Weird behavior GPU (RTX 2060)

[martward]

Hi all, 

 

I have a bit of a conundrum here.

I have a Gigabyte RTX 2060 Windforce OC which I have been using for quite a while now, and I always noticed that it got way hotter than any GPU I have previously owned, usually going to 83C even with a custom fan curve. It bothered me but then again it did stay at about 1905MHz steadily, which is about 135MHz higher than the advertised boost clock of this model . 

So yesterday I decided to do some tests where I would a) keep everything stock b) lower the temperature limit and the power limit completely and not change anything else and c) keep the power limit and temp limit and "underclock" the GPU by 200MHz on the core and the memory. The results are rather interesting. For some reason when I lower the power limit and temp limit without changing the clocks, the GPU runs cooler at a higher clock speed than when I do not alter the power and temp limit but underclock the GPU. Also even though the temperature is lower, it doesn't actually stay below the temperature limit. It kinda feels like the temp limit isn't doing anything and it's just looking at the power limit, but then I still don't get why the temperature is higher when the clock speed (and fps) is lower in scenario c.

 

Pwr limit	TMP limit	CC offset	MM offset	Core Clock	MM Clock	Temp	FPS
100	83	0	0	1905	7000	83	58
73	65	0	0	1845	7000	72	57
100	83	-200	-200	1740	6747	80	55

 

Anyone got any ideas on what is going on here?

Thanks

[samcool55]

Voltages and amps are the key to your answer.

Higher power limit = higher voltage and more amps = more heat.

Even tho the GPU doesn't need such high voltages because it can actually run with less voltage, but it doesn't.

In this case you give it more power than it actually needs from the factory.

 

In scenario B you are pushing it past it specs efficiency wise, because you ask it to perform the same as from the factory, but you limit the voltage and amps it can use. It clocks a bit lower because it's basically on the limit power wise, not temp wise.

 

Temp limit is scenario A, limit is 83 degrees, so that's what holding it back in that scenario. Increase the limit and you will see an increase in performance (but also it temps, of course).

[martward]

1 minute ago, samcool55 said:

Voltages and amps are the key to your answer.

Higher power limit = higher voltage and more amps = more heat.

Even tho the GPU doesn't need such high voltages because it can actually run with less voltage, but it doesn't.

In this case you give it more power than it actually needs from the factory.

 

In scenario B you are pushing it past it specs efficiency wise, because you ask it to perform the same as from the factory, but you limit the voltage and amps it can use. It clocks a bit lower because it's basically on the limit power wise, not temp wise.

 

Temp limit is scenario A, limit is 83 degrees, so that's what holding it back in that scenario. Increase the limit and you will see an increase in performance (but also it temps, of course).

Hmm, I wonder then why it doesn't lower itself to stay below the 65 degrees that I set as a limit. Of course this 65 is just for testing purposes, I will definitely use scenario B as I am fine with 72 degrees and I think 10 degrees lower for 1 fps less is a pretty fair trade. 

My expectation would be that out-of-the-box it would use scenario B since it's still clocked higher than advertised, and then if you'd like you can up the temp and power limit to get 1 fps more. I was mostly surprised since I had a 1060 before this, also a Gigabyte windforce oc, which didn't go over 74 degrees under load at all. 

[igormp]

As sam has already explained everything, I'll just throw in my own experience: those rtx windforce cards with 2 fans heat up way too much. I get my 2060S to 83~86º daily, but since it stays quiet even with the fans at 100%, I don't really mind.

[martward]

4 minutes ago, igormp said:

As sam has already explained everything, I'll just throw in my own experience: those rtx windforce cards with 2 fans heat up way too much. I get my 2060S to 83~86º daily, but since it stays quiet even with the fans at 100%, I don't really mind.

Well, I'd try to lower the limits to the bare minimum. If your card is anything like mine it will drop about 10C without performance loss :P.

[igormp]

Just now, martward said:

Well, I'd try to lower the limits to the bare minimum. If your card is anything like mine it will drop about 10C without performance loss :P.

As I said before, I don't really mind. It has 3 years of warranty, so I can just ask gigabyte for a new one if it ever fries itself over 

[igormp]

38 minutes ago, martward said:

Well, I'd try to lower the limits to the bare minimum. If your card is anything like mine it will drop about 10C without performance loss :P.

Just out of curiosity, I tried to lower down the power limit from the default 175W to 150W. I noticed lower clocks right away, but the GPU still managed to reach 84º (2~3 degrees lower than the default settings), throttling itself down to around 1600MHz.

 

With the default power limit of 175W the card does 1600~1680MHz at 86~87º, throttling itself down to 150W and back to 175W regularly. Doing so actually gives me a little more Gflops of compute (~5600 Gflop/s at stock vs ~5500 Gflop/s at 150W).

 

Upping the powerlimit to its maximum value of 200W (because why not? haha) made the card reach 87ºC right off the bat, with the clocks and actual power usage going up and down like crazy (1600~1700MHz, 145~180W) with performance just slightly above the default PL.

 

Leaving the card at its minimum PL (125W) really limited the clocks (couldn't go past 1500MHz), leaving me with only 5Gflop/s of compute power, while the temperature was stuck at 78º (almost a 10C difference as you said).

 

So yeah, I guess I'll just leave it at stock since it gives me the most clocks without floating too much haha

[martward]

23 minutes ago, igormp said:

Just out of curiosity, I tried to lower down the power limit from the default 175W to 150W. I noticed lower clocks right away, but the GPU still managed to reach 84º (2~3 degrees lower than the default settings), throttling itself down to around 1600MHz.

 

With the default power limit of 175W the card does 1600~1680MHz at 86~87º, throttling itself down to 150W and back to 175W regularly. Doing so actually gives me a little more Gflops of compute (~5600 Gflop/s at stock vs ~5500 Gflop/s at 150W).

 

Upping the powerlimit to its maximum value of 200W (because why not? haha) made the card reach 87ºC right off the bat, with the clocks and actual power usage going up and down like crazy (1600~1700MHz, 145~180W) with performance just slightly above the default PL.

 

Leaving the card at its minimum PL (125W) really limited the clocks (couldn't go past 1500MHz), leaving me with only 5Gflop/s of compute power, while the temperature was stuck at 78º (almost a 10C difference as you said).

 

So yeah, I guess I'll just leave it at stock since it gives me the most clocks without floating too much haha

Yeah in your case you lose too much.