Different mem clocks

[Darkjacky]

Hello,

 

While trying to run folding@home I noticed that my overclocks are different. The weird thing is Folding doesn't seem to be the issue. I will add 2 images showing it.

The card runs at 7500MHz default but somehow in games it adds 250MHz extra on the memory clock. This causes instability in the worst case as I have seen.

Is there any way to fix this?

 

Note that the core frequency goes a lot higher on a more stressful game.

Thanks you

[VeganJoy]

actual boost clocks are handled by the card and are kind of an offset to whatever you set in afterburner. run some torture tests/benchmarks and watch gpu-z/afterburner to see how high it actually goes. 3dmark summarizes these clock speeds at the end of benchmark runs. for example, with my 980ti it'll say it's running at ~1400mhz core in afterburner, but after running time spy the summary tells me it peaked at 1530mhz and averaged 1518mhz. or running furmark, it has an overlay that tells you that sorta thing. in that it becomes more clear that the card's actual boost clocks are temperature dependent as it'll end up at around 1480mhz after a few mins of furmark, which is still higher than the default shown in afterburner.

[-rascal-]

15 minutes ago, VeganJoy said:

actual boost clocks are handled by the card and are kind of an offset to whatever you set in afterburner. run some torture tests/benchmarks and watch gpu-z/afterburner to see how high it actually goes. 3dmark summarizes these clock speeds at the end of benchmark runs. for example, with my 980ti it'll say it's running at ~1400mhz core in afterburner, but after running time spy the summary tells me it peaked at 1530mhz and averaged 1518mhz. or running furmark, it has an overlay that tells you that sorta thing. in that it becomes more clear that the card's actual boost clocks are temperature dependent as it'll end up at around 1480mhz after a few mins of furmark, which is still higher than the default shown in afterburner.

 

The question is about MEMORY frequency, not GPU Core frequency.

The GPU Core frequency will/should boost based on temperature / power headroom.

 

The memory frequency should not 'boost' in anyway.

At least, not from what I know.

 

27 minutes ago, Darkjacky said:

Hello,

 

While trying to run folding@home I noticed that my overclocks are different. The weird thing is Folding doesn't seem to be the issue. I will add 2 images showing it.

The card runs at 7500MHz default but somehow in games it adds 250MHz extra on the memory clock. This causes instability in the worst case as I have seen.

Is there any way to fix this?

 

Note that the core frequency goes a lot higher on a more stressful game.

Thanks you

 

Which RTX 2080 Super do you have (make and model) ?

 

[VeganJoy]

11 minutes ago, -rascal- said:

 

The question is about MEMORY frequency, not GPU Core frequency.

The GPU Core frequency will/should boost based on temperature / power headroom.

 

The memory frequency should not 'boost' in anyway.

At least, not from what I know.

i figured they both did but i'll have to take another look 

[Darkjacky]

1 hour ago, -rascal- said:

 

Which RTX 2080 Super do you have (make and model) ?

 

This one https://www.techpowerup.com/gpu-specs/msi-rtx-2080-super-ventus-oc.b7201 just a copy pasta from GPU-Z Lookup button.

[-rascal-]

19 hours ago, Darkjacky said:

This one https://www.techpowerup.com/gpu-specs/msi-rtx-2080-super-ventus-oc.b7201 just a copy pasta from GPU-Z Lookup button.

 

You manually overclocked the GPU Memory by +600.

 

 

That would give you the ~8350 MHz MSi Afterburner is showing you.

Stock is ~7752 MHz.

 

It looks like, when you are Folding, since the GPU is intensively crunching calculations, the GPU Memory is being clocked lower than what it is set.

There are SOME applications that WILL do this -- and it seems whatever you are running to Folding, is triggering that.

This is something built into the nVidia drivers (also present on AMD).

 

Drop the Memory overclock down back to stock (+0 in MSi Afterburner), or reduce to a more stable level.