it ran dual cards fine.

    they ALL have quality psu's though not all are top name brands....2 are using old superflower kingwin units infact.
     

  5. CrisanT

    CrisanT Guest

    Weird issue...

    It is weird that everyones attention is drawn away from the performance /price of the card to this crap "overcurrent on the PCI-E".
    I demand the retesting of all 900 series from nVidia and 300 Series from AMD(for the same behaviour) to see if this issue can cause wories or it is a bait trown out by the competition.
     
  6. Noisiv

    Noisiv Ancient Guru

    Messages:
    8,230
    Likes Received:
    1,494
    GPU:
    2070 Super
    yeah either that or just read the ****ing title
     
  7. fantaskarsef

    fantaskarsef Ancient Guru

    Messages:
    15,693
    Likes Received:
    9,572
    GPU:
    4090@H2O
    Up until now, I never heard of any card pulling above PCIe specs from the slot, except if you set it to do so manually. Not sure what you think would come out of such a comprehensive test (that takes more time than it's worth), but I would say the reference 480s will not rank as good as before if they are now locked down to 150W (instead of 166W, which means 10% less). Imagine 10% less fps for the cards and it would change a few people's opinion of the cards.

    I think what they should do is go into the BIOSs of the cards and limit the PCIe draw, and rather risk an overdraw on the power plug, as it wouldn't be more than a few W over specs (and should be good). Problem solved that way, but throttling takes performance out of the card that's actually there.
     
  8. Fox2232

    Fox2232 Guest

    Messages:
    11,808
    Likes Received:
    3,371
    GPU:
    6900XT+AW@240Hz
    RX 480 has GPU supplied by both 6-pin and PCIe. It has half of VRMs for GPU connected to each mentioned.
    And they are switching its operation. AMD can therefore adjust this ratio. That's why some sites using oscilloscope to sync them measured those big spikes. Because they did not properly adjusted for edges of rise and fall.

    Only reason why I would not get reference RX 480 (if I had use for it) is lack of dual vBIOS switch.
     
  9. fantaskarsef

    fantaskarsef Ancient Guru

    Messages:
    15,693
    Likes Received:
    9,572
    GPU:
    4090@H2O
    Of course they are using PCIe + power connector... it's >75W. Pretty much what graphics cards have been doing since they required an extra power plug.

    But if they can adjust this ratio, why don't they simply flash new BIOSs? That's what I don't get. And controlling this behavior via a driver update means anybody could fix it 'at home' right now if they would / could get a hold of the right driver part, correct? I just don't understand why it's a driver related behavior, always thought it was on board (vBIOS).
     
  10. Fox2232

    Fox2232 Guest

    Messages:
    11,808
    Likes Received:
    3,371
    GPU:
    6900XT+AW@240Hz
    AMD sets stuff in vBIOS, but a lot of it can be overridden by software as AMD uses I2C compatible controller. Not best solution for 3rd party software to mess with that since AMD driver already talks with controller.

    And reference card vBIOS flash is something more risky that current situation.

    On one side you have very small chance that someone with damn bad MB without any sort of protection will burn PCIe slot/traces. (Still waiting for pictures of dmg. Because till now there are only claims and people who burned it other way.)
    But give all those amateurs around the world flashing tool to flash single vBIOS card have and you are asking for more trouble.
     

  11. AlmondMan

    AlmondMan Maha Guru

    Messages:
    1,036
    Likes Received:
    345
    GPU:
    7900 XT Reference
    Probably it would have never been found out if AMD had just put a 8 pin on there. But then they wouldn't have been able to do the "150watt TDP" marketing thing.

    I am content in holding off on any purchase till quality aftermarket cards are released and reach a price where they compete with the competition. Asus Strix 970s can be had for 30€ less than a 8gb RX 480 here, and around 15$ more than a 4gb.

    Not that I want a 2 year old card, but I also don't want to buy a new card at more than what an older one of similar performance would cost.

    There's also the issue of the 1060 - if that isn't completely throttled at 1440p by a 192bit bandwidth, and priced unreasonably, then that might be interesting as well!
     
  12. pharma

    pharma Ancient Guru

    Messages:
    2,481
    Likes Received:
    1,180
    GPU:
    Asus Strix GTX 1080
    TomsHardware: July 4,2016
    http://www.tomshardware.de/amd-radeon-rx-480-grafikkarte-leistungsaufnahme-pci-sig,testberichte-242143.html
     
  13. Jak Crow

    Jak Crow New Member

    Messages:
    6
    Likes Received:
    0
    GPU:
    EVGA GTX 970 SC
    The only real issue I've heard was from someone in youtube saying when his rig was running a particular benchmark it would turn off at the same point in the benchmark, when the 480 was used in an old motherboard. Using a recent Asus gaming motherboard with a much higher quality power system had no issues.
     
  14. Athlonite

    Athlonite Maha Guru

    Messages:
    1,358
    Likes Received:
    52
    GPU:
    Pulse RX5700 8GB
    how did you find out what the mobo was it doesn't say in his post
     
  15. fantaskarsef

    fantaskarsef Ancient Guru

    Messages:
    15,693
    Likes Received:
    9,572
    GPU:
    4090@H2O
    Ahh, I never knew that AMD was using I2C compatible controller chips, I wish Nvidia would do that too! (At least, would do now, I think they used to.) That way they can toggle things via drivers, avoiding already shipped cards needing a vBIOS flash is a good idea, and they can anyway change it in cards produced / shipped now or later.

    Makes a lot more sense than I thought, I wasn't aware that their drivers are talking to the power controls.
     

  16. Finisterre

    Finisterre Guest

    Messages:
    28
    Likes Received:
    0
    GPU:
    NV
    Doesn't matter much how the power control chips are controlled when three of them are hardwired to the pci power and three are hardwired to the six pin power connector with a split on the gpu vcore. The extra phase near he six pin connector is also doing nothing. This can be clearly seen in the review:

    http://www.guru3d.com/articles-pages/amd-radeon-r9-rx-480-8gb-review,7.html

    The six power phases are blistering hot. The +1 phase is hardly lighting up the ir cam Hilbert uses.
     
    Last edited: Jul 4, 2016
  17. Noisiv

    Noisiv Ancient Guru

    Messages:
    8,230
    Likes Received:
    1,494
    GPU:
    2070 Super
    Seems that 480 running out of PCI-E spec does cause problems, and the issue can be replicated:

    RX 480 shuts down the A78AX mobo + SuperNOVA EVGA 750W PSU in just about every benchmark, while 980Ti runs fine in the same rig.
    Same RX 480 runs without issues with Asus Z170 mobo.

    https://www.youtube.com/watch?v=rhjC_8ai7QA
     
  18. Finisterre

    Finisterre Guest

    Messages:
    28
    Likes Received:
    0
    GPU:
    NV
    You are right, after re watching the vid it's indeed or almost 50/50
    as it is 3 phases on the pci and 3+1 phases in the molex.

    But the gpu vcore power is split over both and are not interconnected.
    Now the problem is that the 3 phases that are powering a part of the gpu from the pci slot already draws overcurrent. The power is split so it is not possible to lower the pci slot current draw and compensate from the molex.
    :bonk:
     
  19. alanm

    alanm Ancient Guru

    Messages:
    12,233
    Likes Received:
    4,435
    GPU:
    RTX 4080
  20. PF Prophet

    PF Prophet Master Guru

    Messages:
    276
    Likes Received:
    31
    GPU:
    Acer A770+1070AmpEx
    i have used that board in a few builds(frys was selling combos cheap and friends bought them.)

    its kinda a POS if you ask me....even for its day, ALOT of matx boards of that era couldnt supply full pci-e spec power either(personal exp with this and some of the "green" versions of various cards. (version that had passive cooling) they had no pci-e power connector, and the power draw was fine till u tried furmark or the like and let it run for 10-15min...the system would either shut down or crash....

    btw, i have been informed that they also tested an old as dirt abit last gen am3/am3+ board that handled furmark for 45min before he shut it off, its their last top end amd mobo....before the company died off. (damn and i loved abit boards..)
     

Share This Page