PSU time again, unbelievable.

[dgianstefani]

Almost 2 years ago (23 months) I had to replace my SS 850 because my 3080 Ti was tripping PSU's OCP. Got SS TX-1000 and was good. Now my 4080S is triggering instant power-off, reading that it's SS to blame once again (even 1300 will trip off, something about faulty supervisor IC). Which PSU is known to work fine with 4080S? I don't need 1000W PSU, only have 4080S + 5800X3D.

Edit: swapped the 850 in, let's see what happens ...

It's not tripping because of OCP. It's some other problem. As you mentioned maybe an issue with sense pins.

The 4080S is a 300 W card with 340 W spikes.

I'm running an overclocked watercooled 3080 Ti on a 750 W SFX PSU, that GPU uses 25% more power than yours and I'm not even close to OCP.

Your entire system uses less than 400 W most likely, definitely no more than 500 W.

 

[xorbe]

Found I had an extra SeaSonic cable, pulled pin #4 from the psu side 10-pin (the 12V sense wire), and put the modified cable and the TX-1000 back in. Working so far, but it's going to take a few weeks before I can assume that it helped.

 

[nomdeplume]

DIT: and I did burn the CPU with Prime (225W limit since im on air) while running 4090 in cyberpunk with raytracing cranked to max at 4K to see if the unit would be enough -- it did pull an average ~680-700W with higher spikes but the unit was fine, no shut offs and was actually pretty quiet all things considered. I think i spiked it over 800W at one point and this PSU didn't flinch.

Yee-haw!

 

[Tomgang]

I can only say my Phanteks Revolt x 1200 watt (pretty much a rebranded seasonic psu, but with added feature to handle 2 system at the same time) handles my rtx 4090 just fine, even with only 3 of the 4 8 pins connected to it. The last one my rtx 4060 takes in the second system.

This does mean i do not take my rtx 4090 above its stock power target, doing so could overload the psu. Pulling up to 600 watt from what is technically rated to max 450 watt, Does not seems like a good idea. I mean 3 8 pins is max rated to 450 watt.

 

[QuietBob]

For months I ran the system in my profile with a 5800X3D and a reference 7900XTX on a 650 W Seasonic Focus GX (SSR-650FX). A stock 7900XTX can spike as high as 551 W for <1 ms and 507 W for 1-5 ms.

I tested dozens of games with uncapped framerate at that time and did not witness a single power-related problem. Also my current Prime TX-750 (SSR-750TR) is running this rig flawlessly.

Depends on how spikey the 4080S is.

 

[Bill_Bright]

Found I had an extra SeaSonic cable, pulled pin #4 from the psu side 10-pin (the 12V sense wire), and put the modified cable and the TX-1000 back in. Working so far, but it's going to take a few weeks before I can assume that it helped.

That's good but IMO, with a history of 2 PSUs from a quality brand having problems in less than 2 years, I still would want to ensure those PSUs are getting good, clean, stable power AND that they are properly grounded to Earth ground. So I would still urge you (and everyone) to get an outlet tester (see post #19 above) and make sure you outlets are properly wired and grounded.

This is also a good time to recommend the use of a "good" UPS with AVR too. Remember, battery backup power during a full power outage is only a minor bonus feature of a "good" UPS with AVR. It is the AVR (automatic voltage regulation) feature of the UPS that is the greatest (by far) advantage. Remember too that a surge and spike is little more than a fancy and expensive extension cord as they do absolutely nothing for abnormal low voltage events like dips (opposite of spikes) and sags (opposite of surges), or long duration sags (brownouts) - any of which can cause your electronics to suddenly stop, resulting in possible data and/or drive corruption.

For abnormal high voltage events, surge and spike protectors merely chop off ("clamp") the tops off the sine waves, leaving a not-so pretty voltage waveform for our power supplies to compensate for.

A "good" UPS with AVR will help shape (regulate) the sine wave into something more easily used by the devices plugged into it. In minor low voltage events, it will use the batteries to boost the voltage up to normal levels, and in extreme high voltage events, it will use the batteries to dump the excess voltage (which batteries can absorb with ease), and/or dump the excess to ground (Earth) (through a properly grounded outlet). And all of this via the AVR circuits without switching over to battery backup!

 

[xorbe]

I just realized I have another TX-1000 in a machine in the closet (i9-9900KS), I could have swapped (identical) PSUs before trying the cable hack. R2105 and R2106!

 

[xorbe]

Nope, it still did an insta-off with the 12V sense wire disconnected after 16 days. Will swap TX-1000 with the other machine and observe.

The easiest trigger is hitting m or j (map or journal) while playing CyberPunk 2077. Harder to hit is with cuda computations.

I suppose it could also be the motherboard (Asus Rog Crosshair VIII Dark Hero X570), or a bad 4080S FE. But the motherboard was working fine with a much more power hungry 3080 Ti FTW3, and the 4080S seems to be fine otherwise.

 

[ThrashZone]

Hi,
Frankly I'd just get a superflower unit and call it a day.

 

[Bill_Bright]

Did you test your wall outlets?

 

[Solaris17]

Did you test your wall outlets?

Two seasonics in a row? No way.

 

[AusWolf]

A 4080 Super shouldn't trip a 1000 W Seasonic PSU, maybe unless you're using it with a 14900K overclocked to the moon. Maybe.

I'd try an RMA if I were you.

 

[Bill_Bright]

Two seasonics in a row? No way.

Yes, way! Two Seasonics in a row being taken out and failing in the same manner in a relatively short period of time would be a rare event. That is exactly why I suggested 2 weeks ago to test the wall outlets to make sure they are properly wired AND grounded to Earth ground. This should be done if for no other reason but to eliminate faulty power from the equation.

In fact, if me, I not only would check for proper wiring and grounding, I would measure the voltage too. A few years ago, the tap on the transformer feeding the houses on my block failed - but all the lights were still on! ??? If it hadn't been for my UPSs alarming, I never would have seen that instead of 120VAC, we were getting 146VAC! Not good.

If interested in the rest of the story, see this TPU post.

 

[nomdeplume]

Testing should be taken to mean determining what else is on the circuit and exploring all outlets, all devices, for a source of this issue. A good example would be a failing appliance that doesn't trip the breaker, but injects something harmful into the power line.

 

[freeagent]

I have my system plugged into a line conditioner, not sure if it actually helps but it gives me peace of mind

 

[xorbe]

I personally updated and checked all the outlets in my house. Crawled through the attic and under the house, it has good copper wiring. The PC was fine with a ~420W evga 3080 Ti FTW3, and then as soon as the FE 4080S (~320W) went in, this occasional insta-off thing started. The 3080 Ti was tripping an older SS 850W OCP (over-current protection) so I got the SS TX-1000, which was fine, until the FE 4080S. So either the TX-1000 (one of them at least, just started testing the second one) doesn't like the 4080S, the 4080S is faulty, or maybe the motherboard is somehow unhappy with the 4080S electrically. With the intermittent nature, all I can do is swap parts and wait. It only seems to happen with CUDA or CyberPunk 2077, and with CyberPunk it's been on the game to map/journal transition.

 

[ThrashZone]

Two seasonics in a row? No way.

Hi,
lol yeah those pesky wall plugs never get focused on hehe

 

[Solaris17]

Yes, way! Two Seasonics in a row being taken out and failing in the same manner in a relatively short period of time would be a rare event.

I was agreeing with you bro. Because the series of events is wack. God communicating with you is difficult lol.

 

[xorbe]

Not two! Just started testing the second TX-1000 as of today! The previous older SS 850W is known to be OCP overly-sensitive, that one is not a mystery, and only takes minutes for the issue to happen w/3080 Ti FTW3. Fwiw, when I ran the SS 850W + FE 4080S for a couple days, it had no immediate problems.

I thought I found something (but no), I noticed CyberPunk's REDlauncher.exe crashed before each power-off in Event Viewer, but actually REDlauncher silently crashes *every time* when clicking play, which is a known issue that most people don't notice. So that was a nothingburger.

 

[Bill_Bright]

I was agreeing with you bro. Because the series of events is wack. God communicating with you is difficult lol.

Well, I mean you quoted me then said, "No way" so not sure how that means you agree with me. But great! I am glad we are (now) on the same page.

 

[ty_ger]

A problem with NVIDIA GPUs is that NVIDIA only limits average power, not peak power. With the 3000 series, they effectively disabled OCP on the GPU completely. A spike will always happen after an idle period, and it could be more than is normally acceptable. I don't think it is always fair to blame the PSU if it is advanced enough and fast enough to closely follow applicable specifications when NVIDIA chooses not to.

 

[ThrashZone]

I was agreeing with you bro. Because the series of events is wack. God communicating with you is difficult lol.

Hi,
Yeah I got it but you did forget the sarcasm tag
/s

 

[dgianstefani]

A problem with NVIDIA GPUs is that NVIDIA only limits average power, not peak power. With the 3000 series, they effectively disabled OCP on the GPU completely. A spike will always happen after an idle period, and it could be more than is normally acceptable. I don't think it is always fair to blame the PSU if it is advanced enough and fast enough to closely follow applicable specifications when NVIDIA chooses not to.

4000 series GPU spikes are lower than 3000 series, and 4080 Supers are lower than 4080. It's not an issue like it used to be.

I run an overclocked 400 W PL (higher sustained draw than a 4080S spikes to) 3080 Ti that can spike to around 500 W, on a 750 W SFX PSU with zero issues.

Besides. If I can read charts right, RDNA2 is the worst offender spike wise, so I don't think it's fair to single out NVIDIA.

6950 XT would be worse.

A 6800XT spikes more than 200 W higher than a 4080S, so I don't think it's the card to blame here, given a 1000 W TX PSU is being used, which should perform better than to power off at 350 W spikes.

 

[tpa-pr]

HX line of units is probably your best bet. I've heard a lot of SeaSonic units tripping on 4090's even and my GT1300w trips on my XTX. System specs are current and solid as a rock.

Just chipping in with an anecdote: I have a HX1000 2022 model and it has awful coil whine (confirmed since I replaced the GPU and moved house). It seems to be a consistent problem with the HX1000s (that may have been fixed with a later revision?).

 

[faye]

Using my EVGA 1300 G2 with my 4080. CableMod E-series cable built for it. Works great.

Zero issues here with an ancient Rosewill Capstone Gold 1KW or the much newer Silverstone SX1000R SFX-L. The 7900 XTX will put the Lian Li 850w SFX in to protection if you look at it funny, though.