EVGA 3080ti crashing w/in 20 mins in GPU intensive games, but only in my new AMD build despite running cooler

[rickyrich93]

I just put together a spanking new AMD Ryzen 7 7800X3D build thanks to some amazing prime day deals I found on amazon, newegg, and Micro Center. I bringing my EVGA Supernova 850w power supply and my EVGA 3080ti FTW3 from my previous intel build into the new AMD build. I bought a new MSI 750w power supply to replace the 850w PSU I had in the old intel build that will now solely function as a plex server.

In GPU intensive games (God of War Ragnarok and Horizon Forbidden West at 4K max settings and frame rates unlocked) my new AMD build will crash to desktop within 20 minutes of gameplay. At first I thought I had connected the GPU incorrectly in the new build, but all the connections are secure on the motherboard, the GPU, and PSU. I did try daisy-chaining two of the three 8-pin GPU connectors to see if that configuration worked better as opposed to the 3 separate 8-pin connectors as I had it originally, but neither connection method proved worse or better than the other as far as stopping crashes within 20 minutes of gameplay. I thought perhaps the GPU was overheating but the temperatures are actually 7-8 degrees cooler in the new Antec case I bought. GPU Temps during crash (Chip=80C, Hotspot=86C, VRAM 88C) are all within spec for this card that is know for running very hot. In fact, it has run hot since the very day I bought it. This is the stock fan curve by the way! I did install fan control to make sure my case fans are cooling the GPU properly and the system is fairly loud now and it's still crashing.

I booted into safe mode and ran DDU and removed all the graphic drivers I could find from NVDIA and AMD, performed a clean installation of the newest Nvidia drivers and it would still crash. I went back and repeated this process for the Nvidia Studio drivers, and even an older Nvidia driver recommended by a Gamers Nexus video (566.36) and it would still crash within 20 minutes.

Because I feared the EVGA PSU might suddenly have become faulty in the hour it took me to move it out of the old build to the new build, I connected the MSI 750w PSU that I had bought to go into the old intel build as a replacement. I can do math, I know a 750w power supply is 100w lower than the 850w PSU that I was already using, but I also know that 750w is considered the minimum for my CPU/GPU combination. Surely enough, it crashed within 20 minutes, so I don't think it's a PSU issue. I even made sure the Eco mode switch is OFF on the back of the PSU.

To go even further, I moved the EVGA power supply and the 3080ti back into my previous Intel i7 9700K build and ran the same test (booting up GOW Ragnarok at full 4K max settings with frames unlocked). Even though the GPU ran MUCH hotter (88C on the chip) in the old NZXT case, it ran for hours with zero crashes. So, because this GPU can run way hotter and longer than its running in its new case, I don't believe GPU thermals are an issue.

I ran a windows memory diagnostic test, and my RAM passed it. I'm running a Passmark MemTest86 as I write this but its been going on for over an hour and half and I'm on pass 3/4 and so far zero errors. The RAM kit is shown as compatible in the memory QVL at the rated speeds and timings.

Keep in mind, this is at stock settings for both the CPU and GPU. I have no overclocks nor undervolts applied to either build or system. This is stock stock stock. I do have XMP/EXPO and re-bar enabled on both systems. The only real thing different is the platform that causes the crash. I'm not very familiar with the Asrock AMD BIOS settings but I did manage to find a setting to lower the PCIE generation on the slot from gen 5 all the way down to gen 3, but doing so didn't improve anything either.

I'm on the most recent BIOS versions for both motherboards on both systems. This is a fresh Windows 11 install (three times now lol) on a brand new Western Digital nvme drive. I've updated all the drivers on BOTH systems. What am I missing? Is there a clear answer here? Troubleshooting can be fun but right now it's getting really aggravating. I don't think it's a GPU thermal thing as I said above because the GPU is running around 8 degrees cooler in the new system and is crashing within minutes.

Old Build: Intel i7 9700K, Gigabyte Z390 Aorus Pro Wifi, 32GB G. Skill Ripjaws 3200mt/s (4x8GB kit), Intel 660 SSD, NZXT H5 case.
New Build: AMD Ryzen 7 7800X3D, Asrock X870E Nova Wifi, Teamgroup T-Create 32gb 6000mt/s (2x16gb kit), Western Digital Sn850x nvme, Antec Flux case.

TLDR: GPU crashes despite better cooling in new build, PSU and thermals are likely not at fault. Nvidia drivers not at play. Possible BIOS setting?

 

[ir_cow]

1) If you aren't on the newest motherboard BIOS, update.

2) Passmark MemTest86 isn't a good test. Try Memtest5 https://www.overclock.net/threads/memory-testing-with-testmem5-tm5-with-custom-configs.1751608/

3) I would have said that EVGA Supernova 850w wasn't up to the amp spikes NIVIDIA cards do, but since it works fine at full load for the old system, hard to see that being the issue.

 

[rickyrich93]

1) If you aren't on the newest motherboard BIOS, update.

2) Passmark MemTest86 isn't a good test. Try Memtest5 https://www.overclock.net/threads/memory-testing-with-testmem5-tm5-with-custom-configs.1751608/

3) I would have said that EVGA Supernova 850w wasn't up to the amp spikes NIVIDIA cards do, but since it works fine at full load for the old system, hard to see that being the issue.

1) I'm on the most recent BIOS version
2) Memtest86 has been recommended by many and surely must be capable enough of finding a bad dimm if that's an issue. I'm also over 3 hours into this test and i've come too far to just back out of it now to run another memtest
3) I agree that the PSU likely isn't the issue

 

[Sammoonryong]

Does your GPU work with stock settings? What is the initial GPU core frequency and voltage? What’s the operating frequency and voltage of the GPU core? Is VRAM overclocked? I had similar problem but I overclocked and undervolted my 3080. It worked fine at 1770MHz than it would rise to 1800MHz (core temps were low) but if the core tried to boost to 1830MHz at 0.806v I had a crash. Quick test that may help you. Install MSI Afterburner and don’t touch anything but TDP lower it to something ridiculous like 60% and see what happens after 20 min.

 

[rickyrich93]

I will try that test. Everything is stock as I mentioned in the post. What am I looking for in 20minutes? What will a crash tell me?

 

[taka]

Did you re-install windows or just moved the disk from Intel to AMD?

 

[rickyrich93]

Did you re-install windows or just moved the disk from Intel to AMD?

As I stated in the post it is a new windows install off a brand new nvme drive

 

[rv8000]

What memory kit are you using exactly (model number, i may have that kit laying around)? It may be just an ASUS thing but enabling expo on my 8000 kit will set the FCLK to 2100mhz instead of 2000mhz on one of the newer bios. Check to ensure that if you’re running 6000 mt/s ddr5 your fclk (infinity fabric) is running at 2000mhz 1:1.

Id also try running the game with the expo/xmp profile disabled and the memory running at jedec standards. This could help rule out memory or IF clock instability.

 

[ir_cow]

2) Memtest86 has been recommended by many and surely must be capable enough of finding a bad dimm if that's an issue. I'm also over 3 hours into this test and i've come too far to just back out of it now to run another memtest

Nope. Memtest is weak. Doesn't stress the memory enough from my testing.

@rv8000 has the right idea. Try and run without XMP/EXPO.

 

[wNotyarD]

Out of curiosity:

What framerates do you achieve in each build? Because the new Ryzen surely gives the 3080Ti more leeway to run free to the point it trips.

If anything, I'd say to try limiting the framerate while using the Ryzen and see if it crashes as well.

 

[rickyrich93]

What memory kit are you using exactly (model number, i may have that kit laying around)? It may be just an ASUS thing but enabling expo on my 8000 kit will set the FCLK to 2100mhz instead of 2000mhz on one of the newer bios. Check to ensure that if you’re running 6000 mt/s ddr5 your fclk (infinity fabric) is running at 2000mhz 1:1.

Id also try running the game with the expo/xmp profile disabled and the memory running at jedec standards. This could help rule out memory or IF clock instability.

The kit is Teamgroup T-create 32gb kit DDR5 6000 CL30-36-36-76 1.35v. So first things first: After running memtest86 and getting a full pass, i went back into my BIOS to switch it to DDR5 defaults and test things out that, only I realized I had XMP enabled the entire time instead of EXPO in my DRAM profile. I was so happy to catch my mistake! Unfortunately, even after switching it to Expo profile I’m still crashing in God of War Ragnarok in 15 minutes or less. Anyways despite my mistake it seems expo is no good either. Switching to the defaults ram speed of 4800 the game doesn’t crash at all. Does this mean the ram is bad or the profile is buggy?

 

[dont whant to set it"']

It could be PSU/ GPU cable/s rail distribution.
Because now you GPU can churn out more frames because it is fed more frames by a better CPU.

 

[Sammoonryong]

I will try that test. Everything is stock as I mentioned in the post. What am I looking for in 20minutes? What will a crash tell me?

My working theory is that your GPU might not be able to operate at stock settings anymore due to degradation, old age or some other unknown reason. Lowering the TDP will drastically reduce the clocks and allow us to confirm/exclude this. Did you try to put the card in other PCI-E slot?

 

[rickyrich93]

My working theory is that your GPU might not be able to operate at stock settings anymore due to degradation, old age or some other unknown reason. Lowering the TDP will drastically reduce the clocks and allow us to confirm/exclude this. Did you try to put the card in other PCI-E slot?

No i didn’t try a different slot, but I moved it back into my previous build where it can run even hotter (due to the older case) for even longer without crashing. Frame rates are virtually identical between the two rigs especially since I’m running 4K max settings frames unlocked. Not a very CPU bottlenecked situation and no crashes in the old build

 

[ir_cow]

Put the GPU back in your old machine and run MSI Kombuster. That doesn't care what system it's in. GPU will be at 100% load with near zero CPU. That will at least eliminate the GPU.

Still everything else is replaced in the new build.

Download MSI Kombustor v4.1.30 - Benchmarking Tool

Download MSI Kombustor for PC to thoroughly test GPU performance by pushing it to its limits. This free utility program is designed for Windows users to conduct graphics card stress tests. DOWNLOAD Simple & Easy To Use The software interface is very easy to use and simple. You don’t need to be a...

msikombustor.com

 

[rickyrich93]

Put the GPU back in your old machine and run MSI Kombuster. That doesn't care what system it's in. GPU will be at 100% load with near zero CPU. That will at least eliminate the GPU.

Still everything else is replaced in the new build.

Download MSI Kombustor v4.1.30 - Benchmarking Tool

Download MSI Kombustor for PC to thoroughly test GPU performance by pushing it to its limits. This free utility program is designed for Windows users to conduct graphics card stress tests. DOWNLOAD Simple & Easy To Use The software interface is very easy to use and simple. You don’t need to be a...

msikombustor.com

It’s not the GPU read my previous comment

 

[nomdeplume]

2) Memtest86 has been recommended by many and surely must be capable enough of finding a bad dimm if that's an issue. I'm also over 3 hours into this test and i've come too far to just back out of it now to run another memtest

BZZZTT - wrong answer

Nope. Memtest is weak. Doesn't stress the memory enough from my testing.

@rv8000 has the right idea. Try and run without XMP/EXPO.

The test that isolates the problem can be strong/weak/outdated/alpha build... we deal with solutions instead of stigma.

Start with forcing and reforcing BIOS settings and then move on to testing hardware.

 

[_roman_]

Keep in mind, this is at stock settings for both the CPU and GPU. I have no overclocks

you do

T-Create 32gb 6000mt/s (2x16gb kit)

depends on the ranks - i see only 5200mt/s

https://www.amd.com/en/products/processors/desktops/ryzen/7000-series/amd-ryzen-7-7800x3d.html

do not choose a psu by watt numbers only

you may share screenshot of windows software which includes all dram and mainboard settings.

 

[ir_cow]

It’s not the GPU read my previous comment

Eh. Just trying to help.

This is what I would do if it was my system and me being lazy.

1) run memtest5 with XMP/EXPO
2) run MSI Kombuster. If it crashes, put in other computer. Still crashes. Either it's the PSU or GPU. Hard to say, but since it never crashed on the old PSU before, probably the GPU at that point .

If neither fails, its something else causing the games to crash. Just saved you hours of pointless troubleshooting in the wrong direction.

 

[rickyrich93]

Eh. Just trying to help.

This is what I would do if it was my system and me being lazy.

1) run memtest5 with XMP/EXPO
2) run MSI Kombuster. If it crashes, put in other computer. Still crashes. Either it's the PSU or GPU. Hard to say, but since it never crashed on the old PSU before, probably the GPU at that point .

If neither fails, its something else causing the games to crash. Just saved you hours of pointless troubleshooting in the wrong direction.

I think it was ram speeds. At Jedec speeds the game won’t crash but at Xpo speeds it does. I bought a new kit of ram and it won’t load Expo profiles into windows even if they’re selected in the bios

 

[nomdeplume]

I think it was ram speeds. At Jedec speeds the game won’t crash but at Xpo speeds it does. I bought a new kit of ram and it won’t load Expo profiles into windows even if they’re selected in the bios

Did you look if either were listed as compatible on your mobo's support page?

 

[rickyrich93]

Did you look if either were listed as compatible on your mobo's support page?

Correct and yes they were both listed as compatible according to the QVL. However neither is working at the advertised Expo speeds.

 

[nomdeplume]

Correct and yes they were both listed as compatible according to the QVL. However neither is working at the advertised Expo speeds.

So on a new mobo this is either BIOS or defect of some kind. Lots could be at fault.

Among what I'd try until a firm clue or two appears is pulling RAM and do a microscopic inspection of all four channels they fit into to make sure nothing found it's way in. Pull the coin battery and do the most complete CMOS reset possible and then breadboard it with a single stick of RAM.

Barring some clue what is borked following reasonable attempts. I'd personally start heading towards the conclusion mobo needs to be exchanged or returned. Stress tests have their place once you have a solid platform.

 

[rv8000]

The kit is Teamgroup T-create 32gb kit DDR5 6000 CL30-36-36-76 1.35v. So first things first: After running memtest86 and getting a full pass, i went back into my BIOS to switch it to DDR5 defaults and test things out that, only I realized I had XMP enabled the entire time instead of EXPO in my DRAM profile. I was so happy to catch my mistake! Unfortunately, even after switching it to Expo profile I’m still crashing in God of War Ragnarok in 15 minutes or less. Anyways despite my mistake it seems expo is no good either. Switching to the defaults ram speed of 4800 the game doesn’t crash at all. Does this mean the ram is bad or the profile is buggy?

It could mean several things.

1) Bad timings or requires more vdd/vddq
2) Requires more vsoc if you have a weak imc
3) Your CPU is unstable at fclk of 2000 or whatever your motherboard is forcing for fclk when your expo profile is enabled > requires vsoc adjustement or vddg adjustment. Your board could be suppling to little or too much of several different voltages.

It might be worth enabling expo again and setting vdd/vddq to 1.4v and seeing if that helps. Either way looks like your running into a stability issue between the memory, your fclk, or an EXTREMELY bad CPU sample (weak fclk/imc).

 

[rickyrich93]

It could mean several things.

1) Bad timings or requires more vdd/vddq
2) Requires more vsoc if you have a weak imc
3) Your CPU is unstable at fclk of 2000 or whatever your motherboard is forcing for fclk when your expo profile is enabled > requires vsoc adjustement or vddg adjustment. Your board could be suppling to little or too much of several different voltages.

It might be worth enabling expo again and setting vdd/vddq to 1.4v and seeing if that helps. Either way looks like your running into a stability issue between the memory, your fclk, or an EXTREMELY bad CPU sample (weak flck/imc).

Yup I think you’re 100% correct. These crashes appear to be a memory stability thing. Someone recommended Event Viewer and I found errors related to memory access which corroborate this and at default JEDEC speeds I have no crashes. Maybe my CPU’s memory controller just isn’t high quality enough to run at 6000mt/s or maybe I just need more voltage on the RAM. Memory overlocking (beyond simply enabling XMP or EXPO) is still outside my wheelhouse and understanding all the acronyms is still something I’m working on.

So on a new mobo this is either BIOS or defect of some kind. Lots could be at fault.

Among what I'd try until a firm clue or two appears is pulling RAM and do a microscopic inspection of all four channels they fit into to make sure nothing found it's way in. Pull the coin battery and do the most complete CMOS reset possible and then breadboard it with a single stick of RAM.

Barring some clue what is borked following reasonable attempts. I'd personally start heading towards the conclusion mobo needs to be exchanged or returned. Stress tests have their place once you have a solid platform.

It’s possible it’s a BIOS thing on Asrock’s side. It could also be the CPU’s IMC. The Teamgroup RAM has a second EXPO profile of 5600mt/s that seems stable so far.