No GPU has a boost clock that matches GPU-Z unless it is overheating or broken. All are much higher when everything is working correctly. Pascal was the same way. My GPU-Z shows 1770 and the actual boost is 2070. That is stock, no overclock. If you look at the second row from the bottom, if the far right cell shows a core boost clock greater that the cell in the bottom row, then the GPU is overclocked, but you cannot tell what the actual boost clock is by looking at that number.
Here is a render test example. You have to look at the sensors tab to see the actual boost clock.
View attachment 169106
The TGP (aka "TDP") can also be misleading. What actually matters is how much the power slider moves and if the voltage scales correctly in unison. If the voltage doesn't increase, the higher available TGP is absolutely worthless. If Alienware is advertising 180W and you can only add +15% that won't be any more relevant or beneficial than 150W with a +25-30% slider unless it holds higher voltage under load. This is controlled by the firmware.
With Pascal and Turing, it's a smoke and mirrors game, and until you actually put it to the test you can have no real idea what the results are going to look like. Desktop Pascal and Turing GPUs are also all over the board when it comes to real world results. Throw in the wild card for temperatures and room temperature thermal throttling algorithms and it gets even more silly.
The firmware is also inconsistent. I have been testing all of the 2080 Ti vBIOSes I can get my hands on and having a higher TGP doesn't necessarily provide any benefit. The Galax HOF 450W vBIOS produces poor results compared to my stock XC2 Ultra, FTW3 and Aorus Extreme firmware, all of which have much lower TGP. What matters more is how well it will hold the maximum voltage. If it does not hold max voltage the boost clocks will be erratic and lower even if the TGP rating is much higher. As I said, it's all smoke and mirrors nonense and the only thing that actually matters is the test results.
In the screen shot below, same load test with 122% power limit with zero improvement in the benchmark result. In both cases, voltage (not TGP) was the limiting factor.
View attachment 169107
Now, in this third example, exactly the same load test, except for I manually selected the maximum voltage. Notice boost clock goes from 2070 to 2220 and the voltage to 1.093V and the VRel (blue bar) performance limit is gone. The watts increased because more voltage was available, but it still didn't exhaust the power limit. There are no GPU-Z PerfCap reasons because the higher voltage limit gave the GPU what it needed and the boost clock was able to go much higher. This is why it is important to have a system not crippled by cancerous firmware. The power limit doesn't matter until it becomes a limiting factor. Voltage control is far more important. If you don't take control of the voltage, you won't be able to take advantage of, or see any actual benefit from, a higher power limit.
View attachment 169108
In both of these full benchmark runs (below) I am using the Gigabyte Aorus Extreme vBIOS with a 380W power limit and the score is MUCH higher than I can achieve using the Galax HOF OC Labs 450W vBIOS simply because the Galax HOF OC Labs vBIOS won't let me lock the voltage at the max 1.093V like the Aorus Extreme, XC2 and FTW3 vBIOS firmware does. The Galax 450W TGP firmware is utterly worthless and there is no point in even having it because of that. The same problem (sucky voltage control) exists when using the MSI Lightning Z vBIOS.
TL;DR - don't get too caught up in the TDP. It matters only when it is too low, but it can be a very misleading specification. This situation is exactly the same for CPUs. You always want the power limit maxed out so you don't have to think about it being a limiting factor, then you can focus your attention on voltage control... that's where you should burn the most calories.
![[IMG]](images/storyImages/image_id_2129127.jpeg)
View attachment 169109
Edit: please ignore the 238.1W reading in GPU-Z. This value is not accurate because my GPU has been shunt-modded to prevent the vBIOS and driver from recognizing how many watts are being pulled by the GPU. It's actually around 500W (860W including the CPU) judging from my Kill-A-Watt meter.
Problems? See this thread at archive.org.
Last edited: Feb 1, 2019
