Question GPU Heat Output (not GPU Temp)

[plotinusredux]

Using an EVGA 1080 Ti SC Hybrid and EVGA Precision, but the general question should apply to most cards.

Is the heat output of the GPU solely or at least primarily determined by the voltage rather than GPU Usage or absolute clock speed? I realize usage can bump up clock speed and that in turn can bump up voltage, but is the heat output effect of the former solely a function of its effect on voltage?

Note I'm talking about heat output, not GPU temperate itself. The hybrid cooling system and custom fan curves keep the GPU between 45 C and 50 C under use regardless of voltage, clock speed, etc., but of course that heat has to go somewhere, which is into my room which becomes uncomfortably hot.

I'm just playing a 4X game (Endless Legends) so don't care much about frame rate. I haven't messed with under-volting, just setting the Power Target down to 50%, which doesn't have a significant effect on frame rate but keeps the frequency generally in the mid 1500's as opposed to the mid 1900's it wants to use otherwise (even though power = adaptive and GPU usage is around 50%). The average voltage goes down to 0.8 from 1.0. The side effect is the GPU usage goes up significantly, from around 50% to 70%.

So my basic question is, does that GPU usage itself increase temperature output if the frequency and voltage remain constant? I suppose I could test it by setting fan speed to constant and seeing if the temp went up significantly in one case rather than the other, but that would be a pain, I'd have to wait for the coolant to stabilize temps between and during tests, etc., so I'm hoping someone just knows the answer.

If voltage is the primary cause of heat output, does anyone know of a way to automatically limit voltage or power target based on running application, rather than manually changing Precision profiles depending on the game I'm playing? Or to set up a custom clock frequency vs usage so it doesn't boost to the 1900's with only 50% usage?

As a side note unrelated to my question, just something strange I've noticed, if I set both the power target to 50% and the clock offset to -200 Mhz, the system becomes unstable with the graphics driver crashing and restarting.

Thanks.

 

[RememberThe5th]

Gpu does not increase heat output TDP, since the 1000 series are locked down you will increase by small bit like 50W max if you oc by voltage and power limit.

My 1050ti msi gaming x would hit 43C 100% fan without oc , with max oc would be 45C.

 

[anotherdrew]

Generally speaking, heat in an integrated circuit comes from 2 source. First is background heat from current leakage. This background heat is proportional to voltage applied. Second, heat is a byproduct of work being done. In your GPU, that work is transistors switching from on to off and back again. The more switching (a factor of frequency and workload) the more heat that is generated. While more voltage can increase the amount of heat generated (both background heat and heat from switching), it is still the switching that generates most of the heat.

So ...

So my basic question is, does that GPU usage itself increase temperature output if the frequency and voltage remain constant?

The answer would be yes. With constant frequency and voltage, heat production is directly proportional to work being done.

**Edit ... just to clarify one thing ... The voltage applied to each transistor does not vary like the voltage applied to the entire chip does.

 

[plotinusredux]

@anotherdrew, thanks, the heat generated by the transistors switching in response to usage is what I was missing for understanding why usage itself could increase heat independently of voltage and frequency.

I also forgot that by lowering the power target to 50%, I'm presumably lower the total watts the card is allowed to consume, and the watts would be the primarily variable determining heat output (if I understand correctly).

At any rate, lowering it to 50% power target did significantly lower the heat being expelled (just as measured by my hand against the metal grill the radiator fan exhausts through and the heat level of the room), though since it did not affect frame rates at all and power is set to adaptive in NVCP, I'm not clear why the card was boosting up to the mid 1900's MHz in the first place at 50% load. In the mid 1500's with the usage up to 70% the FPS is the same as measured by FRAPS while producing much less heat, presumably because that last 400 MHz is getting into the geometric rather than linear power increase, far off-setting the extra heat from increased usage itself at the lower frequency. So why is the card boosting to the 1900's at 50% usage in the first place when it clearly doesn't need to? I suppose that's all under the hood in GPU Boost and its interaction with the adaptive power setting in NVCP rather than something I can change short of custom firmware. I don't normally run any sort of overclocking other than raising the power target to 120%, which I didn't expect to affect anything unless the GPU was under very heavy load.

Now if I could just find a way to automatically set different power targets based on what application is running rather than manually switching Precision profiles....

 

[anotherdrew]

I also forgot that by lowering the power target to 50%, I'm presumably lower the total watts the card is allowed to consume, and the watts would be the primarily variable determining heat output (if I understand correctly).

Not exactly. Lowering the power target limits the amount of work the GPU can do. Doing more work requires more power (it takes more power to make 2 transistors switch than 1). And the work is what generates the most heat.

So why is the card boosting to the 1900's at 50% usage in the first place when it clearly doesn't need to?

I can't give a definitive answer, but I suspect this is done intentionally in the software because EVGA knows that if they sell a card that is advertised to hit XXX MHz and it doesn't (or rarely does), they will have a lot of returned cards.

 

[plotinusredux]

@anotherdrew , clearly I need a basic course in thermodynamics and IC's.

I guess it may be preferring the higher frequency than needed for the load so there's not a slight lag if the load spikes, and since my temps are low it see no reason not to do so.

At any rate, without understanding all the details, the 50% power target is giving me essentially same frame rate without making me sweat (repeated swapping without touching the mouse showed exactly a 1 FPS increase at the 120% target, hitting around 75% reported power usage).

I guess I'll just switch to that profile whenever I'm playing this game or similar 4X games (I tested another with the same results). It's too bad you can't set the power target directly in NVCP so I could just set it in the profile for the game.

Thanks for your explanations--I have a somewhat better (if still far from perfect) understanding of what's going on now.