GPU Usage, FB Usage, Power all dropping to very low % constantly causing micro stutters and FPS drops [LOT OF DETAIL]

Jun 3, 2018
8
0
10
0
Experiencing microstutter/frame drops in games that cause input lag and choppy fps. Usage during superposition benchmark is much more stable with a couple dips during opening or changing scenes/ I mainly play fortnite at the moment but first noticed it in warface months back (when I was running previous driver and old install of windows 7). FPS stays in the 180-200 range and will plummet at the same time my GPU usage drops. The drops cause it to drop to a single digit percentage, sometimes to 0%. The drop only lasts a fraction of a second and it returns to it’s previous load, but very shortly after that it drops again. Recording the graph of a game results in very little stable load percentages with drops happening constantly. My gameplay suffers often but not always that noticeably with the drops. In the graphs the only other element that follows the same pattern are FB Usage and power %. Voltage, gpu and cpu temps, cpu core load, everything else looks normal to me

My rig
Gigabyte p55m-UD4 mobo
- Latest intel series 5 chipset driver from here
- Lan, Audio, Intel rapid storage technology, and BIOS update from here
Western Digital WD10001-FALS
- Passed seagate tests
Intel i7 860 4 core @ 2.8 ghz
- Overclocked to 4.09 ghz (195x21)
- vcore @ 1.35
- vtt @ 1.37
- pll @ 1.88
- Have tried CPU in both stock and overclocked settings
- Doesn't seem to be evidence of throttling
8g 1600 mhz DDR3 ram in dual channel mode
- running at 1560 mhz
- timings set to manufactures specs 11-11-11
PNY NVIDIA GTX 970
- Stock clock and voltages
- Drivers from 05/24/2018
- Have tried older driver from 05/08 and problem persists
- stock clock and voltages
Corsair TX650W PSU
- I’ve had the computer 2011 ish and this is the original PSU that I purchased when I built it. I’ve read some issues where the PSU wouldn’t supply stable power and cause similar issues but I’ve also read things that debunk that. Currently wondering if this is the issue. Power usage in graph spikes make me wonder if the low power usage is causing gpu drops or if the power is dropping because the gpu is dropping.
Monitor is older Acer H23HD5
- 60 hz refresh
- HDMI connection

Idle Temps
- CPU Cores at mid 30's
- GPU at 40

Common Settings
NVIDIA panel
- Power management set to highest performance
- physx set to operate on gpu
- vsync is off
- Texture quality set to performance
Geforce Experience
- Overlay is turned off
Windows Settings
- Power management is custom, set to performance, and turned off all individual power saving options in the advanced tab
Other Settings
- Have tried capping framerates to 60
- Have tried with VSYNC on and off
- Have adjusted max pre rendered frames in nvidia from application controlled, to 1, to 4. 4 was the worst
- DPCLantecy was stable at around 100

Hardware Monitor Logs
- HWInfo Log as a Google Sheet. Lines highlighted with red feature large discrepencies in GPU usage.
- Imgur gallery of MSI Afterburner Hardware Monitor
- Both of these from the same time frame containing two games of fortnite

Other Things:
- No background apps running. On a fresh install of windows with just drivers, hardware monitoring software, and the game
- Stutters do not happen using superposition to stress my gpu. There are a couple dips during the blackouts in between scenes and maybe one stutter but overall the stress usage is stable.
- I have freshly installed windows 10 and this problem persists
- I have tried most solutions I've found from googling and would appreciate some other ideas or solutions.
- I don't have any other monitors, PSU's, or motherboards to test
- The closest solution I found with similar symptoms was regarding a program they had installed called advanced system care that had a "clean ram" setting enabled and would throttle the ram at 50-60% which would slow down the card. I do not have this program installed.

 

fry178

Reputable
Dec 14, 2015
690
0
5,360
135
what res?

(pll to 1.9 ?!)

unless you have an FE/blower style fan, use afterburner to max out the slider for power and temp, no additional voltage or clocks.

set pre-render to max 3 for any game, 2 for most fps.
turn on triple buffer and max out the Nv settings for the game, not using any of the anisotropic (and other ) "optimizations".
max out the game settings.
vsync on, unless you have trouble getting that constant.

make sure your win power profile is at least on balanced (high performance usually just costs more power and produces more heat).
if this doesnt help, i would uninstall all NV stuff, reboot, run DDU, reboot and install 2nd latest driver..
 
Jun 3, 2018
8
0
10
0


res- 1920x1080 Windowed fullscreen
pll - was stock at 1.8 until troubleshooting and reading that increasing it can add stability to overclock. A guide said to set it and forget it at 1.88 as my card was under the 1.9 rating. Either way though my clock is stable without it and will likely be going back to 1.8
NV Settings - all changed and had no effect. Here's a graph

Win power profile set back to balanced, and I wrote in the original post that the problem occurs on both driver versions, as I started with the previous one.

One interesting thing though. In the new game graph you can see in the beginning the stable high usage is when I'm in the lobby with fps uncapped. It's very stable and only starts to spike loading and after loading into game. So it has to be with something to do with communicating to fortnite servers right?
 

fry178

Reputable
Dec 14, 2015
690
0
5,360
135
as im not playing that game, i couldnt tell.

remove some of the hardware monitors/rearrange to get different ones to show:
gpu power
gpu temp
gpu load
cpu load

start playing the game, then tab out and right click on the monitor area and clear the history.
play for about 5 min then look at the graphs..

for network tweak:
got to speedguide.net and dl tcp/ip tweaker.
run it, click on preference and change to 100ms, move slider to the speed your line has (e.g. 25Mbit),
mark the box Modify all adapters, clock on the "optimal" setting on the bottom part, hit apply and reboot pc.
 
Jun 3, 2018
8
0
10
0


Here is graph in game after network tweak and some port forwarding.

So in the lobby everything is stable, and as I start to look for a match things start to spike. The only differences I can imagine are communicating with their servers, and loading extra textures etc. So I am assuming something is interfering with that communication?

UPDATE: It also happens in single player games. Downloaded Crysis Warhead off of my steam acc and I still experience spikes. Why is it not happening in superposition benchmark but it's happening in game?
 

fry178

Reputable
Dec 14, 2015
690
0
5,360
135
at least we can now say it isnt temps/load related, tho with high fps at lower res, cpu bottlenecking becomes more obvious..

but outside spending tons of time on troubleshooting, i would uninstall all Nv related software (and drivers) and any ocing tools/software incl afterburner (keep settings), reboot to safe mode and clean with DDU.
reboot and install the latest driver in custom mode(only driver, no gfe/physx etc), and Dx as well, then see if that helped.
worst case i would wipe windows, as outside cooling/load issues, i could always fix micro stutter/spikes by cleaning Nv driver and/or restore the OS from a previous image.

http://www.guru3d.com/files-details/display-driver-uninstaller-download.html
https://www.microsoft.com/en-us/download/details.aspx?id=8109
 
Jun 3, 2018
8
0
10
0


I've done all of those things multiple times and unfortunately they haven't worked. I might just go back 2 nvidia drivers or something but I'm loosing hope on finding out what the actual issue is. I'm on fresh windows and fresh drivers now. I just don't know what is happening. My game should be smooth as silk, but something is wrong and I haven't found any leads beside stable GPU usage will afk in lobby
 
Jun 3, 2018
8
0
10
0


3D mark score and graphs.

I may be able to borrow a card from a friend to try
 
Jun 3, 2018
8
0
10
0


graph of game after changing game settings in GeForce Experience. The game felt the same, maybe a tiny bit smoother. I think the graph looks stable because the gpu has a limit it hits but you can still see all of the spikes


UPDATE: New driver came out today. Did an intsall via GeForce Experience. Custom, clean install, only graphic driver. Here's a graph of a game after install. Usage is more stable, still having stutters and input lag. This is on same settings as normal. Will grab a more details HWInfo log later tonight of usage.
 
Jun 3, 2018
8
0
10
0
I switched my GPU from it's original PCI-E Slot to the second one on my motherboard and the issue seems to be resolved. Thank you to everyone who contributed!
 

fry178

Reputable
Dec 14, 2015
690
0
5,360
135
good, kinda didnt see anything too worrisome on the graphs, except a little below what a 970 would do.

switch the settings back to max. at 1080p your cpu will be the bottleneck, switching graphics/driver/game settings towards max/ultra etc, will shift it towards the gpu.
when in afterburner, max out the temp/power settings (unlink them) and set voltage to 75% of the sliders maximum.
hit apply and save as a profile, and have AB start with windows.
once before you start gaming apply the "oc" and the chip will boost to the max (possible), without frying the card.
should give you a bit more performance...

 
Jun 3, 2018
8
0
10
0


yeah I played with overclocking my card near the beginning of noticing the problem. Will probably tweak those settings again
 

martinistripes

Prominent
Mar 10, 2017
2
0
510
0
Since 3 or 4 months ago, every new nvidia driver I install causes stutter to single figure framerate for a few seconds. I'm guessing nvidia broke something for the older cards whilst tweaking for the new cards and don't realise it yet. The whole issue caused me to do some testing & benchmarking. Of the few drivers I tested, 365.19 (the oldest one) had the best graphics performance. That's the second to last driver released before the 10 series card came out back in May 2016. I have the 970m by the way, not the 970 but perhaps the same will apply.
 

Similar threads


TRENDING THREADS