Just to throw more into the fire on the high tdp to low tdp mining....
You really need to buy a watt meter to see the FULL draw of the system and you need to understand your specific gpu very well, every gpu acts differently...
I am onboard with the low wattage, high ratio of hashpower per watt numbers, this is the standard i have been adjusting to and i have just recently taken it to even more levels of adjusting after learning some of the coding of the ccminers i use....
It all starts at your PSU, they are rated at X watts @ x% at multiple levels... I found initially that my first 4 card gtx1080ti was pulling like 91 watts more than my 2nd 4 card gtx1080ti rig, with both being identical rigs... made no sense.. settings were the same.. but what i did have a change at was i was powering the 4 cards on the first rig with a 850watt corsair gold rated PSU and the 2nd one was using my 1300watt evga plat rated PSU... i unplugged the corsair and swapped the PSU with a 1200watt evga plat rated psu i had for another rig and the power consumption while mining drop 87watts right off the start with the different PSU....
The second part of the PSU is the ratings @ different % of use... some psu's when you look into them are rated at like 80-83% @ 50% draw, but they get stupid low at like 80%.... my corsair psu rating online at 80% was something stupid like 68% which is just terrible... so running a plat+ rated PSU will increase your profits from the start...
Then you run into thermal/power throttling of GPU's, after reading in ccminer about the coding and such, i found in regards to ccminer's in general.. not sure if this is true among other miners, but in ccminer when you get the hashpower reports every so many seconds, that is actually the PEAK number being reported out of 30 (default) samples over x seconds... not the average... so it may look like your hashing higher than you truly are.... if you go into the batch and lower the sample rate and increase the reporting rate, you will find that your gpu is not hashing at the previous rate on a consistent rate, because your gpu bounces in and out of thermal throttle/power throttle multiple times a second as it trying to maintain power and temps to the levels you set them in your OC software. so by default the split seconds that your gpu is pausing thermal/power throttling, the ccminer is picking up those spikes in hashrates and reporting it as the peak number during that reporting period... a simple way to see if your throttling is using gpu-z, under monitor if your GPU is giving you ALL the hashing it can at the levels you have it set at, it will say pause under throttling, meaning the core and memory are able to perform mining at PEAK power based on the tdp and OC settings you have it set at....
So a good way to tune GPU's is to under power them so you stay out of these throttling events as much as possible... if you look in msi afterburner on the left side there is a white value and a red value above the frequency your gpu is currently reporting its running at. Those are the base and max limits your GPU can perform work at based on the settings you have it currently selected at. If your reported freq is not within a few of the max red numbers, then you are litterally pushing your GPU to work under its capabilities bouncing in and out of throttling... cryptomining blog has a writeup on this stuff, they took a system that was reporting at 80% tdp and an identical rig tuned down into the 60%ish range... even tho the one reporting 80% tdp was saying it was hashing 13% more per second, when they ran the 2 rigs for 24hrs, the difference between the 2 wallets was less than 1% in earnings... nearly within a margin of error range... yet the one at 80% used nearly 40% more power consumed at the wall...
These are mostly Nvidia pascal commands only
If your ccminer your using supports the same codes as mine, you can use the following code in a batch file to see your actual performance also....
nvidia-smi -q -d PERFORMANCE
One of the ways you can tune your GPU's is via the miner directly instead of using msi afterburner... i am wanting to switch to this method very soon as its extremely beneficial if you mine a multi-algo setup, you can tune your gpu's to run at the most efficient settings for that miner/algo combo, then you can set it into the code of the batch on each algo as a change of settings BEFORE starting mining, so it will always be running at its best settings when it switches algo's...
this line triggers a command to your gpu to set MAX memory clock on your GPU to 3505, just as if you were changing it in msi afterburner and clicked accept...
this line triggers a command to your pgu to set MAX core clock on your GPU to 1150, these numbers are just random for example
this line triggers a command to your gpu to set TDP to 100w max, so the miner will mine at 100w or less, it will vary as throttling happens....
this line triggers a sensory sampling, if the gpu hits 80c, the miner will skip the next blocks of data tell the gpu temps drop back down in which you can set a resume code to retrigger the mining again at a different temp....
