Memory leak with latest version

Martin said:
I have tested several systems and have been able to reproduce this only on one with a mobile NVIDIA GPU. So currently it seems like a memory leak in NVIDIA's support libraries (NVAPI), which can be fixed only in NVIDIA drivers.
But I'll keep testing and checking further...

But how you explain that it happens even when i enable only the cpu monitoring, i dont think it can be caused by nvidia.
 
I'm not sure about that yet. Sometimes it can appear as if memory usage would increase over a short period (which can also happen if you perform actions with sensors), but then it stabilizes.
If you have only CPU monitoring enabled, by how much does the memory usage increase and for how long have you tested it ? To get best results, you should leave HWiNFO running and don't make any changes/perform no actions.
I have also noticed that you have fan control active. Can you try to disable it and see if that makes any difference ?
 
Martin said:
I'm not sure about that yet. Sometimes it can appear as if memory usage would increase over a short period (which can also happen if you perform actions with sensors), but then it stabilizes.
If you have only CPU monitoring enabled, by how much does the memory usage increase and for how long have you tested it ? To get best results, you should leave HWiNFO running and don't make any changes/perform no actions.
I have also noticed that you have fan control active. Can you try to disable it and see if that makes any difference ?

Ok, disabled fan control but keeps increasing, do i need to disable gpu monitoring too?
 
How much does the memory usage increase and for how long have you tested it ?
Can you please attach a screenshot of the sensors window ?
 
With Fan Control disabled and only monitoring cpu the memory usage stays the same from now, even y tryed to enable only vrm monitoring of the gpus and it seems it stays the same but maybe because vrm temperature keeps always almost the same if i dont use for games
 
It happens almost the same, im testing with gpuz video benchmark to rise the vram temps and it start increasing hwinfo memory usage...
 
The value of VRM temperatures (or others) should have no effect on memory consumption.
I'm still not sure about what you have determined. So with CPU (and GPU VRM) only monitoring when idle, the memory usage is not growing? And if you raise the GPU load, then it starts growing? I don't think this can have influence. Please double check if you have tested correctly.
 
The memory usage rise with any gpu monitoring enabled, less with vrm monitoring but it is still increasing.

If i disable all the monitorings and the fan control and leaves only cpu monitoring is not increasing at all.

Almost 1 hour using the same ram.
 
OK, so finally we have some conclusion :)
Now try the following - enable all sensors except GPU and GPU VRM (make sure their headings are disabled as well) and check if memory usage is growing.
 
No, is not increasing at all and fan control opened too with the previous custom configuration of gpu, but since monitoring of gpu is not enabled it wont change any of the configurations i set.
 
OK, so you have come to the same conclusion as I did - it's the NVIDIA GPU sensor, that's causing this memory leak.
I have tested a system with a desktop NVIDIA GPU too and there's no such issue, but another one with a mobile GPU was affected (yours is mobile too).
 
Yes, but this starts happening recently, thats why i think is some kind of memory leak and it seems to be with nvidia gpus, can you solve this? or at least give me any kind of temporary fix?

Thanks
 
I'll need to perform more testing...
If it's a leak in NVIDIA drivers (NVAPI), then it might be occurring recently if you have changed NVIDIA drivers.
 
Yes, I did more testing and it reveals my assumption - the memory leak lies inside NVIDIA's support libraries (NVAPI and NVML) and any call to them (to query temperature, voltage, usage, etc) seems to be affected. Only NVIDIA can fix this.
 
Martin said:
Yes, I did more testing and it reveals my assumption - the memory leak lies inside NVIDIA's support libraries (NVAPI and NVML) and any call to them (to query temperature, voltage, usage, etc) seems to be affected. Only NVIDIA can fix this.

Thank, i cant understand why it was working right before...
 
Back
Top