HwInfo causing FPS drop anomalies on Nvidia GPUs with VSync

Well I did some gaming test (avatar FOP, which is know to be quite well optimized, at least for frametime) and the result is that HWinfo64 GPU monitoring is somehow creating frametime spikes when Vsync is on.

See below picture, on the top, with HWinfo64 GPU monitoring (1000ms pooling), it clearly has spikes at a regular pace (1000ms...). The bottom capture was with HWinfo64 GPU monitoring disabled. No spikes.

Frametime.jpg

Sadly I can definitely perceive those spikes, so I would tend to turn off GPU monitoring... But damn that su*ks, as I completely lose any sense of GPU temp and usage, which is a big deal!

I guess ill try to see if I can get RTSS to OSD those metrics at least while gaming ( for the record, RTSS does not create those spikes...).

Note that in both cases NVML was disabled, or else the spikes are much, much bigger.

Edit, I will try reinstalling an older driver, most likely 537.58 to see if this was introduced in a recent driver like the NVML issue.
 
Last edited:
I can confirm, that there seems to be two different issues, which is also in confirmation with reports on both Nvidia forum and also in this thread:


From what I see from this:

1. NVML usage, for example through HwInfo makes it worse => 100% reproducible and larger fps drops
2. Monitoring software also triggers fps drops, but not as bad as NVML usage
3. meaning also MSI AB and also HwInfo GPU monitoring (even with NVML off) causes FPS drops

Maybe internally the Nvidia driver uses NVML if you talk to driver for monitoring, or the issue is deeper than NVML, and not directly in the NVML library, I am not an expert on this of course. There is something broken though since some months in NVidia driver, first the NVML lib, and then also monitoring tools.
 
Last edited:
I have passed this problem to NVIDIA, it should have reached the folks responsible for NVML.
 
It seems to me that the scope of the issue is wider than just NVML. In my case, HWiNFO GPU monitoring is causing frametime spikes even if I disable monitoring of all GPU sensors and NVML (but keep GPU monitoring enabled). By comparison, MSI Afterburner causes frametime spikes only if GPU Power monitoring is enabled. And the root cause of this behavior is NVAPI power call by itself (I verified it myself by calling NvAPI_GPU_ClientPowerTopologyGetStatus). So, I'm making a conclusion that NVAPI calls for GPU Power metrics are flawed. But other metrics should work without noticeable issues. Which is not the case with HWiNFO for some reason even if NVML is disabled.

I also did these tests in Windows 10 with 537.58 drivers and disabled HAGS and the behavior is the same.

Here's a video I made to demonstrate the issue with HWiNFO, comparing it to Afterburner and also showing how NVAPI power call is causing frametime spikes:
 
@Martin maybe you can write the above information @valera5505 wrote to Nvidia for debugging. It was my theory already, that the root cause of this is more deeper, and it might not be inside NVML but on a deeper level just manifesting using NVML. But it seems all monitoring tools causing it too.
 
Nvidia posted this as an answer to the issue:

"We have looked into this and it appears that the application is using non-public APIs that were not meant for external use, especially while gaming. If you are experiencing stutter or FPS fluctuations, we recommend turning off the background scanning while gaming."

Typical answer to avoid looking into it and claim it is the users fault not ours. Seriously? Oh the issue is, youre using non public APIs which are not meant to be used for normal people and especially not while gaming? Seriously? Oh and they also advice to not use any monitoring tools while gaming.... right. Dont they know most people having MSI AB and mostly HwInfo running all the rime, also magazines and sites use them to do benchmarks and testing? So it is not an issue if the benchmark results are flawed?

That is totally nonsense. There is obviously an issue in the drivers.

Hope they will still try to fix it.
 
Last edited:
It is being worked on, its just low priority because as they said, the api's aren't meant for tools such as gpu-z/hwinfo in the consumer field, but for server monitoring, which isn't suffering at all.
 

GeForce Hotfix display driver version 551.46 is based on our latest Game Ready Driver 551.23.

This hotfix addresses the following issues:
Some users may experience intermittent micro-stuttering in games when vertical sync is enabled [4445940]
Potential stutter may be observed when scrolling in web browsers on certain system configurations [4362307]

First test moments ago with MSI AB (monitoring on, RTSS) + HWInfo (7.72-5355, NVML off) + VSync (1080@120) in Rimworld:

1707389969038.png

Tabbing out in out in, stable 120FPS, had dramatic drops before. Stable 120FPS, where I had drops before between 110-117FPS

Ufotest:

1707390021344.png

Looking good so far.
 
Last edited:
@valera5505 what is there to see on that clip? I dont see anything unusual. 240hz is really high refresh rate your pc seems to have issues with it, stuff happening in the background, moving mouse and so on. make sure youre using 500hz polling rate for your mouse for example not more. test in a fullscreen game, not in chrome browser. your issue is not the one reported here.

the problem is fixed, as you see everywhere reported:


 
@valera5505 what is there to see on that clip?/URL]
Regardless of wether I am moving my mouse or not, GPU monitoring consistently causes frametime spikes. In a vsynced game this results in not being able to sync to the monitor. Same thing with testufo and vsynctester as I am showing in the video.
your pc seems to have issues with it
At 0:24 I disable GPU monitoring and suddenly my pc stops having issues with it. And then again starts having at 0:30 when I enable it back.
 
@valera5505 try lowering fps/hz 120 and see how it goes. you also use high polling rate of 500ms instead of 2000ms default in hwinfo.

I did a test right now too with HwInfo running with sensors and https://www.vsynctester.com/ with 120fps and 2000ms polling:

No HwInfo running:

1707414023044.png

HwInfo running:

1707414058544.png
Every 2000ms there is a spike caused by HwInfo it seems. Not sure if this can be made better, MSI AB doesnt cause a spike on sensor date reading it seems.

So youre right on this issue, but the main problem with VSync and fps going way down like 113-117 instead of 120 was fixed with this driver.
 
Not sure if I should open separate topic or just comment here. HWINFO is causing major stuttering and fps loss on nvidia gtx 1070 in at least League of Legends while hwinfo sensors window and graphs are visible. It's much less if they are hidden behind another window or minimized. Other notable applications running are OBS streaming the game and Chrome + Brave + Edge browsers but I don't think it matters much. Also tested with HAGS and without etc... I have 2 monitors one 60hz and other 144hz running at 120hz where I game. HWINFO is running and being shown on 60Hz screen.

Visible window and graphs look like this
0OQnjsn.png
 
Last edited:
Not sure if I should open separate topic or just comment here. HWINFO is causing major stuttering and fps loss on nvidia gtx 1070 in at least League of Legends while hwinfo sensors window and graphs are visible. It's much less if they are hidden behind another window or minimized. Other notable applications running are OBS streaming the game and Chrome + Brave + Edge browsers but I don't think it matters much. Also tested with HAGS and without etc... I have 2 monitors one 60hz and other 144hz running at 120hz where I game. HWINFO is running and being shown on 60Hz screen.

Visible window and graphs look like this
0OQnjsn.png
What system, mainboard is that? Could be due to some sensors monitored like EC if there is such.
 
Back
Top