Hmm please do double check those numbers and report back please, I am basically with the exact same settings but at 44-45 in t-rex too. Which isn’t terrible just trying to understand whats going on 
I just checked and those numbers were accurate, but I forgot that i did just change the overclocks here a little bit ago because like i said, I just got it a few days ago and still trying different settings and such. So the new oc settins i currently have set, that reflects the screenshot of the miner console i have attached are:
core clock locked: 1450
mem clock: +2000
everything else the same as stated before, windows driver, trex version, etc…
The results seem to be 48-49mh/s with a power draw bouncing between 130-140 watts.
Edit: let me add these screenshots that may be helpful for you…
Now you got me a bit confused.
That’s windows right? Shouldn’t that +2000 be +4000 in linux? Best I’ve seen stable with samsung memories was +2600 or so.
Benchmarks do report higher hashrates in windows but they aren’t long running so typically they aren’t stable settings. Did you have those speeds up for over 12h or 24h?
Unfortunately the kh/w seems to always get worse past 43-44MH/s 
Yes this card currently is running in windows and it has Hinyx memory too, which i would assume is the type v2 but that is just an assumption. You are correct though, the mem clocks I am speaking of are the Windows versions so you would need to double them for the Linux equivalent. It will go into a hive rig once i am done trying different settings, algos, etc. I just got it on Friday and through it into a windows rig but didnt have time to play around with it so I just set it on locked core 1350, mem +1400, pl 100% and it ran like that for the first approximate 12hrs or so, then bumped it up to the locked core 1400, mem +1800, pl 100% which it ran like that for about 24 hrs and then i moved it to the settings you see above which all seemed stable without any problems. I did play around with it last night a bit to see how far i could get the lhr-tune up on it, which the highest was 78.5.
I believe I had heard that there were 2 versions of the Hynix memory, a v1 and v2. V1 could not be overclocked very high and sometimes needed to be underclocked even and the v2 which needed/could be overclocked much more similarly like Samsung’s memory. Dont quote me fully on that because i am not 100% on that though. The YouTuber CryptoDonkeyMiner mentions something like this in this video https://youtu.be/k5lFnQC0Mto
It’s possible memory manufacturers improve the same model over time if there are known issues. When you plug it into hive (or if you wanna check gpuz), check the bios number:
Different memory chips require a different bios, so we can make sure we have the same card that way.
But yeah +2000 (4000) is crazy for a samsung mems, let alone hynix… I left it running with 1800 (+900 on windows) on both cards and it has been 40h without crashing, safe to say it’s stable. Will try upping as you did 
PS: looked again and can’t find anyone with stable settings over +3000 (+1500 on windows) and that’s samsung mems… wth There must be something with these cards where the mem values are not halved in afterburner for windows… can’t possibly be +4000
What other settings apart from lhr-tune? Best i can do without triggering the lhr lock is 74.5-75. Anything higher triggers it really quickly and t-rex has to adjust and ends up at 74.5 anyway. But other settings affect lhr lock too, such as intensity, so please share your full miner settings too (this difference could also be due to drivers, hiveos uses older drivers)
Can hold 2200 too it seems. Will let it run for a while and update
Seems to be the same Bios version as mine
Great news then! Don’t think I’ll ever try +4000 but It’d be insane to even get +3000, that’s already super good for samsung mems… currently stable at 2200, will ensure stability before going up as this is already a big jump from 1500 x)
Upping core clock brings more MH/s but reduces the kh/w, so I’m keeping it at 1410 for the time being. I’ve tried all the way up to 1550 and the MH/s can get close to 47 indeed, but not worth for the watts (close to 130-140w for the 47MH/s). Kinda stable at 44.5-45.5 with mems @ +2200.
PS: that finnish guy you linked is really funny. Have watched him before. I also tried that “amazing new boost in performance” from the latest gminer but gained absolutely no boost, still mines slower than t-rex, lhr-tune value can’t go as high (74.6 on t-rex, 73.8 on gminer) cause it defaults to max intensity which ends up at lower kh/w for the same speeds or lower and higher fee…
Getting there… I’ve found meanwhile people doing +3600 on hynix mems, so your +2000 windows numbers aren’t that far off, the question is: is it stable @cmarcum? Did you manage to run those +2000 for a few days already?
I did only 4h each instead of 24h.
+3800 seems stable… playing around with core clocks now to see the limits.
efficiency is much worse with core like this I’m just playing with it out of curiosity. Can probably reach 50 MH/s with a slightly higher core but not worth for the extra watts. Still amazing to see a shitty LHR getting to 50MH/s
1 Like
Update: Yes been running with my mem clocks at 2000 on windows for couple weeks now and stable without any problems. Never crashed yet…
I must have been super lucky in the silicone lottery. My Hynix memory 3060ti XC LHR gets 47.24 MH. I know memory is high, but I have had no crashes and temp is stable. Mined like this for multiple days with no issues.
Your Core is high, not the memory - you’re barely doing any OC at all there.
This is insanely inefficient and nothing to do with sillicone lottery. LHR unlock mechanism works better the higher the core, but that doesn’t mean it’s more efficient. The only reason you even see an increase in hashrate is because of LHR, with FHR cards you actually underclock the Core as low as nvidia allows - it’s the LHR unlock making use of core to fool the LHR lock, not your ETH mining per se.
I can get 47 MH/s with core at 1450, which consumes significantly less than what you do. Lock your core at stock (1410Mhz for mine) and test your Memories’ OC limits first. Once you know how much you can up the memories, then you start very slowly upping the Core clock until you find the sweet spot. For me the sweet spot is at 1470Mhz, higher than that and the hash per watt goes down exponentially.
For comparison, with your displayed 145W consumption I can get nearly 50MH/s, but not worth it due to the hash per watt ratio going down and electricity being expensive here. Currently I am at 1470 locked Core and +3800 Mems (linux numbers, would be +1900 in windows), this produces ~48-49MH/s with 130-140W consumption. Adding PL will not help either, as that consumption is required for the core for LHR unlock reasons described above, or worse it can crash your cards/drivers.
PS: your temps are also shit and your cards won’t last long-term with those temps 24/7. 62ºC core temp means close to 80ºC junction/hotspot temp and depending on your model, most likely around 100ºC. While they are built to handle that, they aren’t built or tested to run that 24/7 over a period of months. And assuming it’s also Winter where you live, it’s only going to get worse.
Sorry for all the edits, I previously had core at 1475 but recently lowered to 1470 as it’s the real efficiency sweet spot for my cards.
1 Like
Hi Guys, at least i’ve found a topic interesting about 3060 Ti LHR
I’ll share you my expercience, I’ve bought about 34 of this cards, waiting for the 100 unlock, and it finally arrived. Until that moment there was everything OK.
just using recomended version of hiveos and drivers, everything worked fine. Just checking OC in hashrate.no and having keep up to date that values.
The Card in Gigabyte Vision 3060Ti.
Unfortunately most of these have Hynix Memory. Only 5 of all of them have samsung memories of 34.
Enviroment was always controlled.
After some days i began to have problems with the cards. One by one, in differents rigs they became to loose mh/s. I began to have error messages in the cards, unlocking fail and finally Mh/s are about 20.
I’ve tried with t-rex, lolminer, nbminer, gminer… and different OC settings, after some hours, results are similar.
Factory warranty didn’t help even this have 2 o 3 months…
the thing is that actually i have 8 card with this issue, in 4 different rigs.
maybe, i want to try limiting LHR, but i don’t get with the point for at least keep them working the best possible.
have you any ideas?
1 Like
Fine tune the oc per individual card. Don’t use core offsets. Find the highest stable memory clock, and the lowest locked core clock that maintains full hashrate. If you’re getting crashes to half or less hashrate, continue to lower the mem clock and reboot each time. Repeat until stable
I’ve 34 cards 
it’s hard for tracking them
Reduce mem clock each time a card crashes and reboot. Shouldn’t take long to get them stable.
Replace risers and wires maybe? Are you getting XID ( Driver Errors ) in the syslog? My guess is you are because it looks like it isnt reading your fan speed on one card currently in the pic. I am on the same Driver version that your showing running with T-Rex v0.26.5, with EVGA 3060ti XC LHR and EVGA 3060ti FTW3 LHR, both with Hynix Memory. Hynix memory you have to clock way higher on the memory to get decent results. Here are a few stable settings I use:
Best Efficiency/Hashrate w/Fixed CoreClock
Core: 1380
Mem: 3500
Pl: 0
Hashrate: 61.27mhs @ 129w
Best Hashrate w/Fixed CoreClock
Core: 1575
Mem: 4000
Pl: 0
Hashrate: 67.73mhs @ 153w
Best Efficiency/Hashrate w/Offset CoreClock
Core: -600
Mem: 3500
Pl: 137
Hashrate: 64.63mhs @ 137w
Hope you get it, or have already got it figured out.
Keep in mind there are 2 different versions of Hynix memory, referred to as v1 and v2. The early v1 will sometimes only do ~1500mhz while the latter will do more than double that without issue