3080 Ti transient spikes triggering OCP on a 1000W PSU?

[Kashinoda]

I've been trying to play The Finals but cannot make it 5 minutes through a match without the computer completely shutting down/power cycling. There's no BSOD and the only thing I see in event viewer is unexpected shutdown (Kernal-Power).

Specs:

3080 Ti FE​

5900X​

MSI B45M Mortar (Original not MAX)​

Crucial Ballistix Sport LT 2x 16GB 3200Mhz​

Seasonic Focus GX 1000W​

This was happening more sparingly with Sons of the Forest a while ago but I let it go.

What I've tried

• Resetting BIOS to stock defaults
• Updating BIOS to latest
• Lower RAM speeds and looser timings
• Different RAM entirely (some budget Crucial 2400mhz stuff)
• DDU removal tool + several different driver versions

It's not an overheating issue, I've can run MSI Kombustor (Furmark) and AIDA64 stability test (CPU, FPU, Cache, Memory) concurrently for hours without issue. Similarly no issues found running OCCT stability certificates.

What did work in the end was frame limiting the game to 60FPS in nVidia control panel and power limiting the card to 80% from default. Note I usually undervolt with a 92% power limit anyway and it did crash then (as well as crashing at default curves)

This was the last thing I checked as even with the issues on 3000 series cards a 1000W PSU should have the headroom. Am I missing something?

 

[Kashinoda]

Yeah it's this one:

Seasonic Focus GX-1000 1000W 80+ Gold Modular Power Supply

Order Seasonic Focus GX-1000 1000W 80+ Gold Modular Power Supply now online and benefit from fast delivery.

www.overclockers.co.uk

I'll check the cabling now.

 

[Kashinoda]

Purchased November 2021, here's the cabling:

do your case fan ramp up when it does power cycle's/restarts?

Nope it just switches off (well, restarts), I did try setting HWINFO64 to 20ms polling rate and logged the output but can't see anything obvious. That's not going to pick up a transient but did confirm otherwise everything looked normal moments before.

 

[Kashinoda]

Is it just a buggy game if it's only happening in that specific scenario?

I mean I can't find anyone else with this particular issue and it was happening with Sons Of The Forest a while ago. It was charactistic of the time I tried to run this setup with a BeQuiet 650W PSU and had to absolutely throttle the card to prevent OCP on any game.

First thing I’d check would be Nvidia’s stupid power connector.

Well it 'looks' fine, I.e. no melted plastic, pins look good, I can't see anyhing obvious. Regardless I've re-seated the connector and flipped the CPU and GPU connectors on the PSU. Just crashed within a minute.

could be the psu, linus uploaded a new video about his personal rig that would just switch off and be very weird trying to power up, he had a seasonic psu too not sure what model, but was replaced with an asus thor when they tested the old unit it would fail at around 100w load, i'm sure his mobo was replaced and when all done it was working correctly.

so it could indeed be a psu problem that may have been caused by a motherboard fault?

Yeah I'm just going to buy another 1000w\1200w PSU so I can at least be certain. Maybe the motherboard is a bit knackered too, I mean it's a budget board made for the Ryzan 3xxxx but it's been otherwise solid.

 

[Kashinoda]

Okay so I get the same issue in Battlefield 2042 with a power cycle within 5-10 minutes. It's also happening in Balder's Gate 3 but very occasionally.

• I replaced my Focus GX with a Corsair RM1200x and the problem actually got worse, I couldn't even load into a game of Battlefield before the PC power cycled, it would just happen in the menus.
• I swapped my 3080 Ti with my housemate's 3080 and bizarrely neither of us had any issues, he also has 5900X but only a 750W PSU. We matched settings best we could, so I don't think it's the GPU.

This has left me confused, it points to the combination of GPU, Motherboard and CPU but I'm not sure why or how? Sure enough, if I lock my CPU to 4.25GHz (all core) it completely fixes the issue - presumably due to the lack of 'bursty' behavior.

I'm sort of okay with accepting the motherboard is knackered, but I'm not sure why this issue is present with my 3080 Ti but was unable to be replicated with another 3080, they both take 75W from the PCI-E slot?

 

[Kashinoda]

I've had shutdowns like this when I had my 3950x, swapping to a 5800x sorted it

It could be the motherboard vrm is getting overwhelmed when you let the 5900x boost to its max frequency, does your housemate have the same board as you or a better model with a beefier vrm, limiting core frequency to fix the problem sugests your board isnt upto the task with a 12 core, you could opt for a 5800x or a x3d as above to fix the problem without changing your motherboard.

Yes he has a all-singing X570 whilst I'm still rocking a B450 . I guess puzzling thing for me is my motherboard seems to keep up when a slightly less demanding (albeit only by 20W) GPU is used. Otherwise the explanation makes complete sense.

I guess there are smaller variables at play which change the output that I haven't thought of. I'll keep it capped for now and use this as excuse to upgrade to a 7800X3D.

Thanks for the replies!

 

[Kashinoda]

Just had a look at some reviews and it appears to be an excellent PSU.

I would look at RMAing the PSU as it shouldn't have any issues running even a high end RTX 3000 series and serious transient loads. See if that fixes things.

I see you did do a power reduction of the GPU. Can you install MSI Afterburner and see if reducing the boosting frequency helps, not just reducing the power limit in the control panel?

It may have got lost in the thread but I did buy a new PSU as I was convinced it was transient spikes causing problems, however the power cycling still happened. That coupled with the GPU running fine on my friends 750W system means I'm discounting GPU issues. Incidentally I always run my card with an undervolt curve and a power limit of 90%.

When I installed a 5950x in a B450 Tomahawk max it shut down when I ran cinebench. If I recall correctly I fixed it by manually setting the PPT TDC EDC values to stock in the BIOS (142W 95A 140A). I assume this was because these were by default incorrectly set higher by the motherboard and it couldn't handle it.

Ah that'll be why most x570 haave a far better vrm to deal with when the cpu pulls 241w, The 12 and 16 cores have power limits that go that high, The 6-8 core chips will max out around 100-150w when hit with a all core load.

But as you say it'll give you the incentive to look for a am5 upgrade, just be sure to look for a denecnt b650 with a decent vrm so it can handle what ever cpu you install

Manually setting PPT TDC EDC to stock values didn't do the trick alone but that in conjunction with a PBO negative frequency override of 150Mhz and setting the Curve Optimizer (all-core) to -30 seems to have nailed it for now without having any notable impact on performance

The 7800X3D can wait for now.