That is how LRH works. The card is detecting the ETH algo and is trying to limit it. The miner lowers tries to trick the card by lowering the hash rate, then it will increase it afterwards. If you are getting constant triggering of LHR then you need to adjust your overclocks.
thats normal, the miners will continue to decrease the percentage until the lock stops happening. trex 24.7 has the highest hashrates currently. make sure youre on latest drivers too.
I can only tell you my tune, only one of which has to be adjusted to try and avoid lhr.
My EVGA 3080 TI FTW3 does 86-87+ mh without ever hitting lhr with CC=1300, Mem= 2800, PL=290(using about 250w)
My EVGA 3080 TI XC3 does 78-80 mh generally hitting 73.5 lhr within a half hour with CC=1300, Mem=2450, pl=290(using about 257w)
With the XC3 the core clock seems to be the trick to avoiding lhr lock.
So the the “lhr-reduce-value”: “0.4” keeps it from dropping LHR at all (so you can adjust tune till it stops hitting lhr lock)?
or is it the value argument of 0.4?
Either way it’s brilliant. I’ve went from 423-426mh total, to 428-434+. Still working out the kinks though.
Looks to me like you’re setting the 3080 ti to lhr mode 2, and then to 72%
I’d lose the “lhr-mode”: “2,1,1” command and set all the gpus to “lhr”: “75,75,75”
And I’d start the 80 ti mem at 2000, and try 1900 for a few minutes before you go the other way.
btw, my 3060ti is set at Core=1400, Mem= 2600, and PL at 135 (45+ mh)
my 3060’s are at Core=1500, Mem=2500, and Pl at 120 (36.4 mh)
Both with lhr set to 75%. I think the core being locked and it’s level has the most to do with setting off the lhr lock.
Yep.
U can make a fine tune for every single LHR gpu by using this method. Key is locked core clock. If it is not, ur chance to get locked again increases because of high volatalite on mV which will drive miner to locked state again.
Also you may decrease the value once more by changing it until u see there is no “locked” state left.
After change your LHR% by increasing %0.1 and if u see u did not get locked nor decrease hashrate u may add another 0.1 to lhr.
Lock your cores and dont give them PL, they will find optimum mV by themselves.
and 2700 mem is too much. reduce it to 2.5k at the start and increase it 50 by 50 every 2 mins if you did not get locked.
in hiveos i just set the cclock and mclock, this is lock, i tried to put the cclorck and mclock command in T-rex but show a error that in Linux this command dont work