Need help with crashes and verifying VBIOS on 3090

[PeterEU]

So recently I've bought used RTX 3090 from Inno3d exact model is ichill x4, everything seemed to run fine except I've noticed when i uncap my fps games close after 20 - 60 min after they are opened without any crashlogs or errors (they simply close and desktop is shown)

So i want to eliminate possibility that it's faulty VBIOS so I submitted my VBIOS it seemed to be already in database https://www.techpowerup.com/vgabios/244807/244807 but I am not sure if it's really correct because it's unverified, I'm also attaching gpu-z screenshot if something is needed let me know,
I want to verify it's correct because freshly Installed windows with automatic driver installed (I checked date it was from 2020 most likely release driver) was giving gpu error 43 in device manager newer drivers seem to work fine

My specs are:
GPU: inno3d geforce rtx 3090 x4 (Driver 516.94 but tried older ones too, Stock I haven't applied any OC)
CPU: Intel I7 12700 (non-k)
RAM: A-Data XPG Lancer DDR5-5200 1x16GB (XMP enabled, 1 stick for now)
Motherboard: ROG STRIX B660-I GAMING WIFI
PSU: Corsair SF 750W
Case: MasterBox Nr200p


For now I'm using stock cooler on my CPU with power limited to 75W and boost to 100W in asus bios
Max temps reach 82 C while gaming on CPU and 81 C on GPU
I've bought arctic liquid freezer II 240 and will try it soon too to cross out CPU as cause.
The 2nd thing I suspect is PSU it's only 750W because i was using 3060 ti earlier and I thought 3090 will be fine with it especially when CPU boost power is limited for now from 238W boost to 100W in asus bios giving GPU more room.

Things i've tried already: memtest86 for ram, Furmark for 10 minutes

 

[Mussels]

It's not your VBIOS.
Don't mess with the BIOS without a good reason.

Since you have an SF PSU, i'm just gunna say you're running into the infamous power spikes, and the SF PSU doesnt have the capacity (literally, big enough capacitors) to cover the spikes - a 12700k and a 3090 uncapped is pretty much as power hungry a system as you can possibly get


Oh and 1 stick of ram is absolutely hurting performance, work on that when you can too.



Some cards can do stupid things, while your PSU isn't bad - it's not going to handle spikes like this

Oh and if you're running the GPU vertical mounted, you'll need to drop to PCI-E 3.0 for stability - few risers handle 4.0 well, and often they weaken and crap out over time.



Use an undervolt curve on the 3090 - found 1700Mhz at 750mv got me ~50Mhz lower than stock, but 110W less on average (and without the spikes)
(Ignore right side image, clearly i'm at idle)

 

[PeterEU]

Thanks for quick reply!

I have undervolted my GPU with values you've provided

My default curve looked like this

And now it looks like this

Starts at 1215 mhz and maxes at 1695 mhz hopefully it looks right

Also yes i plan to get 2nd stick of ram I'm just waiting for a good deal, my cpu is capped at 100W turbo and 75W without because I am yet to mount new liquid cooler and with stock one cpu overheats at uncapped power, GPU is also mounted horizontally

I will test the card in few games and write back again

 

[Mussels]

Undervolting ampere is incredibly helpful - i wish we could do BIOS mods and save the values

Between that and your CPU Capped values you should have prevented those millisecond long spikes and solved the issue - you might have spiked to 800W at stock for 5 ms, but now it might only spike to 400W

Pure coincidence, i checked HWinfo for my UPS stats since i normally sit around 400W with a 5800x and 3090 (incl monitors) and today... it went woosh.
Both undervolts are still active, so *something* spiked long enough to register

 

[PeterEU]

Hello again, After some testing yesterday it seemed like crashing was gone, I will keep the GPU undervolted since the temps also dropped by 10 C and I lose only 1-4 FPS , I will also keep CPU power limited to 100W even with new cooler since default 238W would probably make system unstable again with little gains to it.

 

[Mussels]

Hello again, After some testing yesterday it seemed like crashing was gone, I will keep the GPU undervolted since the temps also dropped by 10 C and I lose only 1-4 FPS , I will also keep CPU power limited to 100W even with new cooler since default 238W would probably make system unstable again with little gains to it.

Try 125W on the CPU - from other users testing here on the forum that's become a pretty good limit for gamers, giving the full gaming performance while preventing multi threaded blowouts on wattage

 

[PeterEU]

I've set my long duration power to 125W and short duration to 150W, it seems cinebench likes those values for example when i set it to 100/125W the score was about 19k and 125/150W gives me 21k, with uncapped power limit which goes above 230W its still under 22k, my question is if there is some kind of test to stress both CPU and GPU at the same time without going into game because some games aren't always stressing both CPU and GPU to 100% and if there is one how long should I run it for?

 

[GerKNG]

You run a 3090 on a 750 SFX PSU.
The transients are higher than the whole capacity of the PSU.

 

[dgianstefani]

It's not your VBIOS.
Don't mess with the BIOS without a good reason.

Since you have an SF PSU, i'm just gunna say you're running into the infamous power spikes, and the SF PSU doesnt have the capacity (literally, big enough capacitors) to cover the spikes - a 12700k and a 3090 uncapped is pretty much as power hungry a system as you can possibly get


Oh and 1 stick of ram is absolutely hurting performance, work on that when you can too.



Some cards can do stupid things, while your PSU isn't bad - it's not going to handle spikes like this
View attachment 257826


Oh and if you're running the GPU vertical mounted, you'll need to drop to PCI-E 3.0 for stability - few risers handle 4.0 well, and often they weaken and crap out over time.



Use an undervolt curve on the 3090 - found 1700Mhz at 750mv got me ~50Mhz lower than stock, but 110W less on average (and without the spikes)
(Ignore right side image, clearly i'm at idle)
View attachment 257829

Actually Corsair SF750 PSU can sustain over 950w of power, being the best SFF unit available. Unlikely it's that. I have an overclocked 3080ti under the same PSU and have had zero issues.

 

[Mussels]

Actually Corsair SF750 PSU can sustain over 950w of power, being the best SFF unit available. Unlikely it's that. I have an overclocked 3080ti under the same PSU and have had zero issues.

It's very likely with a 750W SFX, as the primary capacitors are smaller. Sustained isn't like the peaks, there's a lack of time for the PSU to adjust
Also and this is something obvious that people forget - PSU's age. My corsair HX750i conked out after years of use and being fine on my 3090. Went from fine to whine within 2 months.
The problem is that your CPU also spikes at the same times, which is where this problem may occur

MSI afterburners curve undervolting would solve any spikes and likely drastically lower power consumption - on average i'm 50MHz down for 110W less, without any spikes.

 

[PeterEU]

I've been using the undervolt for longer now and i can tell it works because when i turn it off crashes are coming back so I guess it's spikes, I am also using 1440p monitor not 4k so power draw isn't reaching above 300W when i last checked rivatuner overlay, I've tried 8k resolution for fun with custom dsr and the card was drawing lots of power there but i got no intention to go above 1440p

 

[GerKNG]

I've been using the undervolt for longer now and i can tell it works because when i turn it off crashes are coming back so I guess it's spikes

as expected.
in the long run i'd recommend to buy a bigger PSU. (at least 1KW or maybe 1200W if it's just 20 bucks more.)

 

[Mussels]

I've been using the undervolt for longer now and i can tell it works because when i turn it off crashes are coming back so I guess it's spikes, I am also using 1440p monitor not 4k so power draw isn't reaching above 300W when i last checked rivatuner overlay, I've tried 8k resolution for fun with custom dsr and the card was drawing lots of power there but i got no intention to go above 1440p

Lower res can increase power consumption, in a weird twist of logic vs common sense

Because you're more likely to be GPU limited at 4K, the lower FPS means less work for the CPU - lower resolutions means more FPS, so the CPU draws more power to deliver them

Obviously this isn't always the case since any bottleneck anywhere, including Vsync or frame rate limiters can alter that behaviour.

as expected.
in the long run i'd recommend to buy a bigger PSU. (at least 1KW or maybe 1200W if it's just 20 bucks more.)

Being on SFX he may not have much choice for higher wattages, and will likely need to focus on a quality or newer unit

 

[Hyderz]

my 2cents on your 1stick ram... sell that and grab yourself a dual kit
even if you get another identical 1 stick, xmp might not work properly and you have to set it on lower default speed.