Hi! I know this has been posted a couple of months ago. Maybe you already solved it, but I'll share my experience anyway. It may help you and in case you have better suggestions you can help me. I've found out with my XC3 Ultra that I can get up to 124Mh/s in NiceHash using NBminer and DaggerHashimoto with the following settings in MSI Afterburner:

• Power limit: 104%
• Memory clock: +1300 (though I've found this variable not affecting much)
• Core clock: -502
• Fan speed: 100% 

Now, there's a couple of things to keep in mind. Fans running at 100% 24/7 is not good for longevity. In Tom's Hardware article I read that keeping it at 50%, or at a maximum of 80% is the best for long-term use. The other very important factor to keep in mind is memory temps. To check this temperature you'll need to install HWiNFO and check the temperature of "GPU Memory Junction Temperature". Apparently, GPUs using GDDR6X memory (3080 and 3090) get REALLY hot quickly, especially when mining. These settings that I just shared with you will make the memory temp go up to 110ºC. At that temperature, the card will limit itself because reaching 120ºC would permanently damage the memory modules. The best I could do to keep it at down as possible was open the case and a big fan blowing directly at it and got an average of 102ºC. This is not ideal since it will accumulate dust very quickly, GPU fans still run at 100% and 102ºC is still too hot.
 
Now, I'm not using those settings now. Instead, I'm using these:

• Power limit: 80%
• Memory clock: 0 (though I've found this variable not affecting much)
• Core clock: -502
• Fan speed: 100%

In this configuration, I get around 105Mh/s with 92ºC. I'm way more comfortable with these temperatures and I get paid around 0.00025000 BTC per day. This can still be improved since 92ºC is still not ideal (but WAY better than any triple-digit temperature) and fans are still working at 100% 24/7. I just bought a couple of heatsinks to put over the metal backplate with some good thermal pads to see how it works. This would be done over the plate and without disassembly since that would void your warranty. A guy in the NiceHash's Discord did something similar with 4 small heatsinks for M.2 NVMe SSDs over the memory modules on the back of his FTW 3090 and saw memory temperature decrease by 10ºC.
 
Of course, having a case with good airflow really helps with these. Overall you'll find that memory temperature is your enemy and your main trade-off. Another suggestion that I saw for not ruining the GPU fans is to use come cheap external ones that, if they fail, are nowhere near as costly as replacing your 3090 fans. I would recommend playing it safe and not pushing your card to the limits at least until it pays itself. Once you reach that point, you can push it as far as you'd like. Also, try every resource and alternative you have to cool it off. The more you cool, the safer for your card, the more performance you'll get and the more profit you'll earn.

You should undervolt directly (by adjusting the curve in PX1/afterburner) to reduce power, rather than doing it indirectly fiddling with the limit. If the 3090 is anything like my 3080, you'll get better and more consistent results by dialing it down as far as it goes (~0.731V).

Just to further clarify the best setup is mem+1200 core+0 power 85% (about 280-290Watts) internal card fan at 80% with radiatior fans at 90% (I recommend getting 4x HD120 rgb pro PWM corsair fans, theyre 2350rpm and not too loud if bought nee) hybrid kit (lather the pump and copper plate that goes around it and where the two connect in thermal paste so it gets optimum contact as that is limited by the cooler design as it is), extra thermal paste on the mmeory modules under the backplate, extra fan set on the back of the card, if my tiny XC3 gets cooled that much better those of you with a larger one like the ftw3 or an asus strix or a kingpin should be in even better shape than I am.