PC shuts off during Baldurs Gate?

Jump to latest Follow Reply
P

Pocky Is God

Ars Legatus Legionis
24,268
  • #1
    So...whenever I pan the Camera around my battlefield of death, PC just shuts off. Hard off. No freeze. No Bluescreen. Just off.

    Googling seems to immediately point to a shoddy PSU, but this is a new one, gold rated. And nothing worth noting in Event Viewer that I can find.

    5800X
    6900XT Most Recent drivers
    64GB of 3600 RAM
    Win 11

    Posting here in Audio/Visual since there is obviously a link between this only being during BG3. Happens LESS with DX11, more so with Vulkan. Where in Event Viewer would I double check to see if there is anything being logged by windows?

    Also, monks rock. They are the best.
     
    • Like
    • Hug
    Reactions: JamesB0007, JohnCarter17 and blath
    theevilsharpie

    theevilsharpie

    Ars Scholae Palatinae
    1,199
    Subscriptor++
  • #4
    A PSU being "gold-rated" (assuming you mean 80 Plus Gold) is merely a rating of minimum efficiency under load. It doesn't say anything about the PSU's quality or reliability.

    Modern GPUs are very power-hungry, and can overload the PSU due to split-second spikes in load. Paradoxically, higher-quality PSUs are particularly prone to shutting off under this kind of load due to their various protection mechanisms, whereas cheaper PSUs are more likely to shrug it off (with an associated higher risk of fire and/or damage to equipment if the overload persists long enough).

    If possible, reduce your GPU's clock speed and voltage, as well as it's power limit. A setting of 80% of stock is a good starting point. Then see if the shut offs continue. On Nvidia cards, this can be done with MSI Afterburner. I'm not familiar with overclocking tools on AMD hardware, but they should have similar functionality.

    If this does "fix" the problem, you may want to reach out to your PSU vendor to see if perhaps they have a fix (I believe SeaSonic has replaced shutdown-prone PSUs under warranty). Also, the ATX spec has been updated recently to ATX 3.0, specifically in response to high GPU power draw, so you may want to read up on that.
     
    • Like
    Reactions: JamesB0007, JohnCarter17 and davidtheweb
    I

    IceStorm

    Ars Legatus Legionis
    24,871
    Moderator
  • #7
    Pocky Is God said:
    Googling seems to immediately point to a shoddy PSU, but this is a new one, gold rated.
    Click to expand...
    And yet, you won't tell us which PSU :)

    It doesn't matter what the rating is on the PSU. Cooler Master makes a 850W SFX PSU, but I know someone who routinely saw their system shut down when running a 4090 around the 500W mark. Replaced it with a Corsair SF750, and the problems "went away".

    The 6900 XT is a power hog. I wouldn't be surprised if the PSU can't handle its transient spikes.
     
    • Like
    Reactions: JamesB0007, Astrallionheart and davidtheweb
    P

    Pocky Is God

    Ars Legatus Legionis
    24,268
  • #9
    malor said:
    But you say the power supply is new, and suddenly you're having a problem.

    You need to tell us what the power supply is. We cannot give you an informed answer unless you do.
    Click to expand...

    Honestly I had totally forgot what kind: Corsair RM850x.

    Also, just finished a ~50 minute game of Darktide with nary a hiccup.
     
    • Like
    Reactions: JamesB0007
    U

    Ulf

    Ars Legatus Legionis
    12,551
    Subscriptor++
  • #12
    I did a search on "6900XT" and "transient spike" and apparently there have been people whose computers shut down with 850 watt power supplies. They actually had to use 1,000 watt units to handle it.

    My own RTX 3080 Ti caused shutdowns on my prior power supply (Seasonic 650 Watt), and I said "Screw it" and just dropped in a 1000 watt Corsair unit.
     
    malor

    malor

    Ars Legatus Legionis
    16,093
  • #14
    Nevarre said:
    While this is nominally a GPU troubleshooting thread (possibly related to GPU power spikes), I had a similar issue and it turned out to be RAM.

    If you've got 64 GB, pull half and see if it improves? If not try the other 32 GB?
    Click to expand...
    Particularly with the claim that the PSU was recently replaced, I'm of the opinion that it's probably the issue, despite its good rating.

    RAM seems like a pretty unlikely cause for instant hard shutdowns, but it's an easy thing to test, much easier than swapping supplies. As a just-in-case, the half hour invested to check it might be a good idea. PSU swaps can take a long time.
     
    • Like
    Reactions: Jeff3F
    Nevarre

    Nevarre

    Ars Legatus Legionis
    24,110
  • #16
    I mention it because it seemed a very unlikely root cause in my case... but it was the root cause. The RAM had been stable for a long time prior before causing shutdowns in games.

    Power supply swaps are the worst, and it's very possible that particular unit is faulty, but it would have to be a defective unit. It's properly sized and not old. It could also be the motherboard, but of the three possibilities, testing the RAM is the least pain factor by far. Since Pocky is starting with 64 GB (4 sticks?), going to 32 GB isn't a huge performance regression during testing. If it's a really old AGESA for that motherboard, a BIOS update might not hurt either.
     
    malor

    malor

    Ars Legatus Legionis
    16,093
  • #17
    asbath said:
    I had a similar issue a year ago, but it was the motherboard that was the culprit.

    Motherboard swaps are so much worse!
    Click to expand...
    Yeah, I hope for Pocky's sake that that's not the underlying problem here. Such a PITA to replace those. PSU swaps suck much less. Those are getting easier, too, as 3.5" drives are kind of going away. Fewer cables to mess with.

    I'd have to double-check, but I'm pretty sure I have just three wires connected from my PSU; the two for the motherboard, and one for the GPU. It was a lot harder when there were a bunch of SATA connectors to detach and then reattach.
     
    E

    Exordium01

    Ars Praefectus
    3,977
    Subscriptor
  • #19
    I’ve seen this exact behavior with a power supply. It would happen in Elite: Dangerous while turning towards a star to scoop fuel right after entering into a system.

    Panning is similarly graphically intensive.

    The PSU is working as intended and protecting itself from damage. I’d try a larger PSU before anything else. Corsair may even be willing to work with you on the replacement.
     
    malor

    malor

    Ars Legatus Legionis
    16,093
  • #20
    GKH said:
    My first troubleshooting step would be to use older graphics drivers. Like, 3-6 months old. We live in an age where bad drivers aren't supposed to hard crash, and even if older drivers work it won't prove the newer drivers are bad (could just be stressing your system in a weird way), but always start simple.
    Click to expand...
    BG3 is pretty new, though, and AMD drivers aren't noted for quality, so old drivers could be troublesome.
     
    malor

    malor

    Ars Legatus Legionis
    16,093
  • #22
    theevilsharpie said:
    Bad drivers, faulty RAM, etc., are all things that could cause a machine to lock up, but it's not going to suddenly and immediately power the machine off.
    Click to expand...
    Well, @Nevarre is saying he had exactly that symptom, and it was bad RAM.

    Computers are complex enough now that I'm wary of absolutely ruling out most components as trouble sources. I just try to rank from most to least likely, and least to most difficult to test, and try to navigate a shortest-time path for troubleshooting.
     
    Nevarre

    Nevarre

    Ars Legatus Legionis
    24,110
  • #25
    SnowGhost said:
    RAM can be easily tested using memtets86. It costs nothing but time.
    Click to expand...

    Again, I'm 100% not saying that RAM is Pocky's problem and while also AM4 my system was not identical to his. I did not use an AMD GPU for any of my testing. My system would pass overnight runs of Memtest86/Memtest86+ with zero errors or shutdowns but still exhibit system power off events in games. Likewise extended benchmark/stress test runs would fail to cause a system shutdown and it very rarely happened at the desktop (and when it did, the system was usually idle.) It took forever to diagnose because RAM isn't one of the likely culprits and it passed all the cursory tests. Eventually I threw parts at the problem and it hasn't reoccurred a single time.

    The old RAM in this case with G.skill DDR4-3466 (run at that EXPO profile-- because that's pretty much as good as I could get at the time, (long story). It was 100% stable for a year and a half. Instability didn't coincide with any major hardware changes or UEFI updates or anything like that. I only got to that point after swapping a known-good PSU, getting a new GPU, etc.-- all the low hanging fruit. Logs were unhelpful as everything seemed fine until power-off, and it never threw any BSODs or anything else loggable before shutdown.

    I even tried running the RAM one stick at a time in single channel. That was my clue. The lockups reduced significantly but did not stop, and did not follow one stick or the other. Either one installed could cause a shutdown but the rate--up to few days between shutdowns when gaming vs. at least once daily clued me in to trying new RAM.

    There's a high chance that's not the problem here, but I'm merely mentioning it in case it is because such an unsual situation happened to me.
     
    • Like
    Reactions: malor
    M

    mpat

    Ars Praefectus
    5,951
    Subscriptor
  • #26
    I know that we have all moved on to more complex solutions here, but... You connected all of the possible power connectors, right? All EPS 12V connectors to the motherboard in particular, and no they are not interchangeable with PCIe 8-pin power even though it looks like it at first glance (the keying really only works one way, preventing using EPS 12V in a PCIe 8-pin, so you can absolutely burn out a motherboard here).

    When it comes to reducing the power limit to the GPU, note that AMD allows you to set power limits and voltage settings for specific games in the driver. Also note that the value you set in the driver is the limit for the GPU itself, not including the VRAM. The default for a 6900XT is thus 255W. Reducing this a bit would be a good start. Also check that your motherboard manufacturer isn't automatically overclocking the CPU in the BIOS.

    If you do reduce the power limit on the GPU, causing lower frame rates, you can try to claw that back by undervolting.

    EDIT: You can see the power draw of the GPU in AMD's overlay, or on your phone using the AMD Link app.
     
    • Like
    Reactions: Nevarre
    C

    cerberusTI

    Ars Tribunus Angusticlavius
    6,449
    Subscriptor++
  • #30
    • Like
    Reactions: continuum
    C

    cerberusTI

    Ars Tribunus Angusticlavius
    6,449
    Subscriptor++
  • #31
    malor said:
    I have a vague idea that 5% loss of output per year is about normal, but I have no idea where I learned that, or if it's accurate.
    Click to expand...
    That will certainly be highly variable. You can build a power supply which will last decades reliably, especially if you do not load it heavily compared to its potential output and keep it cool, or use solid capacitors. You can alternately optimize on cost, and getting more than a few years out of it can take a bit of luck.

    Most should be somewhere in between, but you can somewhat tell by the warranty. Six years on a generic power supply is probably about right before it has lost 30% of its capacity (although they may very well account for this).

    Capacitors tend to be rated at 85 or 105 degrees in thousands of hours, however lifetime basically doubles for each ten degrees under the rated temperature it is. When a manufacturer says they are using 105 degree or solid capacitors, they usually mean they have used components which are expected to survive longer than the useful life of the system.

    On the other side of this, it is easy to make a mistake and end up with greatly reduced life.
    en.wikipedia.org

    Capacitor plague - Wikipedia

    en.wikipedia.org en.wikipedia.org
     
    Last edited:
    • Like
    Reactions: malor
    Saikaici

    Saikaici

    Ars Praefectus
    3,357
  • #33
    I've had a few shut down issues over the years.

    1st: 2008 era machine.
    - Bad stick of RAM.

    2nd: 2019 build
    - Bad motherboard

    In order I'd try:
    Test RAM - Memtest86 (Easy to run on one stick while MFG RMA goes through)
    Check for BIOS updates - see if there's relevant updates to memory compatibility/etc
    Underclock GPU
    If underclock is successful, try a different PSU
    Lastly - Replace Mobo
     
    • Like
    Reactions: ACinn
    P

    Pocky Is God

    Ars Legatus Legionis
    24,268
  • #36
    Nope. Still happening. Most recently was in the Grymforge, finally finished that annoying "move these platforms around to get to the waypoint" part, and I had Shadowheart playing maestro on the switches. Camera was flying back to her so I could have her rejoin the party and blammo. PC just shut down. Had to do it all over again since I didn't save in time.

    Memtest86 didn't reveal any ram issues. So either PSU or Motherboard. Which right now I simply don't want to replace. I have too much shit to do.
     
    malor

    malor

    Ars Legatus Legionis
    16,093
  • #39
    NervousEnergy said:
    Yep, and if it's the PSU as expected it's a 5 minute swap as long as you make sure the cables are compatible. Corsair is generally good stuff, though they have a huge range like any other large OEM.
    Click to expand...
    Now that SATA drives are going away, PSU replacements are faster than they were, but I'd plan on at least 20 minutes, probably 30, with the average case, especially if you're doing any cable management.

    IMO, you shouldn't leave the existing cables in place, you should always swap those out, because it's just too easy to make a mistake and plug in something that's not compatible and fry the system, even if you've looked it up and ordered carefully. There's no guarantee that website is correct, and manufacturers change components in manufactured goods without notice. Take the extra time and just replace everything, keeping the old cables with the old supply. IMO, it's too risky to re-use anything.

    It's usually only three or four cables anyway, these days, two motherboard and one or two PCIe, unless you've got an older install with a bunch of drives. It still takes about a half-hour, but it's just a little time-consuming (and a bit annoying), not physically difficult.
     
    R

    Rodge

    Seniorius Lurkius
    5
  • #40
    I had the same symptoms as you a few years ago, whenever I panned in a recent 3d based shooter (Alien:Fireteam), hard off. GeForce GTX 1070 in an Asrock Z77 based board. Could run Furmark for hours with no issues. Older graphics cards worked fine, so I assumed a power draw spike, so upgraded the power supply. No change. At that stage I just moved to Factorio and older games and parked it as a problem for future me*.

    I tried Alien:Fireteam again about 9 months later and instantly cooked the PCI Express lane the 1070 was in. No card will work in that slot now, though onboard graphics still work fine. I'm assuming some capacitor on the motherboard gave something out of spec, but I’ve never had the courage to test the 1070 on another board. It lives in a drawer now, with a little skull and crossbones on it. The Asrock still happily chugs along as a headless media server.

    *timed it right for the GPU mining / supply chain crisis. Yay past me.
     
    You must log in or register to reply here.
    Jump to latest Follow Reply
    Jump to latest Follow Reply