New build with an RTX 3080 - constant spikes/stuttering/clock drops with any game

[mattcaves]

Hey wonderful people! if anyone with experience would be able to help me pin down the problem with my new rig I'd very greatly appreciate it, I haven't been able to work this week

this problem doesn't occur with a capped framerate but I'm doing some real-time graphics work and I often need to keep things running at max when testing.

Basically when running any 3d application with an uncapped framerate I'm getting huge spikes, a stable 170 frames would go down to 15 for a few seconds, and this happens every 8 seconds or so. GPU-Z stats are telling me that all temperatures are fine, and the PerfCap was triggered by VRel Vop, or not enough voltage.

Here are some threads with the same issue that weren't resolved:
https://forums.tomshardware.com/thr...d-games-stutter-then-come-back-again.3753122/
https://www.techpowerup.com/forums/threads/rtx-3090-unstable-performance.286810/page-2
https://linustechtips.com/topic/1437891-rtx3070-fps-drops-and-stuttering/

Things I've tried

• Swapped out GPU for a different model
• Upgraded Power supply to 850W gold
• updating bios/windows/drivers
• changed pcie to gen 3 in bios
• undervolting GPU

Specs

• MSI Ventus3x RTX 3080 10GB
• Ryzen 9 5900X
• 32GB ddr4
• Biostar b550mh mobo
• 2tb m.2 gen 4 ssd

What I might look into

• downgrading to a weaker GPU (not preferable)
• swapping out other parts, in case there's some sort of bottleneck
• other bios tweaks

Any help or consideration is very much appreciated!

 

[Mussels]

This isnt going to be an issue with the GPU itself

It's either a software issue (background program screwing with things)
or a config issue in software or hardware - such as hitting Vsync limits, VRM's for the CPU thermal throttling, other overheating issue


the unlimited FPS going up and down tends to imply that something else is happening in the background - you got anything like wallpaper engine or rainmeter running? RGB software with ambient lighting?

Run HWinfo in sensors only mode, run a problematic game for 5 minutes and take a screenshot of everything at once - show us everything

 

[mattcaves]

Hey thanks for responding! here's a screenshot, sorry the window is cropped I can send more info if you need.
Something I just noticed from hwinfo is that the CPU clock also drops super low during the spikes.
but yeah I've tried to minimise background apps and no rgb's or unnesscesary components drawing power.

 

[Chrispy_]

Yeah, this is software. If it was hardware the period of each fps drop would be less regular, since 3D workloads are rarely constant unless you are testing a static scene. That variance in the workload means that any thermal/voltage/TDP limits you were bumping into would also vary in their timing and threshold.

If you haven't already, DDU, and clean-install the latest Nvidia driver. Also, look at task manager and see if anything else running has activity in the GPU column of the performance tab.

 

[mattcaves]

Yeah, this is software. If it was hardware the period of each fps drop would be less regular, since 3D workloads are rarely constant unless you are testing a static scene. That variance in the workload means that any thermal/voltage/TDP limits you were bumping into would also vary in their timing and threshold.

If you haven't already, DDU, and clean-install the latest Nvidia driver. Also, look at task manager and see if anything else running has activity in the GPU column of the performance tab.

This makes me think it is hardware, because the spikes are usually a lot less regular, especially if the pc has just been booted up, and the problem is mitigated if I use a framerate cap. Sorry I forgot to mention I've already DDU'd and tried several nvidia drivers, and then reverted back to the newest. someone suggested on another forum that was the low end motherboard (biostar m550h) that doesn't have a proper temp reg solution which could lead to the CPU throttling. I'll try out any other software options but I'm leaning towards getting the motherboard replaced

 

[Chrispy_]

If you want to confirm or rule out thermal, Furkmark it and CPUBurner it (included with Furmark). Keep monitoring with GPU-Z and Ryzen Master IMO to see if you're hitting limits for temperature or power on either GPU or CPU.

Realistically, if you have the time and a spare SSD kicking around, I'd swap to a spare drive, kick off a clean install of Windows and test again. It sounds like a big deal but installing a new drive is a 10 minute job at most, installing windows only actually takes 5 minutes at the start and 5 minutes at the end. The rest of it is just progress bar you can walk away from.

So that's 20 minutes to a fresh Windows install without affecting your existing install in any way. Worth a shot, given that you've probably spent more time than that gathering screenshots and writing your three posts here on TPU forums

Just in case you're not a veteran of multi-boot, or multi-drive OS installs, I'd recommend physically disconnecting or removing your original windows drive to avoid the new install modifying your existing Windows Boot Manager. Otherwise, fixing that typically requires about 10-15 minutes of trying various command-line Bootrec and BCDEDIT works to work out and then repair exactly which of many possibilities you need to recover your original OS to a bootable state later.

 

[Hofnaerrchen]

If you want to rule out hardware issues without doing a complete install use a Linux distro from an USB-stick and run some tests there.

 

[mattcaves]

If you want to confirm or rule out thermal, Furkmark it and CPUBurner it (included with Furmark). Keep monitoring with GPU-Z and Ryzen Master IMO to see if you're hitting limits for temperature or power on either GPU or CPU.

Realistically, if you have the time and a spare SSD kicking around, I'd swap to a spare drive, kick off a clean install of Windows and test again. It sounds like a big deal but installing a new drive is a 10 minute job at most, installing windows only actually takes 5 minutes at the start and 5 minutes at the end. The rest of it is just progress bar you can walk away from.

So that's 20 minutes to a fresh Windows install without affecting your existing install in any way. Worth a shot, given that you've probably spent more time than that gathering screenshots and writing your three posts here on TPU forums

Just in case you're not a veteran of multi-boot, or multi-drive OS installs, I'd recommend physically disconnecting or removing your original windows drive to avoid the new install modifying your existing Windows Boot Manager. Otherwise, fixing that typically requires about 10-15 minutes of trying various command-line Bootrec and BCDEDIT works to work out and then repair exactly which of many possibilities you need to recover your original OS to a bootable state later.

Just formatted the drive and gave it a fresh win 11 install, installed all nescessary software, unfortnatley the issues are still there.

Really appreciate the time taken to help me out though! I'll probably just do a quick motherboard replacement and report back

 

[Hofnaerrchen]

Did someone already mention the GPU memory temperatures? Might be just a bad mounted GPU cooler.

is about a similar issue.

 

[Chrispy_]

The VRAM is hot. looks like it's 88C at the bottom of one of it's up and down waves. I believe 95C is the limit for GDDR6 and 105 the limit for 6X.

 

[mattcaves]

It was the motherboard! it's all working well now, thanks for the help!

for those scrolling through, If you're getting a ridiculously powerful gpu then don't skimp out on a motherboard.

 

[Chrispy_]

It was the motherboard! it's all working well now, thanks for the help!

for those scrolling through, If you're getting a ridiculously powerful gpu then don't skimp out on a motherboard.

Wow, that is some bargain basement motherboard there, I didn't know anyone made a B550 board with such weak VRMs!

Honestly, the MSI Pro VDH is pretty weak, but this thing is a 5+3 phase design that's clearly aimed at 65W APUs, Not only is it at least 3 phases short of most other designs, it also has NO heatsinks on those VRMs

Glad you found the culprit, anyhow!

 

[Mussels]

You can click the arrows in the bottom left of HWinfo to show more on the screen

Heres how mine looks (Customised and some values disabled since they used some CPU cycles)


It's everything as a whole that needs to be seen, for example seeing a 90C maximum on the CPU could be worrying - but if we see the CPU was using 150W at the time, it becomes normal

The motherboard itself doesnt affect performance unless something is physically wrong, even an A320 can power a 3090 and 5950x without issue

I'd be checking the thermal pads on the VRM's and under the chipset heatsink
If it has no VRM cooling, then you need to make sure PBO is off or set custom safe values if you re-use the board