[Imprezzion]

I had the exact same issue happening to me the last days: computer just shuts down randomly during gaming: power still seems to be on
as the RAM still has RGB but nothing else is on.

When running OCCT power test the system instantly powers off.

This leads me to believe that the power supply is indeed to blame.

its an Asus Rog Strix 850 W White edition and no more than 1,5 years old.

It only started happening the last week and i never experienced any trouble before.

Should i go for an RMA or just bite the bullet and go buy something else?

Event viewer gives this error: (70368744177664),(2) kernel power

Can the problem be something else or is it definetly a PSU issue?

Im not experienced to problems like this: any help would be greatly appreciated.

PSU. That unit is a rebranded Seasonic. Afaik it's based on a Focus unit.

 

[GamBoTron]

PSU. That unit is a rebranded Seasonic. Afaik it's based on a Focus unit.

I see: any suggestion for a new PSU ? its kind of a stupid question i know, just something that will probably last more than 1,5 years and can handle a 3080 lol

 

[shilka]

I see: any suggestion for a new PSU ? its kind of a stupid question i know, just something that will probably last more than 1,5 years and can handle a 3080 lol

If you dont want Seasonic a good CWT unit is probably your second best option
Something like a Corsair HX (NOT the HXi)

There is also Be Quiet

 

[Always Counterclockwise]

There is also Be Quiet

XPG Cybercore is another, based on CWT's CST platform.

 

[exploiteddna]

• Same Difference, whatever fix they do will only take seconds...
• The Dark Power Pro 12 is a fully digital unit or majority digital with digital signal processing ... just a bit inferior to Corsair AX1600i (but still the closest thing to it):

Ok makes sense
Yeah I read those reviews and overall are pretty solid.. Honestly wish they made a smaller version, like AX1200i (looks like they used to make them but stopped at some point.. wonder why??)

Because a TX model should in theory have the issue fixed from the factory.

thats what worries me about even getting the newer model. If im coming out of pocket for a new unit, id rather go with something that will give me more confidence. Seems even the new TX units can still be hit or miss (just less so compared to the older units like my current one)[/QUOTE]

It's just sitting in my closet ... I paid for shipping under $30, waited over a month and received the same unit (prime titanium 1kw) ... I'd just grab another unit that doesn't have these issues.

thats the plan im leaning towards now. and i wanna get something thats best of the best right now.. would go for the AX1600i but that thing is massive, I'd like to stick with units that are approx. the "normal" size, if possible. Either way, I want something thats going to be viable for the next 5-10 years, including potential upgrades to 4000-series gpus, raptor lake or ryzen 7000-series, etc etc. And trends indicate more and more power is going to be needed going forward, so im considering units between 1200-1600

If you dont want Seasonic a good CWT unit is probably your second best option
Something like a Corsair HX (NOT the HXi) ... There is also Be Quiet

Yeah CWT makes the dark pro 12 units. Also, whats wrong with the HXi? How would you rank the HX compared to the AXi, or compared to the EVGA T2 (which isnt digital at all but is an absolute tank, super reliable, can handle massive loads

XPG Cybercore is another, based on CWT's CST platform.

ill have to look into those, not sure ive ever seen or heard of them before, thanks!

 

[exploiteddna]

screw it, i just ordered the Dark Pro 12 1500W, on sale for 350 @ amazon (new, prime, sold/shipped by amazon) .. cheaper than the 1200W

EDIT: now ive gotta decide whether to get a replacement from SS or sell the one i have. my fear is theyll get the unit and say its fine, wasting a month of my time and shipping costs..
their definition of "working fine" is 'it works so long as you undervolt your card and dont overclock anything' .. yeah ok, the standards should be a lot higher for an industry leader and the flagship model. which audience do they think is buying their products? non-enthusiasts? mmk

 

[exploiteddna]

PSU. That unit is a rebranded Seasonic. Afaik it's based on a Focus unit.

Are all ASUS PSUs seasonic?

 

[shilka]

Are all ASUS PSUs seasonic?

The Strix are Thor are made by Seasonic but the TUF models are made by Great Wall so no

 

[postem]

screw it, i just ordered the Dark Pro 12 1500W, on sale for 350 @ amazon (new, prime, sold/shipped by amazon) .. cheaper than the 1200W

EDIT: now ive gotta decide whether to get a replacement from SS or sell the one i have. my fear is theyll get the unit and say its fine, wasting a month of my time and shipping costs..
their definition of "working fine" is 'it works so long as you undervolt your card and dont overclock anything' .. yeah ok, the standards should be a lot higher for an industry leader and the flagship model. which audience do they think is buying their products? non-enthusiasts? mmk

That is what bugs me at SS, and they keep selling these defective units. It's very dumb because it harm a solid reputation and the issue is already old

I brought 850 xpg, no complain, but o would buy any ATX 5 psu but there are almost none available now

 

[Ice009]

Never heard of this issue about Seasonic PSUs and high end graphics cards. Is the issue mostly with the Nvidia 3080/3090 series cards? I've currently got a Seasonic Prime Platinum 1300W, but am using an AMD 6900XT. I am assuming if I used a 3080/Ti like I intended to before I got the 6900XT, would I have issues if I swapped graphics cards down the line?

If you dont want Seasonic a good CWT unit is probably your second best option
Something like a Corsair HX (NOT the HXi)

There is also Be Quiet

Just curious, what is wrong with the HXi compared to the HX series?

 

[shilka]

Just curious, what is wrong with the HXi compared to the HX series?

The HXi lacks the capacitors on the cables which means it has worse ripple suppression than the RMi/ RMx which uses the same platform and are cheaper besides being better
The HXI is the most pointless series Corsair sells its worse and more expensive than almost everything else they sell

The HX/AX/AXi/RM/RMi/RMx are all better cheaper or both

 

[Ice009]

The HXi lacks the capacitors on the cables which means it has worse ripple suppression than the RMi/ RMx which uses the same platform and are cheaper besides being better
The HXI is the most pointless series Corsair sells its worse and more expensive than almost everything else they sell

The HX/AX/AXi/RM/RMi/RMx are all better cheaper or both

Thanks. Wish I knew that before buying the two HXi series PSUs I have .

As for the Seasonic issue, are people experiencing the issues only with high end Nvidia 3xxx series cards? Also, if I ever have the issue, would the PSU have to be RMA'd, or is it possible to get a replacement cable without the sense wire? I really don't want to pull the whole system apart as a have a Dark Base Pro 900 Rev. 2 and the PSU is hard to get to in this case . What is the easiest way to test for this issue? Do I need to install something like OCCT?

 

[Always Counterclockwise]

As for the Seasonic issue, are people experiencing the issues only with high end Nvidia 3xxx series cards?

I've seen reports across both AMD and Nvidia's upper ends over a fairly broad range of power supplies. I haven't tabulated all that and worked back from the supplies to rank platforms by probability of being affected or GPUs by probability of triggering, though. Also, reports aren't always clear as to the specific AIB and PSU model involved.

However, everything I've seen suggests the distribution skews heavily towards 3080s, 3080 Tis, and 3090s and towards Seasonic Focus and Prime platforms. As Seasonic's never formally acknowledged the issue it's unclear if they've managed to fix it with running changes—I've seen several threads where someone's claimed they have only for someone else to report OCP with a more recently manufactured supply later on in the thread. A few people have had 3070s, 6800 XTs, 6900s, and 6900 XTs take down systems but I haven't come across anything at the 3060 or 6800 non-XT level or below.

Also, if I ever have the issue, would the PSU have to be RMA'd, or is it possible to get a replacement cable without the sense wire?

It's unclear to me how effective Seasonic RMAs are in fixing OCP issues. Lots of unhappiness there and lots of people changing to different supplies and then selling off their Seasonic when the RMA returns. I'm actually not thinking of a single thread where I can recall someone coming back and saying the RMA fixed their OCP issue. Usually if the returned PSU is kept it's used in a secondary rig as the primary rig's supply's already been replaced.

So far as I know you have to cut or ferrite the sense wire yourself. There might be a third party making such cables but I didn't find anything last time I searched a few weeks ago.

My default for GPU stress is Heaven but there's a number of similar options. There is some evidence to suggest gaming is less likely to trip OCP than GPGPU so testing CUDA, OpenCL, or AI workloads may also be of interest. AMD doesn't have enough of an AI presence for there to be any data but, anecdotally, it appears Nvidia OCP interactions are sensitive to the specifics of AI tasks with some things being bombproof and others tripping OCP every time within seconds of starting.

 

[Dogzilla07]

It's unclear to me how effective Seasonic RMAs are in fixing OCP issues.

Just to clarify it's not OCP. It's noise feedback on the 12v sense line, that makes the Supervisor IC go haywire, that's why it happens on 1300W units as well, far away from any triggering points

1. Removing the 12v sense wire on the 24-pin cable fixes it
2. Putting a ferrite choke around the that wire fixes is a temp solution (but it only lasted a few months and reverted)
3. Running a 6-pin cable to the GPUs fixes it (as it bypasses the PCi-E power delivery from the board (new info that Jonnyguru mentioned this just last week) [MSI B550 Unify, and other more expensive AMD and Intel boards]

A) Any Seasonic Prime-based unit with a new Supervisor IC doesn't have it (EVGA G6/P6)
B) Any Seasonic Prime-based unit with custom cables doesn't have it (Asus THOR, THOR II)
C) Certain motherboards filter out the noise, and thus additionally cut-down the instances of the issue appearing

Cards most impacted by this are EVGA 3090s (and FTW %-wise the most if I remember correctly), and Sapphire 6900XTs (mostly XTXH-based). A Gigabyte 3080 Ti showed up twice, Though other 3080 Ti + and AMD 6900XT did show up (but in vastly smaller instances, exactly under the radar enough to not show up GamersNexus, and get widespread attention). If Jonnyguru didn't show exactly what the issue is and worked with the guys behind the tier list on the details, no one would have probably known it was a separate issue, that is this specific. (And the only reason why he broke his mantra of being diplomatic around industry-related people/companies and things, and brought this to light, is because Seasonic (well specifically 1 or a few people at Seasonic, soured the relationship and didn't want to cooperate with Corsair to a minimum level he could accept [for the Seasonic-based AX series units]).

Lots of unhappiness there and lots of people changing to different supplies and then selling off their Seasonic when the RMA returns

This can be chakled to time-constrained/up-time constrained people(for any of the many reasons, and most of the very valid) + people who formed an emotional bond with a corporation/product that is now compromised. I haven't noticed any objective, impartial, purely logical situation where someone had the issue re-appear after RMA.

It's unclear to me how effective Seasonic RMAs are in fixing OCP issues.

Going by people experiences on LinusTechTips forum, and secondary info from cultists Discord, and other various PSU discords, 100% effective in solving the issue (just cable change).

Also, reports aren't always clear as to the specific AIB and PSU model involved.

They're very clear on reddit, LinusTechTips, and various PSU discords, Jonnyguru site/youtube (maybe I'm misremembering and he only showed which wire to cut, and tested on a sunmoon).

A couple of New World killing cards incidents were a compounding issue regarding this, and an invaluable trove of information to cross-reference (as going by RMA correspondence pictures on reddit, at that time some Seasonic reps, inadvertently mentioned to 2 (or 3 different users), about not worrying about having the same PSU unit returned, as they just removed a sense pin cable. No more instances of this popped up, every other RMA interaction reported on the internet had Seasonic reps lips sealed xD

Of course if it those RMA responses were not fabricated*, proof altered*

An additional instance of a RMA communication from hardforum, a while after the one above popped on reddit, also mentioned showed talkative Seasonic reps it's just a cable change, and that's why that person got the same PSU. (after that the next dozen and half (though only 1 or 2 more provided picture proof of RMA communication), only mentioned stock responses

Likely, because these 3 or so were posted on internet This isn't he first time, reps within company going above&beyond on customer satisfaction, just for it to stop abruptly due to afformentoned customers posting those instances (for all reasons from completely innocent to everything else on the scale) in great detail on the internet (thus breaking the unspoken, unwritten social contract off "if you get above&beyond with a public company, do not advertise, do not brag, do not post in great detail, be happy, keep shush, or at most just commend in vaguer terms, without posting detailed proof"

 

[Always Counterclockwise]

makes the Supervisor IC go haywire

Do you have the part number for the IC and a citation of a source showing the measurements necessary for diagnosis? I've searched on this in the past but haven't been able to find anything more specific than posts like this one. As someone who's done power supply design with point of load sensing that brief description jonny's used across multiple posts leaves me with more questions than I started with. Haywire isn't a shutdown mechanism and, often, when people use such terms it indicates they've a hypothesis of where the problem might be but lack the detail to confirm a specific mechanism of action.

If supervisor means voltage supervisor here then, yes, it would be OVP/UVP rather than OCP. Because cutting the sense line opens the supply's outer feedback loop enough things change in the circuit it's not much of a diagnostic, though. I can say that usually when I've seen problems in this direction the supply protection is functioning as designed. While I'm not speaking from ATX power supply design experience, fixes are usually small enough engineering changes it surprises me Seasonic still seems to be struggling after 1.5+ years.

 

[Dogzilla07]

Do you have the part number for the IC and a citation of a source showing the measurements necessary for diagnosis?

It's on Jonnyguru video/site, there's also a couple of posts on LinusTechTips forum. (maybe I have them bookmarked, they're pretty buried across 2 years of posts and thousands of posts, i'll check). Additionally, it should be the one on cybenetics/tomshardware/techpowerup reviews, but Jonnyguru wrote once, that Aris sometimes copy-pastes it, so I don't know the exact accuracy.

here's the footnote on the cultists tier list (summary of how Jonnyguru explains I guess + possible their cross-referencing):

"Seasonic PRIME based units experience shutdowns with RTX3080/3090 (and possibly RX6900 XT) GPUs. The cause is not the OCP tripping but a PSU design flaw as evident by the PSU not latching off on shutdown and 1000W+ models being affected too. Doesn’t manifest in 100% cases as it’s also dependent on motherboard model and GPU OC. A DIY fix is possible, of disconnecting a specific pin on the PSU-side of connector 24-pin motherboard cable. Sub 850W Prime GX/PX models are not affected since they’re based on Focus platform. There’s no public information from Seasonic on the issue and whether it was fixed but it was confirmed by JonnyGURU to be present at least in late 2020 revision units and there are user reports with units past this date. "

The EVGA G6/P6 have a different named Supervisor (part number present in the reviews) [This is the new one: "Weltrend WT7527RA (OCP, OVP, UVP, SCP, PG) "]. This should be the old one that gets confused and I guess falsely triggers OCP (Weltrend WT7527V (OVP, UVP, OCP, SCP, PG )

(I really don't remember the specifics, and I did write this in an approximate way, which might have been confusing as I'm not using correct nor precise terminology). It's highly possible it's a falsy OCP triggering [without the current actually redlining].

OVP/UVP rather than OCP.

Jonnyguru specifically mentions everytime when asked, that no protection goes out of spec when this happens. Extra noise builds up on the sense wire and the supervisor IC gets confused and shuts down the PSU.

 

[Always Counterclockwise]

It's on Jonnyguru video/site, there's also a couple of posts on LinusTechTips forum. (maybe I have them bookmarked, they're pretty buried across 2 years of posts and thousands of posts, i'll check).

Thanks! jonnyguru.com remains down, so I suspect any original content on this is probably gone (it doesn't show in the last snapshot of reviews or news and web.archive.org didn't copy the forums). I've gotten no hits searching through jonnyguru's twitter and have the same burial problem trying to find anything at Linus, even when attempting fairly targeted searches. There might be specifics somewhere in one of GN's chats with Jon but usually they don't get that detailed (and it's rare I reach the level of boredom needed to go through that much video looking for one specific thing ).

Extra noise builds up on the sense wire and the supervisor IC gets confused and shuts down the PSU.

Taken literally, that's unlikely as the sense wire really should be a low impedance net. If anyone out of Nvida, 30 series AIBs, or Seasonic did something to convert parts of the sense route high impedance that's, well, frankly, incompetent. And, even then, it's still pretty weird if they got the net the IC pin's on to pump.

I could see some of the undergraduate electrical engineering students I've taught stumbling into something like this but it's strange to see it out of a major OEM. So I'm left feeling like things don't add up, though it may never be clear what actually happened or why.

 

[Sylas]

Yes, sometimes a GPU can fail, but it really isn't that often.

I have found 3 usual reasons why shutdown happens..... and they are all PSU related.

One is the current PSU just doesn't "play well" with the 3080/90 series GPUs based on transient power spikes, gamersnexus does an excellent job explaining this problem.

The Brewing Problem with GPU Power Design | Transients - YouTube

The second is the PSU is just underpowered for the 3080/90 GPUs. Personally, I wouldn't recommend anything under 1000W. There are a number of really good PSUs to consider, I like the Be Quiet Dark Power Pro 12 PSUs the 1200W and 1500W are f'n awesome.

And the third is a bad or loose PSU wires ... and also it can be caused by poor quality extension cables.

I'd focus on the PSU side of the problem first before getting into a RMA loop on returning and replacing GPUs.

I once had a shutdown issue, and after many days of trying to solve the issue, I discovered that the power switch on my case was shorting out.

 

[Dogzilla07]

Thanks! jonnyguru.com remains down, so I suspect any original content on this is probably gone

I was thinking his personal blog, but it's not there

Jon Gerow's personal website

Also for the youtube it's just 16AWG vs 18AWG testing to show how Corsair pig-tail works with 3090 TI type of power consumption, and how regular 18AWG don't

https://www.youtube.com/c/JonGerow

and have the same burial problem trying to find anything at Linus, even when attempting fairly targeted searches.

Same experience just annoyance and time wasted, LinusTechtips doesn't have the save post (with labels and tags) option like overclock, which means I couldn't find the posts within a reasonable amount of time I'll double check my archived notes, but I don't know if I compiled all the useful posts as I went along, so not holding my breath for it being anything but a giant chore to pull out useful Jonnyguru posts from LinusTech tips.

Taken literally, that's unlikely as the sense wire really should be a low impedance net. If anyone out of Nvida, 30 series AIBs, or Seasonic did something to convert parts of the sense route high impedance that's, well, frankly, incompetent. And, even then, it's still pretty weird if they got the net the IC pin's on to pump.

That's how it was always explained by a few people, bad design on part of NVidia and AIBs, the noise feedback loop keeps getting bigger until the IC gets confused. (He mentioned Seasonic used a few old components (specifically this supervisor IC) from the KM2 platform, and that the old X-series had same issues. He noticed it while testing Seasonic based AX series (Which got an early retirement due to broken cooperation, for this specific reason).

For the specifics you'd have to ask him, or join the power cultists discord and ask for a summary, link to a discussion there.

There's rumours swirling around Nvidia is using the extra time now, to minimize the "bad behaviour" the new 4000 series will do with regard to power excursions (spikes) and this issue as well. Heck the EVGA cards were so badly designed, New World game killed dozens if not many dozens of them (power virus like code situation). And all double-sied GDDR6X are showing signs of early EOL, due to how wonky the double-sided memory design was in relation to overheating, and how wonky and specific about thermal pad thickness, and softness/hardness the cooling systems were (Buildzoid's video on what miner cards to avoid like a plague, which is anything GDDR6X, double-sided explicitly, HBM memory, or generally with an interposer):

 

[Always Counterclockwise]

There's rumours swirling around Nvidia

Yeah, world+dog has been unhappy with the state of GPUs for a while. I'm no exception and wish AMD luck in catching up to Nvidia in AI. All indications are Navi 2 would run our workloads more quickly, cooler, and more reliably than Ampere if the low level parts of the software stack we need were in place and solid. They're not, but it'd be nice to have a choice.

That's how it was always explained by a few people, bad design on part of NVidia and AIBs, the noise feedback loop keeps getting bigger until the IC gets confused.

Which also doesn't make sense. Nothing in the circuit geometry is changing, so the loop isn't going to change its physical or electrical size. Maybe I'll find time for further digging but, I dunno, we don't have any Seasonics and don't need to buy any in the foreseeable future.