Memory leak (HWiNFO using 10G memory)

[hugalafutro]

Hi,
as the title says, this is using HWiNFO 64bit 6.40-4330 portable on latest win10 x64.

screenshot:

any further logs or info I can supply let me know.

 

[Martin]

Do you have the "Log all values for Report (consumes memory)" option in sensor settings enabled perhaps?

 

[hugalafutro]

Nope, it's unchecked.

edit: I restarted hwinfo after original post, at the moment it uses almost 700MB

 

[Martin]

In which mode are you running HWiNFO, what settings?
Can you try to completely Reset Preferences if that will help?

 

[hugalafutro]

I don't think I changed many (if any) settings, I use it to feed sensor readings into Rainmeter skins.

If I reset preferences will that not mess up the HWiNFOSensorId, HWiNFOSensorInstance, HWiNFOEntryId values? I read a lot of them in the skins and it would be not fun having to change them. Otherwise I can reset the preferences and report the mem usage tomorrow.

 

[Martin]

It should not change those values, only user-defined values like custom sensor settings (colors, names, order, etc.). You can do a backup before resetting it.

 

[hugalafutro]

I reset preferences yesterday around midnight, only changed the minimize at startup options to my liking and this morning the process mem usage is close to 500Mb. I noticed there's new beta available, think it would be worth trying that ?

 

[Martin]

Yes, it's worth to try. Also make sure to upgrade drivers, especially GPU.
What are your system specs? A HWiNFO Debug File might be helpful to see the details.

 

[hugalafutro]

Hi, beta is the same, ie mem usage growth ~500M in half a day. Your question about specs made me realize I only noticed this leak around after I upgraded my pc about month ago, so maybe related to new hw?

I replaced mobo/cpu/gpu:
Msi Z490 Gaming Plus mobo
Intel i7-10700K cpu
Msi RTX 3090 gpu

Drivers are always latest stable, I'm a bit OCD about that.

hwinfo debug and report attached.

 

[Martin]

I have analyzed the data, but don't see anything suspicious except Hyper-V being active. Can you first try to disable Hyper-V if that will make a change?
If not, please try to disable sensors one by one (hit Del key over the sensor heading) and watch if disabling one of them will stop the memory leaking. I'd start with the GPU or disk sensors.

 

[Lukas0610]

Just noticed HWiNFO used quite much memory on my machine too (~750 MB after 12 hours of running), Hyper-V is disabled, disabling GPU monitoring seems to fix the memory leak.

 

[Martin]

So you both have an NVIDIA GPU, it might be a leak in NVIDIA drivers. But let's first wait for results from @hugalafutro.

 

[hugalafutro]

Definitely looks like gpu monitoring is the issue, after disabling the gpu section in sensors and restarting hwinfo the memory usage haven't moved from 70M in 8 hours.

 

[Martin]

Then it really seems to be a leak in NVIDIA drivers. You might try to disable only particular sensor values under the NVIDIA sensor to see which of them are causing this.

 

[hugalafutro]

I'm enabling the sensors bit by bit so far I've got all input voltage and temps selected and the leak is not there. This will probably take a while, but I'll keep enabling the sensors every half a day or so and report back.

 

[Martin]

That's a tough job to test all values. But I think you don't need to wait so long, it should be sufficient to watch the memory usage in Task Manager for a couple of minutes it it's constantly growing.

 

[hugalafutro]

Well this was faster than I thought, but I got the culprit 100% reproducible and recorded it. It's GPU Power. In the vid notice how the memory rises every ~2 sec when it is enabled. I then enable all the other sensors but leave GPU Power off and the memory usage doesn't move. Then after enabling GPU Power the memory rises again constantly. Hope that helps. For now I left the sensor disabled.

File is on my nextcloud as it was too large for here: https://turbomrak.o5.ddns.net/s/rQXA2PAkfyjgePw

 

[Martin]

Thanks for the testing and feedback! This is clearly an NVIDIA bug in their NVML library and should be submitted to them. I think we already submitted similar bugs, but they didn't quite listen.

 

[red-ray]

From which driver version did the leak start? I know they broke nvmllInit_v2() with the 460.79 driver so can you check if you still get the issue with the 457.51 driver?

They fixed nvmllInit_v2() with the 460.97 driver (https://nvidia.custhelp.com/app/answers/detail/a_id/5145/~/geforce-hotfix-driver-version-460.97) and am wondering if they also fixed nvmlDeviceGetPowerUsage().

If it's not fixed then if we know the driver version where the issue started I find it's easier to get NVidia to fix things

 

[red-ray]

I have logged https://partners.nvidia.com/bug/viewbug/3216492, but you probably need an NDA account to see it.