Question RTX 3070 Utilization low in multiple games

Feb 10, 2022
9
0
10
Hey all! I have a NVIDIA GeForce RTX 3070 (EVGA 8GB) and the GPU usage is always low in games even when uncapped, and I get low frames as a result. For example, in FFXIV on all max, I get around 70 FPS but it only uses about 28% of GPU despite no frame limit being set. I have the same issue with PSO2 - Uncapped FPS but it only uses like 43% of the GPU max and it goes to 80 - 90 FPS. I get similar results in other games. I've tried updating my drivers, enabling the 4G address thing, the rebar thing, disabling the high precision event timer, reinstalling Windows, nothing changes it. I also set it to Max Performance in the NVidia control panel, nothing.

System Specs:
CPU: AMD Ryzen 9 3900X 3.8 GHz 12-Core Processor
GPU: EVGA GeForce RTX 3070 8 GB XC3 ULTRA GAMING Video Card
Motherboard: ASRock B550 Steel Legend ATX AM4 Motherboard
RAM: Patriot Viper Steel 32 GB (2 x 16 GB) DDR4-3200 CL16 Memory (x2)
OS Drive: Samsung 960 EVO 250 GB M.2-2280 NVME Solid State Drive
Power Supply: Corsair RMx (2018) 650 W 80+ Gold Certified Fully Modular ATX Power Supply
Monitor: Pixio PX248 Prime 24.0" 1920x1080 144 Hz Monitor
OS: Windows 10/11 (it happens in both)

UserBenchmark: https://www.userbenchmark.com/UserRun/50379771
 
Last edited:
GPU will work as hard as CPU will allow it to do so
if you see quite low GPU usage while also having low FPS, then it means game isnt using your CPU properly (too old game engine without AVX support for example)
the only thing that helps here (as old games doesnt get CPU codepath updates) is CPU overclocking, sometimes just helps disabling SMT (simultaneous multithreading aka hyperthreading)
on the other hand you have so much spare GPU horsepower that you can run max details alongside with antialiasing and better resolution (or DSR), that wont give you better FPS, but better image quality while having same FPS
there are also 3rd party tools which can enhance old game graphics with extra postprocessing shaders like ReShade tho some anticheat software might not allow you to run it, better to use on single player games

for testing your CPU, you can run some benchmarks and compare online if it runs as it should
also make sure that dual channel with atleast 3200MHz is set on your RAM modules, as it may hinder CPU performance aswell if it runs below 3200MHz or in single channel
 
Feb 10, 2022
9
0
10
GPU will work as hard as CPU will allow it to do so
if you see quite low GPU usage while also having low FPS, then it means game isnt using your CPU properly (too old game engine without AVX support for example)
the only thing that helps here (as old games doesnt get CPU codepath updates) is CPU overclocking, sometimes just helps disabling SMT (simultaneous multithreading aka hyperthreading)
on the other hand you have so much spare GPU horsepower that you can run max details alongside with antialiasing and better resolution (or DSR), that wont give you better FPS, but better image quality while having same FPS
there are also 3rd party tools which can enhance old game graphics with extra postprocessing shaders like ReShade tho some anticheat software might not allow you to run it, better to use on single player games

for testing your CPU, you can run some benchmarks and compare online if it runs as it should
also make sure that dual channel with atleast 3200MHz is set on your RAM modules, as it may hinder CPU performance aswell if it runs below 3200MHz or in single channel

I'm not sure FFXIV uses AVX or that stuff, but I'm 150% sure PSO2: NGS does. You can't play the game unless you have an AVX2 Enabled CPU (it literally will not run). The image quality is fine, I just really want at least a solid 90 FPS on these games =/

I added a userbenchmark to the first post as well, forgot that. https://www.userbenchmark.com/UserRun/50379771

EDIT: I just checked the video userbenchmark recommended me that's similar to my build - They got an average of 231 FPS in fortnite but I just loaded up mine and I get an average of 70 - 110 with a lot of frame drops when set to unlimited. I'm definitely not getting the same performance they are and their CPU is much weaker than mine.

Thank you!
 
Last edited:

Zerk2012

Titan
Ambassador
I'm not sure FFXIV uses AVX or that stuff, but I'm 150% sure PSO2: NGS does. You can't play the game unless you have an AVX2 Enabled CPU (it literally will not run). The image quality is fine, I just really want at least a solid 90 FPS on these games =/

I added a userbenchmark to the first post as well, forgot that. https://www.userbenchmark.com/UserRun/50379771

EDIT: I just checked the video userbenchmark recommended me that's similar to my build - They got an average of 231 FPS in fortnite but I just loaded up mine and I get an average of 70 - 110 with a lot of frame drops when set to unlimited. I'm definitely not getting the same performance they are and their CPU is much weaker than mine.

Thank you!
First place I would start is free up more space on the OS drive it says just 11GB left.
 
your userbenchmark result aint great tbh
here for comparison with mine r7 3800x https://www.userbenchmark.com/UserRun/47132492
ignore that 100% ranking..thats from memory score which got 90points and has barely anything to do with gaming...thats just good for database operation...
ram latency is quite high and single core cpu clock isnt boosting to 4.6GHz, 1core at 4.5GHz would be like 150 and ure at 130
can you run this app and post picture from it?
 
Feb 10, 2022
9
0
10
Was windows a full Fresh install or a reset/repair install?
It can make a huge difference.

Fresh install

First place I would start is free up more space on the OS drive it says just 11GB left.

I'll do that when I get home, just to be safe.

your userbenchmark result aint great tbh
here for comparison with mine r7 3800x https://www.userbenchmark.com/UserRun/47132492
ignore that 100% ranking..thats from memory score which got 90points and has barely anything to do with gaming...thats just good for database operation...
ram latency is quite high and single core cpu clock isnt boosting to 4.6GHz, 1core at 4.5GHz would be like 150 and ure at 130
can you run this app and post picture from it?

That's really odd - I'll run that and post the results when I get home tonight. EDIT: Actually was able to get it done before work, here ya go: View: https://i.imgur.com/EvITlOr.png


I don't think this would effect it, but I noticed programs like Chrome lagging hard when I turned off the high precision event timer , so I re-enabled that before I took these timings. I turned it off and on with the window open and the timings didn't change, so I don't think it affected it. I'll re-run the userbenchmark when I get home to make sure it wasn't affected by that either, but it didn't lag during the test.

Thanks for the help, this has been driving me crazy for a while!
 
Last edited:
Fresh install



I'll do that when I get home, just to be safe.



That's really odd - I'll run that and post the results when I get home tonight. EDIT: Actually was able to get it done before work, here ya go: View: https://i.imgur.com/EvITlOr.png


I don't think this would effect it, but I noticed programs like Chrome lagging hard when I turned off the high precision event timer , so I re-enabled that before I took these timings. I turned it off and on with the window open and the timings didn't change, so I don't think it affected it. I'll re-run the userbenchmark when I get home to make sure it wasn't affected by that either, but it didn't lag during the test.

Thanks for the help, this has been driving me crazy for a while!
MCLK, FCLK and UCLK doesnt have same frequency
go to bios
in OC Tweakerpage, set infinity clock and dividers to 1600MHz, then go to advanced page, AMD Overclocking, there enable DRAM latency enhance (if present) and openup SoC/Uncore OC mode and change it there to enabled
once done, zentiming should report same clock for FCLK, MCLK, UCLK and ram latency should considerably drop

PC laggin without HPET...hmm you could disable power saving CPU C-states to solve that...or reconfigure your windows timers or just keep HPET enabled :)
 
  • Like
Reactions: Unolocogringo
Run HWiNFO64 in the background (sensors only, logging on) while you experience bad performance, for at least 30 mins. Go through the log file afterwards to see if anything jumps out.

Userbenchmark is telling me that your OS drive is full and that your CPU isn't boosting to its potential (possible thermal issues).
 
that odd cpu boosting behavior can be from windows power plans
if you switch to different power plan, youll get different performance level, even if you compare high performance with ryzen high performance power plans
they all get different max boost clock

but i wouldnt thrust userbenchmark on CPU score that much, that test literaly takes 1 second to test your CPU
use cinebench to test your CPU, check with hwinfo your CPU boosting behaviour - frequency boost global limit (feel free to run CB on 6 cores to see gaming performance level), you can also switch windows power plans during CB test to see which gives you better performance (but usualy its teh windows high performance power plan), balanced power plan usualy gives -300mhz on max boost clock
 
Last edited:
Feb 10, 2022
9
0
10
MCLK, FCLK and UCLK doesnt have same frequency
go to bios
in OC Tweakerpage, set infinity clock and dividers to 1600MHz, then go to advanced page, AMD Overclocking, there enable DRAM latency enhance (if present) and openup SoC/Uncore OC mode and change it there to enabled
once done, zentiming should report same clock for FCLK, MCLK, UCLK and ram latency should considerably drop

PC laggin without HPET...hmm you could disable power saving CPU C-states to solve that...or reconfigure your windows timers or just keep HPET enabled :)

Done, the numbers are all the same and it could be placebo but my OS does seem a lot more responsive and faster!

View: https://i.imgur.com/wELzDID.png


I saw the memory timings too.
Something off there.
Do as suggested above.
Hopefully they will run and act nice together.
When mixing memory kits ,sometimes weird things happen.
Your memory shows as 2x 32GB kits for 64GB.

Please correct if this is wrong.

I have 4 x 16GB sticks (for a total of 64GB) but they're all the same brand.

Run HWiNFO64 in the background (sensors only, logging on) while you experience bad performance, for at least 30 mins. Go through the log file afterwards to see if anything jumps out.

Userbenchmark is telling me that your OS drive is full and that your CPU isn't boosting to its potential (possible thermal issues).

Just cleared out 70GB on my OS drive, and I'll check HWiNFO64.

that odd cpu boosting behavior can be from windows power plans
if you switch to different power plan, youll get different performance level, even if you compare high performance with ryzen high performance power plans
they all get different max boost clock

but i wouldnt thrust userbenchmark on CPU score that much, that test literaly takes 1 second to test your CPU
use cinebench to test your CPU, check with hwinfo your CPU boosting behaviour - frequency boost global limit (feel free to run CB on 6 cores to see gaming performance level), you can also switch windows power plans during CB test to see which gives you better performance (but usualy its teh windows high performance power plan), balanced power plan usualy gives -300mhz on max boost clock

I used to have a Ryzen High Performance power plan on my previous version of Windows but I don't seem to anymore even after installing the Ryzen drivers. And will do and post results.

Thank you all!

EDIT:
HWiNFO64 stats while CB was running it's multi-core test (I couldn't find a way to specify the cores):

Beqn32D.png


vAG77ih.png


HwCYxfg.png
 
Last edited:
system being more responsive with lower timings is normal
all core test 4GHz is okay as it hits PPT and EDC power limit walls (power tracking and current design), no thermal throttling
for the 6core test, you open task manager and change core affinity on cinebench, or try single core test

another advice wold be changing FSB in bios from auto to 100.0 MHz or you can go up to 101.0 MHz without affecting system stability, that will sligtly improve system buses bandwith and latency
 
Feb 10, 2022
9
0
10
system being more responsive with lower timings is normal
all core test 4GHz is okay as it hits PPT and EDC power limit walls (power tracking and current design), no thermal throttling
for the 6core test, you open task manager and change core affinity on cinebench, or try single core test

another advice wold be changing FSB in bios from auto to 100.0 MHz or you can go up to 101.0 MHz without affecting system stability, that will sligtly improve system buses bandwith and latency

kk, I'll try that

just remembered something, can you run this command in powershell with admin rights and post what it prints there?
Code:
Get-CimInstance –ClassName Win32_DeviceGuard –Namespace root\Microsoft\Windows\DeviceGuard

Just realized I never did this, sorry.

PS C:\Windows\system32> Get-CimInstance –ClassName Win32_DeviceGuard –Namespace root\Microsoft\Windows\DeviceGuard


AvailableSecurityProperties : {1, 2, 4, 5...}
CodeIntegrityPolicyEnforcementStatus : 0
InstanceIdentifier : 4ff40742-2649-41b8-bdd1-e80fad1cce80
RequiredSecurityProperties : {0}
SecurityServicesConfigured : {0}
SecurityServicesRunning : {0}
UsermodeCodeIntegrityPolicyEnforcementStatus : 0
Version : 1.0
VirtualizationBasedSecurityStatus : 2
VirtualMachineIsolation : False
VirtualMachineIsolationProperties : {0}
PSComputerName :
 
Just adding that I read somewhere that some monitoring programs do not record your GPU usage properly. You should consider using MSI Afterburner (or its cousin EVGA Precision) to get an accurate gauge of your GPU usage. You can also try to shift more load onto your GPU by using DSR and playing your games at 4K resolution (downscaled to your 1080p monitor).
 
Feb 10, 2022
9
0
10
Both games now use about 10% more of the GPU, so I think the bottleneck might be whatever is causing my CPU to not run at it's intended speed.

another advice wold be changing FSB in bios from auto to 100.0 MHz or you can go up to 101.0 MHz without affecting system stability, that will sligtly improve system buses bandwith and latency

I couldn't find anything labeled FSB in my bios. Here's some stuff that I thought might help (sorry for the picture quality).

Please note in this picture that I changed the CPU Frequency and Voltage (VID) Change from Auto to Manual to show what the options were, as I think it might have been related to the FSB setting you mentioned. It's current setting (CPU Frequency and Voltage (VID) Change) is "Auto".

NWvP4sx.jpg



FYMp1de.jpg


disable virtualisation in bios settings, its called SVM, should be around advanced CPU configuration settings
HVCI has some performance drops even with hardware baked support

Done (y)
 
I couldn't find anything labeled FSB in my bios. Here's some stuff that I thought might help (sorry for the picture quality).

Please note in this picture that I changed the CPU Frequency and Voltage (VID) Change from Auto to Manual to show what the options were, as I think it might have been related to the FSB setting you mentioned. It's current setting (CPU Frequency and Voltage (VID) Change) is "Auto".

NWvP4sx.jpg
its that overclock mode (bus speed)
 
Feb 10, 2022
9
0
10
I changed the Overclock mode to 101.000 and then the tRFC value to 350 and my system wouldn't post. It threw error 22 which is "RAM could not initialize". I reset the CMOS and I'll try the overclock mode later today when I get home.
 
We are working with 4 sticks and 2 different kits.
You may never get low cas and tight timings in the specs.
Mixing kits is a crap shot.
Some times it works.
Some times it works with terrible timings and lower speeds.
Some times it sort of works but you get crappy performance and random glitches and errors. That you can never diagnose.
And some times it fails to even boot.
Looks like everyone in the thread missed that important part.
And Op fails to actually state it is his description, and futher ignores when asked directly.
 
Feb 10, 2022
9
0
10
We are working with 4 sticks and 2 different kits.
You may never get low cas and tight timings in the specs.
Mixing kits is a crap shot.
Some times it works.
Some times it works with terrible timings and lower speeds.
Some times it sort of works but you get crappy performance and random glitches and errors. That you can never diagnose.
And some times it fails to even boot.
Looks like everyone in the thread missed that important part.
And Op fails to actually state it is his description, and futher ignores when asked directly.

From the OP:

RAM: Patriot Viper Steel 32 GB (2 x 16 GB) DDR4-3200 CL16 Memory (x2)

It's the same kit, twice. And you asked me and I responded to it, saying so:

lRfS4ge.png
 
Do a simple test.
Take them out and match them in their original pairs.
Do a bios reset.
Try an original matched pair and compare.

You can look on the serial number to determine pairs.

Look at performance and responsiveness, between 1 pair and 2 pairs.
This comes up a LOT so a member made a sticky to explain it all.
 
Last edited:
Feb 10, 2022
9
0
10
Do a simple test.
Take them out and match them in their original pairs.
Do a bios reset.
Try an original matched pair and compare.

You can look on the serial number to determine pairs.

Look at performance and responsiveness, between 1 pair and 2 pairs.
This comes up a LOT so a member made a sticky to explain it all.

Interesting... Could this be the reason for my bad GPU performance? I'll take care of it either way (thank you!), just curious.