3090 Rog Strix OC 3.74 MHs Low Hash Rate. New Thermal pads. Need Help

[onicoworker]

Detected 6 NVIDIA cards

ERROR: X Server is not running! Some settings will not be applied!

GPU BUS ID : 02 04 05 06 07 08
CLOCK : 1300 0 -150 0 0 0
MEM : 2650 3000 3000 2900 1000 1000
PLIMIT : 317 339 335 335 327 311
FAN : 90 90 90 90 90 90
FANCNT : 2 2 2 2 2 2
Unable to init server: Could not connect: Connection refused

ERROR: The control display is undefined; please run nvidia-settings --help for usage information.

(exitcode=1)

=== GPU 0, 02:00.0 GeForce RTX 3090 24268 MB, PL: 100 W, 350 W, 365 W === 16:29:22
nvtool failed by timeout (exitcode=124)
nvtool failed by timeout (exitcode=124)
Max Perf mode: 4
Unable to init server: Could not connect: Connection refused
ERROR: The control display is undefined; please run nvidia-settings --help for usage information.
(exitcode=100)

=== GPU 1, 04:00.0 GeForce RTX 3090 24268 MB, PL: 100 W, 350 W, 365 W === 16:30:12
nvtool failed by timeout (exitcode=124)
Max Perf mode: 4
Unable to init server: Could not connect: Connection refused
ERROR: The control display is undefined; please run nvidia-settings --help for usage information.
(exitcode=100)

=== GPU 2, 05:00.0 GeForce RTX 3090 24268 MB, PL: 100 W, 390 W, 480 W === 16:30:24
SET GPU CLOCKS: 0 MHz
SET POWER LIMIT: 335.0 W
Max Perf mode: 4
Unable to init server: Could not connect: Connection refused
ERROR: The control display is undefined; please run nvidia-settings --help for usage information.
(exitcode=100)

=== GPU 3, 06:00.0 GeForce RTX 3090 24268 MB, PL: 100 W, 390 W, 480 W === 16:30:24

[onicoworker]

i was able to get out this code lines by pushing something lol. the thing is i had 4 cards 3090s running no problem, i then added 2 new ones with a new PSU just for them 1000w, and like i said if i put them completely stock no OC the rig EVENTUALLY starts after multiple hard reboots, i can remote it , and then they run, i apply some light OC and then try reboot, nothing again

[keaton_hiveon]

Poor clocks, use locked core clocks on all cards, as low as maintains full hashrate. No need for power limits that way.

Your memory is pretty high as well which I would guess is the cause of your xserver/driver crashing. Set some more conservative mem clocks and see if it’s stable

[onicoworker]

i have benched all cards individually and very spesifically in windows using t rex afterburner so the clocks are not random, each and every one including power limit is very spesific for each card and have been working really well until i added the 2 last cards, never had a crash etc, it could be the oc settings but they should be able to take 10% more before crashing at least. Also it was my understanding that setting core at 0 is locked core ? just locking it at 0 or do you have to have a fixed value like -1 or +1 etc?

[keaton_hiveon]

You should be using locked core clocks for all cards, you’re using it for one.

1300mhz is way more than needed on a 3090 for eth alone unless you’re trying to set a world record on ln2 or something.

Leaving the other cards at default core clock and limiting them with power limit is counter intuitive. Just set a proper core clock (lowest that maintains full hashrate as I mentioned above) and there’s no need for power limits at all as the card will only draw what is needed.

[mrnoise]

cc and mem clock 0

[keaton_hiveon]

That’s good settings for gaming, terrible for mining.

[onicoworker]

well those settings would drop the total hashrate 100mhs

[onicoworker]

the reason i choose that powerlimit is because of the heat and when i was testing it, lowest powerlimit for the least ammount of hashrate drop. some of my 3090s are hasing 126.1mhs that is like peak world performance on eth and with 3090s.

I wasnt aware that locking core will stop the card from drawing power, i was thinking it can still fluctuate up and down, 99% of all hive miners i see on youtube reddit etc with 3090 power limits

[keaton_hiveon]

Use the core clock to regulate power draw

[mrnoise]

im use
cc 1050 or 1333
mem 1900 to 2333

[onicoworker]

well i have seen all there is of mining blogs and youtube seems to me like everyone is power limiting 30 series with using PL. and i never had any issues for 3 weeks running the 4 x 3090 when i added the last 2 is when all the problems started

[keaton_hiveon]

Well as long as you’re aware of the better way to run them now, you can use whichever way you’d like. Core offsets and power limits are the old less efficient way.

[onicoworker]

thanks for all your help keaton, much obliged!!!