Vega 64 on multi rail PSU

Discussion in 'Videocards - AMD Radeon' started by thetimelord, Nov 26, 2017.

  1. thetimelord

    thetimelord New Member

    Messages:
    6
    Likes Received:
    0
    GPU:
    Sapphire Vega 64
    Hello,

    Recently I ran into an issue that my newly built PC was shutting down unexpectedly (loosing power) under kinda heavy load. I've noticed that behaviour first while running FurMark tests, but didn't pay much attention since the issue happened only a few times. But recently I've been playing BioShock Infinite, and my PC shut down two times with exactly the same symptoms. So I decided to investigate further.
    First thing that I noticed, that without flipping the switch on the PSU the computer won't boot, so that meant that protection systems were triggered in the PSU itself. OK, that means I have an over-current, likely.
    My RM850i supports Corsair Link, so I decided to play around with FurMark trying to repro the situation while monitoring the parameters (like voltage, etc.). After several attempts I managed to repro the situation. Looking at the power logs, I found nothing interesting. But I noticed that my PSU was running in Multi-Rail configuration. OK, spec says that it has 40A protection for +12V line in multi-rail mode. OK, that means, most likely, I have an over-current on +12 line. I switched to Single-rail mode and repeated exactly the same test. After several test attemps, I finally managed to see 47A spike while launching FurMark test.That's not so common, I didn't see more than 25A in total while running FurMark. Since I was in single-rail mode, the protection wasn't triggered. I was able to repro this spike a couple of more times, and I was not able to repro original issue at all.

    All that means that sometimes my Vega 64 draws more than 40A in spikes, causing the PSU to switch the power off when running in Multi-rail mode. I have no problem running in Single-Rail mode, but still prefer to run in Multi-rail. Right now the card is powered by a single cable. There is an option to use two cables instead, each connected to a distinct slot on the PSU.

    What do you think here?
     
  2. Rich_Guy

    Rich_Guy Ancient Guru

    Messages:
    12,740
    Likes Received:
    723
    GPU:
    MSI 2070S X-Trio
    Welcome to the forums :)

    Ive heard of Vega cards not running on a single cable (that has the 2x 8pins coming off it), so best to use 2x seperate ones.

    I just got my Liquid Cooled 64 a couple of days ago, and im on a single rail PSU, Corsair AX860w, and using 2x seperate cables, had no issues at all so far, runs fine.
     
    Maddness likes this.
  3. thetimelord

    thetimelord New Member

    Messages:
    6
    Likes Received:
    0
    GPU:
    Sapphire Vega 64
    Thank you! :)

    Well, I can't really see any difference between using two separate cables and 2x8p cable when in single-rail configuration, unless there are any internal restrictions on the output to a single port in the PSU itself (I don't really think there are any).
    I'll continue to use one cable in single-rail mode and write back here if there are any issues.

    However, I wonder how can one debug/analyze power issues without any special hardware?
     
  4. yasamoka

    yasamoka Ancient Guru

    Messages:
    4,859
    Likes Received:
    248
    GPU:
    EVGA GTX 1080Ti SC
    Why do you prefer to run in multi-rail mode?
     

  5. user1

    user1 Ancient Guru

    Messages:
    1,855
    Likes Received:
    685
    GPU:
    hd 6870
    If you can figure how to run an 8 pin on rail 1 and the other 8 pin on rail 2, you can probably keep multi rail mode on.

    you should only need a multi meter , if you put load rail 1 , you should only see significant voltage drop on 12v pins connected to rail 1, same for rail 2

    multi rail mode is safer, In the event of catastrophic failure,you'd much rather have 2 lines capped at 40Amps, than 1 line at 80 amps, it can be the difference between a burning smell, and bursting into flames
     
    Last edited: Nov 27, 2017
  6. user1

    user1 Ancient Guru

    Messages:
    1,855
    Likes Received:
    685
    GPU:
    hd 6870
    edit: ignore double postwhoops
     
  7. Athlonite

    Athlonite Maha Guru

    Messages:
    1,325
    Likes Received:
    39
    GPU:
    Pulse RX5700 8GB
    IF in multi rail mode could you not use two 8 pin cables from separate ports on the PSU instead of a single cable from one port that way even at full draw the current wouldn't go over the max allowed per rail
     
  8. thetimelord

    thetimelord New Member

    Messages:
    6
    Likes Received:
    0
    GPU:
    Sapphire Vega 64
    as @user1 said, it's a bit safer, especially when you have lots of power-hungry consumers on different rails. If one of the consumer fails and short-circuits, PSU will react faster, potentially saving other equipment.
     
  9. thetimelord

    thetimelord New Member

    Messages:
    6
    Likes Received:
    0
    GPU:
    Sapphire Vega 64
    Well, there is nothing said about rails in the documentation to the PSU, apart from drill-down by rails and current. But there is a number of available +12V rails there, and it is magically the same as the number of 8Pin connectors on the PSU. Looks like each connector has it's own rail.
    I'll also play around with multimeter when I have free time.
     
  10. thetimelord

    thetimelord New Member

    Messages:
    6
    Likes Received:
    0
    GPU:
    Sapphire Vega 64
    Well, now I have a different issue: random system crashes, but not like the original ones, the system just shuts down like after a bugcheck or something, retaining the power (so I don't need to flip the switch on the PSU to boot up again, the power is present). The crashes are semi-random, they can happen everywhere (while gaming, surfing internet, whatever), with or without load. Most likely it's not power-related, since I ran FurMark + Prime95 for half an hour, no power issues.
    However, I the memory was overclocked @2933 14-14-14-26 with some subtimings also tweaked. Running memory tests for prolonged time didn't find any issues, so I was pretty sure that this memory config is pretty stable.
    Also, there is nothing in Event Viewer, no bugchecks, driver crashes or power failure events.
    The are some issues people describing that looks similar, but not exactly the same.

    I've loosen the timings a bit, now running at 14-14-14-35 with Auto for all other timings, no crashes so far, but who knows...
     

  11. Agent-A01

    Agent-A01 Ancient Guru

    Messages:
    11,501
    Likes Received:
    994
    GPU:
    3080Ti Strix H20

    What are you using to test for errors?

    Memtest86 or such isn't very good.

    At minimum you want to use HCI memtest to 300%; a better option is booting linux and using google's stressapptest, it will find errors much quicker.
     
  12. thetimelord

    thetimelord New Member

    Messages:
    6
    Likes Received:
    0
    GPU:
    Sapphire Vega 64
    I used Google's stressapptest running on WSL, and it found no errors at the old memory config.

    But apparently the issue was not in the memory at all... As usual, blame AMD for drivers :) I was on driver version 17.11.2 when experienced the crash. They released a hotfix, version 17.11.3, which addressed this random hang/crash issue. Now they've released even the next version, 17.11.4, hope it improves stability.
     

Share This Page