Don't look at averages, look at the end of each test when the card reaches peak temperature. Benchmarks are not good indicator because they run too short. Start a GPU hungry game and record data from it then look at last hour of play for example, or at least play 30min and look at last 5min. Takes time for a card to heat up, Timespy doesn't even do that really so averages are low. I only look at this last portion where the card actually heated up briefly finally.
- GPU Core Clock: 2032 MHz
- GPU Temperature: 68°C
- Fan Speed: 52%
- Fan Speed: 1590rpm
- Power: 236W
- Power: 110%
- Vcore: 1027mV
You want ideally to have power around 10% lower at least than max limit, so instead of 111% to have 100% and instead of 240W to have 216W. It's still hitting the power limit hard, it's not going to say 111% but 110% is damn close.
I think the cards only go truly unlimited at around 300W at these <1.1V. And if someone would manage to use the whole GPU with all it's extra features/cores for tensor, rt, what have you at once, then oh my it's essentially a 1080Ti size die and 300W is certainly not gonna cut it.
Well 200W running say 1975MHz and at 240W running 2032MHz? 57MHz extra for 40W, 2.9% performance in clock for 20% extra power. Sadly all these smaller nodes run best at around 0.75-0.9V, past that they turn power hungry and performance gains are small with power skyrocketing, so Nvidia limits it to 1.1V max but many cars run 1.0V power limited or even lower. AMD on the other hand pushes the limits and volts higher and then their cards run hot and power hungry a lot more, only so they get that tiny extra performance gain to keep up with NV's performance. On AMD many people actually undervolt to reduce power consumption instead of push more volts and clocks. They reduce volts and temperatures so the boost goes automatically to higher clocks.
I don't have that many games installed but I can try Doom Demo or Q2VKPT again and record some stats. The game I play often runs 40% GPU when online and even much slower GPU doesn't get fully loaded. No Doom Demo won't cut it, even at max settings this card is a beast and kept hitting 200fps limiter in some scenes. from what I remember. Q2VKPT will load it full, recording that now. A 6min run and gun without any level loading, attached. Runs around 40fps or so, almost unplayable at this near fullscreen.
Average:
- GPU Core Clock: 2000 MHz
- GPU Temperature: 63°C
- Fan Speed: 46%
- Fan Speed: 1191rpm
- Power: 198W
- Power: 113.6%
- Vcore: 983mV
Doom is not cutting it, 82% load only at max settings, 200fps limiter forced there.
Gonna try heaven. Why didn't I think of that before. LOL Windows started installing some update at the end of Heaven run, stupid M$. I don't know how many times I've set the updates and they always change something or change their update services completely, start ignoring older settings, etc.
Average:
- GPU Core Clock: 1984 MHz
- GPU Temperature: 63°C
- Fan Speed: 48%
- Fan Speed: 1235rpm
- Power: 197W
- Power: 112.7%
- Vcore: 973mV
I always run around 3min to warm up and then start 5min recording in HWiNFO.
Try that and see what it runs for you as Timespy runs so short the card doesn't heat up only at the end of it at least with your custom fan profile. Run Heaven 1440p max settings, that seems the easiest, let run it's scenes on automatic cinematic mode.
I think my fans spin up on PC start and this is normal behavior for all GPUs I have seen, then they stop, not sure if before or after Windows and driver load since that happens fast nowadays. I can check by booting to UEFI.