what happens if power supply is insufficient?

Colonel Sanders

Supreme [H]ardness
Joined
Sep 26, 2001
Messages
5,582
we all know that current (and upcoming) GPUs can be extremely hungry for power, in some cases exceeding 350 watts under load. same with many higher end CPUs.

my question is, how does an insufficient power supply manifest in terms of symptoms during normal use? like say you tried to put a 3080 ti on a 550 watt power supply - what actually happens? googling this topic indicates the symptoms should be pretty "severe" i.e. immediate system shutdown or serious errors (hard lock, blue screen, random restarts.) but is that the only thing that can happen? is there a point where a psu is close to its limit and might cause other problems even if not enough to actually trip the protections and shut down the PC? can monitoring software like HWinfo provide the necessary info to determine if the psu is causing issues?
 
If protections fail, you could...

Get an under/over-voltage condition on the lines, or high ripple, as the supply struggles to provide the requested current.
Burn through the insulation on the wires or melt connectors, if the wires or connectors have more current running through them than they are designed for.
Burn up or blow up the switching components in the power supply, which may start a fire if they fail short to ground.

I think that covers most of the failure modes for supplies with bad or no protection, or where the protection failed.
 
Nobu is spot on, but I'll add a little insight if the protections work. Immediate system crash, black screen, then you'll see BIOS splash as it reboots. Once there is draw greater than supply it doesn't take much to push those protections out of tolerance.
 
If protections fail, you could...

Get an under/over-voltage condition on the lines, or high ripple, as the supply struggles to provide the requested current.
Burn through the insulation on the wires or melt connectors, if the wires or connectors have more current running through them than they are designed for.
Burn up or blow up the switching components in the power supply, which may start a fire if they fail short to ground.

I think that covers most of the failure modes for supplies with bad or no protection, or where the protection failed.
thanks, but i'm not saying what if protections fail - i'm more asking, can issues occur *before* protection trips?
 
thanks, but i'm not saying what if protections fail - i'm more asking, can issues occur *before* protection trips?
Pretty much the same, just less likely. As a PSU ages, it becomes less capable of supplying current up to it's rating, so you'll get more ripple and voltage will run higher or lower. Eventually the switching components also become a liability, especially if it's been run near the limit and depending on how well they were cooled.
 
thanks, but i'm not saying what if protections fail - i'm more asking, can issues occur *before* protection trips?
Usually nothing. Your PC will just reboot as the protections are tripped for whatever reason. This is a common cause for the numerous posts you'll find across the internet where people don't want to accept that their cheap power supply is not up to the task.

In severe cases you can experience hard locks (no BSOD) as various components in your PC run into issues from the power delivery coming from the power supply.
 
The most common symptoms are likely to be either random BSOD-like errors or in more extreme cases, hard shutdowns.

However.

There are graphics card designs that can actually damage themselves if the power supply is unable to keep the 12V supply voltage in the proper range. This happens when the voltage used to drive the gates on a large high-side transistor comes directly from the 12V power supply rail, and when that droops, the transistor doesn't turn on all the way, and you then get a ton of resistance through it, which causes it to burn way more energy off as heat than it normally would. If this continues for a long enough period of time, either in a single session or across many over longer term, the transistor can eventually fail, and that frequently means it becomes a short directly to the output. The GTX 690 is one such card that is for sure prone to this. That's obviously an old and mostly irrelevant board design, and newer designs usually drive the transistor gates with 5V or 7V from a dedicated regulator on the board, but it's not out of the question that some designs may still be prone to this, especially on minor rails that are required, but most folks don't ever think about. On the 690, the rail that's most susceptible to this is the 0.95V rail, which supplies power to circuits on the GPU die, even if it's not THE GPU power.

Another possibility is that the power supply itself can fail if it's run at or beyond its designed capacity for a long period of time, and what happens then can be unpredictable.
 
thanks guys. the reason i'm asking is because i recently started running into games CTDing at random over the past few weeks. at first i thought it was an Nvidia driver problem but that's definitely not the case as i tried rolling back to various older versions (including ones i definitely had no problems with before) and no change in crashing.

if you check the memory subforum, you'll see a thread i made about finding defective ram - i've fixed that particular problem, but the crashes continue. my Seasonic 650w psu is a) below the recommended wattage, which i knew might become a problem when upgrading to my 3080 ti and b) it's like ten years old now and has a lot of hours (and stress) on it. was wondering if it could be causing symptoms like crashing, but not shutting down the PC or doing other more "obvious" PSU symptoms.

this is likely the first time in my ~25 years of building pcs that i've got components likely drawing at or beyond the rating of the PSU. and it's also probably the longest i've ever used a single PSU.

edit: i have an 850w PSU on the way from Bezos, as well as a Kill-a-Watt meter. i figured it would be very interesting to see just how much power is being drawn at the outlet.

edit2: went and looked up the actual order date for my power supply.. February of 2011 :eek:
 
Last edited:
It's the output voltage from the power supply that matters, and not the wattage draw at the wall. The ratings are presumably based on component selection for the power supply, but they're ultimately just kind of arbitrary.

You can measure the voltage on each rail with a meter. Start up Heaven or some other game-like looping benchmark app (not Furmark! Fuck Furmark!), and use the meter to measure the voltage on the 12V wires. Ideally, you'd want to do this on the card (I use the solder joints for the 8 pin connector), but you can measure it elsewhere and get close enough most of the time.
 
It's the output voltage from the power supply that matters, and not the wattage draw at the wall. The ratings are presumably based on component selection for the power supply, but they're ultimately just kind of arbitrary.

You can measure the voltage on each rail with a meter. Start up Heaven or some other game-like looping benchmark app (not Furmark! Fuck Furmark!), and use the meter to measure the voltage on the 12V wires. Ideally, you'd want to do this on the card (I use the solder joints for the 8 pin connector), but you can measure it elsewhere and get close enough most of the time.
If your hands are a bit shakey, maybe use a sticky note or two to separate the pins so you don't accidently short to ground.
 
If your hands are a bit shakey, maybe use a sticky note or two to separate the pins so you don't accidently short to ground.
Or kapton tape. That's probably the "proper" way.

You'd figure out which pins are 12V (farthest from the latch on normal, non-goofy nvidia proprietary connectors), cover all the others with kapton tape, and use the case or I/O bracket as the ground.
 
It's the output voltage from the power supply that matters, and not the wattage draw at the wall. The ratings are presumably based on component selection for the power supply, but they're ultimately just kind of arbitrary.

You can measure the voltage on each rail with a meter. Start up Heaven or some other game-like looping benchmark app (not Furmark! Fuck Furmark!), and use the meter to measure the voltage on the 12V wires. Ideally, you'd want to do this on the card (I use the solder joints for the 8 pin connector), but you can measure it elsewhere and get close enough most of the time.
thanks. are the readings shown in HWinfo basically bunk/not helpful?
 
thanks. are the readings shown in HWinfo basically bunk/not helpful?
The problem with software monitoring is that some power supplies have more than one 12V rail, and so the one that motherboard's monitoring hardware is seeing isn't necessarily the same one that's powering the graphics card or even the CPU. So, it can be worth looking at, and if you're seeing way less than 12V there, that's a clue, but it won't always give you the whole picture.

Most of this being moot if you've got a new power supply on the way anyway.

Edit: This is notably different from the fancy power supplies that have software monitoring. I have a Corsair AX1200i that monitors and reports each rail individually via the Corsair software, and that's generally accurate enough for this purpose.
 
there are plenty of thread around here talking about the under powered psus and 3000 series cards. systems turn off under gaming load.
 
11.7 would be scary to me. I cringe at 11.8-11.9. Once I see on my DMM the 12v line drop to 11.8, I yank the psu and get a new one.
 
Pretty much the same, just less likely. As a PSU ages, it becomes less capable of supplying current up to it's rating, so you'll get more ripple and voltage will run higher or lower. Eventually the switching components also become a liability, especially if it's been run near the limit and depending on how well they were cooled.
Are you saying that a l situation where the PS runs well under rated wattage, no issues at all, still needs to be replaced after X years of operation? How do you know what X is?

With all the cabling, etc, I would have to replace my Corsair AX 85o with something "pin compatible." The power cabling is a bitch to redo.
 
Are you saying that a l situation where the PS runs well under rated wattage, no issues at all, still needs to be replaced after X years of operation? How do you know what X is?

With all the cabling, etc, I would have to replace my Corsair AX 85o with something "pin compatible." The power cabling is a bitch to redo.
No, if it's always run well below the limit, you'd probably be fine. But if you ever stressed the supply after 10-20 years, you could cause failure even if you are under the design limit.

Materials breakdown, even under perfect conditions. Nobody can say exactly how long yours will last, so probably your best course if you choose not to replace after 10 years, is to test the supply (under load) periodically for voltage deviation, and also ripple if you are able.

Or you can just not care. That's pretty much how I've been going -- I replace when I do a new build and the old supply is insufficient or incompatible. Usually that's less than 10 yrs anyway, or I'm not 32.
 
Are you saying that a l situation where the PS runs well under rated wattage, no issues at all, still needs to be replaced after X years of operation? How do you know what X is?

With all the cabling, etc, I would have to replace my Corsair AX 85o with something "pin compatible." The power cabling is a bitch to redo.
Corsair use type 4 for all their newer PSUs (corsairs pinout standard) so if your PSU is new enough then you can just swap for any corsair PSU with modular cables. If you have type 3 (80plus platinum versions) then you need to redo the 24v but if you have 80 plus gold then you will need to redo everything while 80plus titanium is type 4. Check link below to see compatability.

https://www.corsair.com/us/en/psu-cable-compatibility
 
The recommended PSU on the GPU box is there for a reason. Too small or too low quality PSU has generally been the top reason for system instability along with memory. I wouldn't even consider running a 3080ti with anything lower than an 850w PSU and if running Intel I would probably consider getting a 1000w PSU due to the boost drawing in excess of 200w. There are lots of other things that draw power in addition to the CPU and GPU. While their powerdraw is low in isolation it all adds up. E.g. my cooling and lighting probably draws 50w, but set to full cooling it will probably draw 80-90w. Then there are SSDs, chipset etc. which also each draw a little power but it adds up.
 
The recommended PSU on the GPU box is there for a reason. Too small or too low quality PSU has generally been the top reason for system instability along with memory. I wouldn't even consider running a 3080ti with anything lower than an 850w PSU and if running Intel I would probably consider getting a 1000w PSU due to the boost drawing in excess of 200w. There are lots of other things that draw power in addition to the CPU and GPU. While their powerdraw is low in isolation it all adds up. E.g. my cooling and lighting probably draws 50w, but set to full cooling it will probably draw 80-90w. Then there are SSDs, chipset etc. which also each draw a little power but it adds up.
i just got done swapping in a 850w psu, we'll see how she do.
 
thanks guys. the reason i'm asking is because i recently started running into games CTDing at random over the past few weeks. at first i thought it was an Nvidia driver problem but that's definitely not the case as i tried rolling back to various older versions (including ones i definitely had no problems with before) and no change in crashing.

if you check the memory subforum, you'll see a thread i made about finding defective ram - i've fixed that particular problem, but the crashes continue. my Seasonic 650w psu is a) below the recommended wattage, which i knew might become a problem when upgrading to my 3080 ti and b) it's like ten years old now and has a lot of hours (and stress) on it. was wondering if it could be causing symptoms like crashing, but not shutting down the PC or doing other more "obvious" PSU symptoms.

this is likely the first time in my ~25 years of building pcs that i've got components likely drawing at or beyond the rating of the PSU. and it's also probably the longest i've ever used a single PSU.

edit: i have an 850w PSU on the way from Bezos, as well as a Kill-a-Watt meter. i figured it would be very interesting to see just how much power is being drawn at the outlet.

edit2: went and looked up the actual order date for my power supply.. February of 2011 :eek:
You should get a sine wave UPS for your PC if you don't have one.
i just got done swapping in a 850w psu, we'll see how she do.
Should be just fine for a 3080 Ti.
 
With good PSUs, nothing bad happens, just OCP is triggered and the system shuts down. Happened a couple of times on my mining extension when OCs got reset for some reason and the cards started pulling 1200W+, the main system with the single 3090 was still up and running and a reboot resolved that issue.
 
Aside from the mentioned things, in some cases, you can hear stuff.

Squealing noises from the PC itself and various whistles and clicks in the audio you're listening.

USB devices going offline needing a restart - normally you'd suspect the USB device, but no! PSU bad!

The fan in the PSU is working overtime and making noise.

If you follow strict precautions, you can look inside. A visual inspection of areas around/beneath big transistors, the bridge rectifier, brown PCB near resistors, weird gray non-silvery solder. Weak spots like to manifest themselves pretty clearly after a few years of use.

It's often a diagnosis by exclusion and source of many gremlin class problems that pop up once a week when you sneezed the wrong way.
It's not always a binary deal with failing PSUs - reduce the load and suddenly it operates "good enough" (but not really).
 
No, if it's always run well below the limit, you'd probably be fine. But if you ever stressed the supply after 10-20 years, you could cause failure even if you are under the design limit.

Materials breakdown, even under perfect conditions. Nobody can say exactly how long yours will last, so probably your best course if you choose not to replace after 10 years, is to test the supply (under load) periodically for voltage deviation, and also ripple if you are able.

Or you can just not care. That's pretty much how I've been going -- I replace when I do a new build and the old supply is insufficient or incompatible. Usually that's less than 10 yrs anyway, or I'm not 32.
I have a Thermaltake PSU tester, which I use may be once in five years, for my own system or a friends. Not that much, and it's a great way to ID or not a source of problems
 
Also, if you use a UPS then what's it rated at. I have a APC 750ES. When I really push my EVGA 3080 12GB (MSFS with everything ultra and render scale pushed up to 200) the UPS alarm will start beeping. These new cards draw a ton of watts. People think a UPS is only for backup power, but they have a throughput rating too while running on A/C power. They can limit the power going through to your computer's PS (mine's a EVGA 1000W).
 
what do you mean by "crashing"? the games ctd, windows bsod or power off/lock up?
actually that was a false alarm. my PC has been solid so far, every day since swapping the power supply. i have not had a single Nvidia driver-related event since. interestingly though, the GPU 8 pin voltage values reported in hwinfo were the same or maybe even a tinge less - but i know that's not conclusive without checking manually with a multimeter.
 
I am not sure what happens for most PSUs but I am sure that if its Gigabyte PSU then it blows up :)
lol, thankfully it was a Seasonic and not Gigabyte.

still no crashes whatsoever since swapping ram and upgrading psu :cool:
 
Gigabyte is known for their explosive power delivery ;)
gigabyte_psu.jpg
 
If power supply fails it created a black hole singularity that will absorb your computer and desk and throw it into the void.

Or you may hear a spark and a burning smell with little to no damage to other components.

Unless it’s a Gigabutt PSU. Then it will take a few components with it.
 
always stick with Corsair PSU's and they will never fail
Depends on what model it's rebranded from. Most of their low- to mid-range units are rebranded CWT or GreatWall. High-end to premium units are Seasonic or Flextronics. CWT is a "get what you pay for" manufacturer, so generally the more pricier units are more reliable.
 
Back
Top