3

I've started playing some newer games lately and encountered frequent, intermittent crashes printing errors such as DXGI_ERROR_DEVICE_HUNG, DXGI_ERROR_DEVICE_REMOVED, or similar. Researching reduced the issue to these possibilities:

  • faulty graphics card,
  • underpowered PSU.

My question is: how do I differentiate between the two? Is there some diagnostic that logs a "failure due to insufficient power" event for NVIDIA cards, or something to that effect?

The card is based on an NVIDIA GTX 6xx series chipset, and the system in question is Windows 7 64-bit (dual boot with Linux, in case that's required for performing the troubleshooting).

Notes:

I'm purposefully omitting the rest of the configuration, since I'd like the answer to be as general as possible - for the benefit of the community. If you think you have a solution that's still general enough and absolutely needs some more info, drop a comment.

Underclocking is not a viable solution (since it doesn't differentiate between the two conditions), and, according to the specs of the card and the PSU, the wattage provided should be sufficient (albeit with only a narrow overhead).

For completeness, here's a graph of the cards params just after the crash. Each label describes the graph below it, topmost graph is temperature in Celsius. The spike in GPU Usage and the drop in Core Clock frequency corresponds to a crash:

Graphics card stats during normal operation and a crash

Improve this question
2
  • For even more complete completeness and to save everyone's time: I did try other possible solutions, including: tweaking PhysX settings, tweaking power management settings (both for the card and for the system), bringing everything up to date (OS, graphic card drivers, motherboard BIOS, graphic card BIOS, the games themselves), shutting down all other programs, running the DirectX installer, running the DirectX diagnostics, physically reinstalling the card, and, obviously, checking whether the graphics card overheats.
    – mikołak
    Mar 4, 2014 at 21:32
  • I am having the same issue with my RTX 2080 ti, I suspect that I have an underloaded supply, i will replace it and see if the problem persists will post my findings here. my specs: 9700k with RTX 2080 ti on a 500w psu, 1 SSD, 1 SATA drive, 3 fans, 1 DVD writer
    – Ahmad Karim
    Mar 7, 2019 at 10:48

2 Answers 2

Reset to default
2

If you use a no-name PSU, this is probably the source of your problem. Even if it is sometimes the case with known branded PSU, no-name generally counts all the power-line (-5, -12, 3.3, 5, 5VSB, 12V) for their total wattage, where only the +12V is really used by a graphic card (if it has an external power connector) and CPU (with a P4 connector). Moreover the 3.3 and/or 5V often deliver high ampere current, whereas the +12V is lower. (already seen a PSU with useless 40A on +3.3V and barely 18A on +12V) No-name PSU can also have undersized components, making them unable to deliver more than 80% of the total wattage correctly.

Knowing all the hardware specs of your computer would already help to know if the PSU should be sufficient or not.

One thing you can try though : block the CPU multiplier to a lower value, undervolt it so it would consume a bit less current, but this will not affect directly the graphic card. Obviously you will get lower performances in-game.

You can try the same with the chipset, but make sure the PCIe link frequency is fixed at 100Mhz.

Finally, some graphic cards own an onboard LED which indicates failure in powering correctly the card. (my Sapphire Radeon HD4850 has one for exemple but here we're talking about NVidia)

Improve this answer
3
  • 1
    I purposefully omitted most of the configuration, because I'd prefer for the answers to be as general as possible - I'd like this QA to be useful to not only myself, but any other poor sod that encounters the same problem ;). For the record, the PSU (and other components) are branded, although unfortunately your query has not surprised me at all. The LED remark is useful, I'll try to hunt for one, thanks.
    – mikołak
    Mar 4, 2014 at 22:36
  • Okay, like I said you should also try to reduce the CPU usage too, although it would not do much difference especially if it's not a "powerful" one. And since you're asking for an event logged indicating an unsufficient PSU, I think that it would not be possible to differentiate between a PSU failure or a failure in the power managment of the graphic card. But I'm not aware of a such log. But I know some graphic card (NVidia 6xxx-8xxx or ATi 9xxx) complaining at boot time when there's a problem with the external power.
    – piernov
    Mar 4, 2014 at 22:47
  • Underclocked CPU by reducing the multipliers by a factor of 4 - same thing happens. Given the diff of max power consumption for the CPU and graphics is about 100W (in favor of the card), this was unfortunately to be expected. Neither did I find a LED on the card. No complaints on boot time, but then the power consumption is well under 100% at that point. So, back to square one.
    – mikołak
    Mar 6, 2014 at 19:42
1

So, the card got replaced under warranty due to an unspecified "hardware fault".

To summarize, if:

  • you have problems like the in the question (intermittent card poweroff without a temperature spike),
  • you've confirmed that your PSU is not underpowered w.r.t. your graphics card.

Then simply arrange to check whether the fault occurs when the card is inserted in a completely different hardware config. The machine that provides verification should have equal or greater specs, of course (especially the PSU and the motherboard).

In the end, this is exactly what I did - asked a friend to check the card in their computer. The exact same fault popped up in about 7 minutes.

Of course, there is a very remote possibility that the graphics card might be faulty in such a way as to do damage to the motherboard - however, this isn't very likely, and checking just the graphic card on another configuration is safer than than the other way around, since you only have a single possible faulty component in play here.

Improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .