PC restarts after GPU swap

[Vergil44]

I've been lucky enough to get my 3080 Aorus Xtreme after a nine-month on backorder, so installed it and everything seems fine, after trying out red dead and a few other games, after a few minutes of playing the pc would restart. At first, I thought one of the power cables were not plugged in correctly but that does not seem to be it. I checked thermals and even increased the max temp but still pc restarts even at 62c. I run a few GPU-Z logs but the power draw seems to be "normal" at the time of the restarts when the log ends, or at least there are points on the logs that the power draw has been higher than it was at the end when the pc restarted.

I also tried lowering the power draw limit of the card but it seems it still didn't work, looking around it seems that people do recommend a 750W but it is cutting it close.
Mind you PC works fine and most games run fine as well, it just seems to happen at really high settings and after a good amount of time.
 

My system:
Gigabyte Aorus 3080 Xtream
Intel Core i9-9900K
Seasonic Prime 750W Titanium 
32.0 GB Ram
 

I was recommended to upgrade my PSU but I don't wanna just to conclusions and spend more money for it to not actually work. And If I do swap PSU should I go with an 850 or go all out for 1000?

[Guest]

2 minutes ago, Vergil44 said:

Seasonic Prime 750W Titanium 

This is why.

 

But you don't need > 750W.  You just need one that's not Seasonic.

 

 

[KaitouX]

You can try to power limit/undervolt more as a possible temporary fix for the issue, maybe lower the power limit to the minimum to test and increase from there. The permanent fix was already mentioned.

[Juular]

5 hours ago, Vergil44 said:

after a few minutes of playing the pc would restart

5 hours ago, jonnyGURU said:

This is why.

Shouldn't it shutdown latching off in that case ?

RAM maybe ? Tried to termporarily disable XMP ?

[Vergil44]

10 hours ago, jonnyGURU said:

You just need one that's not Seasonic.

Damn, I thought Seasonic was a good brand with especially since it was recommended to me and saw a few sites also recommend it. Any suggestions in this case? Cause I was leaning into just getting a higher W Seasonic one just so I don't have to swap out the cables.

4 hours ago, Juular said:

RAM maybe ? 

It's not the ram, before the GPU swap everything was running normally, and I also tried with two other graphics cards and it was running normally, so it has something to do with the GPU.

[electropical]

4 hours ago, Juular said:

Shouldn't it shutdown latching off in that case ?

Apparently, Seasonic Prime doesn't latch, the PC just restarts.

Linus witnessed the exact same behavior when he tried running 2 3090s in SLI with a TX1000 (at 7:33) :
https://youtu.be/i1dGQiNfCAc?t=453

Don't ask me why though, no idea

[taavi]

12 hours ago, jonnyGURU said:

This is why.

 

But you don't need > 750W.  You just need one that's not Seasonic.

 

 

Are Seasonic PSU's bad with the new cards? Even the new (rebranded?) prime series?

[Pwnstix]

7 hours ago, taavi said:

Are Seasonic PSU's bad with the new cards? Even the new (rebranded?) prime series?

I can comment on this. I posted about it at greater length a few months ago.

 

So, last fall, I got an EVGA RTX 3080 (FztW3 Ultra), and just in case my Seasonic Prime Titanium 750W PSU from around 2017 wasn't capable enough,  (although it had never failed me before) I went ahead and bought a new Seasonic Prime TX-1000.

 

I installed the new PSU and the 3080 at the same time. One of the first things I did aftering booting my PC back up was to run a benchmark (I ran a few tests in 3DMark). 

 

My PC rebooted during a run of Timespy. I thought maybe it was a fluke, maybe the drivers needed some time. It didn't happen again...until a few weeks later, while I was browsing the internet. That time, I thought it might've been an issue with one of the browsers I had open, so, again, I didn't think much of it. But it started happening more and more frequently while I was playing Cyberpunk 2077. 

 

There was no overheating, I didn't mess with voltage, and I also played around with lowering the power limit.

 

I thought for sure that my brand new TX-1000 PSU couldn't be the culprit, but just in case, I swapped it out for my older Prime Titanium 750W. The reboots persisted, so it couldn't be the PSU, I thought. I also ruled out my uninterruptible power supply (UPS) by plugging directly into the wall. So, I figured it was maybe the video card.

 

I decided to cross-ship RMA my 3080, but the same thing happened with my replacement. I set all my motherboard settings back to default, disabled XMP, everything. Nothing helped. 

 

I panicked and thought maybe there was something wrong with my motherboard, CPU, RAM...I ran memtest86 for half a day and passed. I ran several of OCCT's stress tests, and passed, but the GPU test would almost always cause my PC to reboot. My thoughts were that it might've been the PCIe slot, or the CMOS battery, so I swapped it out..no joy.

 

The reboots were happening more frequently and while the PC was idle, and eventually it wasn't just rebooting--it was shutting off.

 

So I panic bought a new motherboard, CPU, RAM. Did a little minor upgrade to a 5800X. Things seemed fine for a few days, but then I ran a stress test and my PC shut itself off. Tried a few games--same thing. And then it started shutting off on its own again. 

 

So I bought another damned PSU, the one I'm currently using -- an EVGA SuperNova P2 850. I was a bit wary of "downgrading" my wattage from 1000W to 850W, but it's still within the system requirements for the RTX 3080. I haven't had any more issues since I made that swap, which was almost 4 months ago.

 

So, yeah, I made some stupid choices, panic bought a whole new rig. But it was just really hard for me to believe that there was a problem with my then-brand new TX-1000, especially since I'd read that Seasonic had fixed the transient response compared to the older Prime Titanium units  -- until I started reading more about Seasonic PSUs here, and reading more about transient response, etc. I just hadn't done enough research and didn't understand a number of things beforehand, and that's on me. 

 

I RMA'd my TX-1000 and Seasonic was good enough to send me a replacement, although I did do an advanced RMA which basically cross-shipped the new one. They received my old unit several months ago, and I can still access the RMA status page, where I can see that my old unit is in "testing." I'm really curious to see their testing results, but it's been in the testing phase for around two months now, I think...

 

So far, this EVGA unit hasn't given me any problems. I do feel a little silly for building a new PC, although I had planned on upgrading from what I had, eventually. But it seems like the culprit to my problems was my  previous TX-1000. I haven't used the replacement unit in anything yet, and as far as my older 750W unit showing the same behavior, I suppose that might have to do with it also being an older model in the Prime Titanium series: there was a minor upgrade about a year after I bought that one, to Prime Ultra Titanium -- something about transient response, again.

 

Anyway, sorry I got kind of long winded again, so: TL/DR: my PC was rebooting with a RTX 3080 and my brand new Seasonic TX-1000 ended up being the problem component; another (EVGA SuperNova P2 850) PSU solved my problem.

 

 

edit: as jonnyGURU says in the post below mine: the whole transient spike thing doesn't seem to be causing these issues or tripping OPP. I kept seeing things about "transient spikes," etc., and I bought into all that as well, because of my own relative ignorance regarding PSU construction. I always figured most of the problem had to do with the general consensus that Ampere's power delivery/PCB construction is...not always that great. That's a whole other topic entirely, but...it does tie into how certain power supplies behave. 

 

 

 

 

[Guest]

4 hours ago, taavi said:

Are Seasonic PSU's bad with the new cards? Even the new (rebranded?) prime series?

Yeah. Though really it's more the card to blame than anything.  The power management on Ampere is horrible.

 

And it's not transient spikes tripping OPP.  People keep saying that, but I don't know where that rumor started.  I mean, it's not a bad guess, 

 

The problem that was observed is that under high loads, the ampere cards actually feeds noise back to the PSU.  The PSU sees this noise as a fault and shuts down.  Depending on the design of the PSU, this noise can be filtered out.  I've even seen some OEMs do something as simple as putting a ferrite bead on the +12V sense lead.  For most Seasonic users, I just tell them to remove (cut, de-populate, whatever) the +12V sense wire from the 24-pin cable and everything rights itself.

[taavi]

3 hours ago, jonnyGURU said:

Yeah. Though really it's more the card to blame than anything.  The power management on Ampere is horrible.

 

And it's not transient spikes tripping OPP.  People keep saying that, but I don't know where that rumor started.  I mean, it's not a bad guess, 

 

The problem that was observed is that under high loads, the ampere cards actually feeds noise back to the PSU.  The PSU sees this noise as a fault and shuts down.  Depending on the design of the PSU, this noise can be filtered out.  I've even seen some OEMs do something as simple as putting a ferrite bead on the +12V sense lead.  For most Seasonic users, I just tell them to remove (cut, de-populate, whatever) the +12V sense wire from the 24-pin cable and everything rights itself.

Thank you for this info!

 

I am having the same exact trouble as Pwnstix mentioned above. I have Strix 3090 and prime TX-1000 (bought sept. 2020).

 

So basically to kind of "fix" the issue I must unplug the 12V sense plug from PSU side. If I understand correctly I can apply the same pin layout for Seasonic Prime TX-1000 that is on jonnyGURU's page: jongerow.com/Corsair_pinouts/Corsair_pinout_AX_Platinum.htm ?

[electropical]

3 hours ago, jonnyGURU said:

Yeah. Though really it's more the card to blame than anything.  The power management on Ampere is horrible.

 

And it's not transient spikes tripping OPP.  People keep saying that, but I don't know where that rumor started.  I mean, it's not a bad guess, 

 

The problem that was observed is that under high loads, the ampere cards actually feeds noise back to the PSU.  The PSU sees this noise as a fault and shuts down.  Depending on the design of the PSU, this noise can be filtered out.  I've even seen some OEMs do something as simple as putting a ferrite bead on the +12V sense lead.  For most Seasonic users, I just tell them to remove (cut, de-populate, whatever) the +12V sense wire from the 24-pin cable and everything rights itself.

It is understood that the problem is - as you explained months ago - "noise sent down the +12 sense wire causing the supervisor IC to freak out and shut the PSU down".

But why does the system restart?

I used to believe that the only way a PSU would shut down "on its own volition" is by latching after the triggering of one of its protections.
The fact that it shuts down without latching is already surprising to me (admittedly, my knowledge is very limited), and seeing the system restart only makes things weirder.

Can you please explain what's exactly going on?

[Guest]

3 hours ago, taavi said:

Thank you for this info!

 

I am having the same exact trouble as Pwnstix mentioned above. I have Strix 3090 and prime TX-1000 (bought sept. 2020).

 

So basically to kind of "fix" the issue I must unplug the 12V sense plug from PSU side. If I understand correctly I can apply the same pin layout for Seasonic Prime TX-1000 that is on jonnyGURU's page: jongerow.com/Corsair_pinouts/Corsair_pinout_AX_Platinum.htm ?

Yes.  But don't just leave the pin dangling.  It's live with +12V. 

 

[Guest]

2 hours ago, electropical said:

It is understood that the problem is - as you explained months ago - "noise sent down the +12 sense wire causing the supervisor IC to freak out and shut the PSU down".

But why does the system restart?

I used to believe that the only way a PSU would shut down "on its own volition" is by latching after the triggering of one of its protections.
The fact that it shuts down without latching is already surprising to me (admittedly, my knowledge is very limited), and seeing the system restart only makes things weirder.

Can you please explain what's exactly going on?

Without Seasonic's help, I can't analyze the problem any further.  All I know is that if you scope the +12V sense line, you can see noise on it increasing exponentially until the PSU either shuts down or restarts.  We're not sure exactly what protection is thinking there's a problem.  We can only capture the problem itself.  And we know that it's that because cutting the +12V sense makes the problem go away.

 

[taavi]

5 hours ago, jonnyGURU said:

Yes.  But don't just leave the pin dangling.  It's live with +12V. 

 

Yes of course I will isolate it. Going to try this and contact Seasonic too and will report back here if this fixes it for me.

[dag_dg]

Double check your PCIe power connectors. I was having this exact problem with a 3090 and it turned out to be a loose PCIe power connector. I lucked out because the card I had(Strix) featured little LEDs that showed if the connector was loose but looking at the connection it appeared as though it was fully plugged in. Double checked to make sure it wasn't a bad cable or loose cable by jiggling the connector and watching for the warning LED to come on but it never did. Was just *barely* loose and that was enough to cause it to reboot once the GPU usage started ramping up.

[Vergil44]

So any suggestions for a good PSU? I find it hard to find an up-to-date PSU that is both quiet and efficient with a good rating on the PSU rating list.

[Dogzilla07]

Adata XPG Core Reactor 850w is a very good choice. Cheap, excellent, high quality, silent, compact.

[Sleepycat3]

Just joined to share with you all as I had this same issue last night when I started using AI based video processing methods. The GPU power draw was not much, at 300W, and after 5 seconds of processing, my PC would just shut down immediately as if power was cut off. My PSU is a 2017 model Seasonic Prime Platinum 850W. The system's total power draw at the time of shutdown was under 550W.

 

I spoke to Seasonic and they had approved the RMA. However, i read Jonnyguru's post here and thought that I would give the ferrite choke on the +12V sense wire a go before returning my PSU. Lo and behold, it worked! I powered through 2 hours of video processing and it finished without a restart.

 

So for those who have this issue, it is worth trying it as it is non-destructive as you don't have to cut anything, just clip on the split choke. I recommend a 3.5mm split choke, but I used a 5mm instead as I had one lying around. I did use some heat-resistant foam so ensure that the sense wire was pushed up against the ferrite choke.

 

Please don't mind the dusty case, the NF-A14 PPC3000 intake fan is wrecking havoc with the fan filter. This is 2 weeks after proper cleaning.

 

 

[IIIIIIIIII]

40 minutes ago, Sleepycat3 said:

Just joined to share with you all as I had this same issue last night when I started using AI based video processing methods. The GPU power draw was not much, at 300W, and after 5 seconds of processing, my PC would just shut down immediately as if power was cut off. My PSU is a 2017 model Seasonic Prime Platinum 850W. The system's total power draw at the time of shutdown was under 550W.

 

I spoke to Seasonic and they had approved the RMA. However, i read Jonnyguru's post here and thought that I would give the ferrite choke on the +12V sense wire a go before returning my PSU. Lo and behold, it worked! I powered through 2 hours of video processing and it finished without a restart.

 

So for those who have this issue, it is worth trying it as it is non-destructive as you don't have to cut anything, just clip on the split choke. I recommend a 3.5mm split choke, but I used a 5mm instead as I had one lying around. I did use some heat-resistant foam so ensure that the sense wire was pushed up against the ferrite choke.

 

Please don't mind the dusty case, the NF-A14 PPC3000 intake fan is wrecking havoc with the fan filter. This is 2 weeks after proper cleaning.

 

 

Thanks for the useful information. What GPU did you use?

[Sleepycat3]

5 hours ago, IIIIIIIIII said:

Thanks for the useful information. What GPU did you use?

My GPU is an EVGA 3080 XC3 Ultra.

[Guest]

10 hours ago, Sleepycat3 said:

Just joined to share with you all as I had this same issue last night when I started using AI based video processing methods. The GPU power draw was not much, at 300W, and after 5 seconds of processing, my PC would just shut down immediately as if power was cut off. My PSU is a 2017 model Seasonic Prime Platinum 850W. The system's total power draw at the time of shutdown was under 550W.

 

I spoke to Seasonic and they had approved the RMA. However, i read Jonnyguru's post here and thought that I would give the ferrite choke on the +12V sense wire a go before returning my PSU. Lo and behold, it worked! I powered through 2 hours of video processing and it finished without a restart.

 

So for those who have this issue, it is worth trying it as it is non-destructive as you don't have to cut anything, just clip on the split choke. I recommend a 3.5mm split choke, but I used a 5mm instead as I had one lying around. I did use some heat-resistant foam so ensure that the sense wire was pushed up against the ferrite choke.

 

Please don't mind the dusty case, the NF-A14 PPC3000 intake fan is wrecking havoc with the fan filter. This is 2 weeks after proper cleaning.

 

 

Nice!

 

I would've just cut the wire myself. 

 

[Guest]

1 hour ago, The Unknown Voice said:

What mix # for the bead or Choke?

 

I think you're meaning to ask @Sleepycat3, not me.  I have no way to know what mix # ferrite bead High Power uses in their PSUs. 

 

[Sleepycat3]

3 hours ago, The Unknown Voice said:

What mix # for the bead or Choke?

 

Not sure what the mix # is for this choke. It was just one of those that came off those old yellow video out plug cables that connected between those old Sony Handycams and your CRT television to transfer the analog signal.

 

The markings on it say: ZCAT, 2035 0930 and when I looked it up, it is made by TDK, the specs say that it has an average impedance of 50 Ω in the 10 - 100MHz frequency range and 100 Ω for the 100 - 500 MHz range.

 

TDK doesn't mention the mix # in their data sheet, but the Z value for the frequency range seems to be high for a very wide frequency range. So the mix is probably proprietary to have such impedance over a wide frequency range.

 

https://product.tdk.com/system/files/dam/doc/product/emc/emc/clamp/catalog/clamp-filter_commercial_zcat_en.pdf

[taavi]

I removed sense pin and for now I have not experienced any random restarts.

 

Can there be any negative side effects when removing the pin, or is it just a control for "smooth" voltage delivery?

[IIIIIIIIII]

18 hours ago, taavi said:

or is it just a control for "smooth" voltage delivery?

Yes.

http://jongerow.com/no-sense/