Teamblackminer crash after start

[joethebirder]

I have 6 1660Ti cards at the moment and this miner produce me the best hashrate for these cards (+33MH/s), but crashed a few minutes after start.

GPU4 Dag buffer verified ok!
GPU3 Dag verification failed. Exiting....
Shutting down threads.
Stratum thread exited.
GPU2 thread exited.
Control thread exited.
GPU1 thread exited.
GPU4 thread exited.
GPU0 thread exited.
GPU5 thread exited.
Api thread exited.Exiting.

The RIG will freeze and no remote way to reboot.

lspci | grep 1660
01:00.0 VGA compatible controller: NVIDIA TU116 [GeForce GTX 1660 Ti] (rev a1)
02:00.0 VGA compatible controller: NVIDIA TU116 [GeForce GTX 1660 Ti] (rev a1)
03:00.0 VGA compatible controller: NVIDIA TU116 [GeForce GTX 1660 Ti] (rev a1)
07:00.0 VGA compatible controller: NVIDIA TU116 [GeForce GTX 1660 Ti] (rev a1)
08:00.0 VGA compatible controller: NVIDIA TU116 [GeForce GTX 1660 Ti] (rev a1)
0c:00.0 VGA compatible controller: NVIDIA TU116 [GeForce GTX 1660 Ti] (rev a1)

I use the same OC settings for each: CORE-200, MEM2200, PL80.
Could anybody advise me how can I fix?

[nanial]

Hi,
What type of memory have your 1660ti ?

[keaton_hiveon]

try switching to a locked core clock, and lower the memory clocks on any card that is crashing

[joethebirder]

Hello! Samsung memories.

[Lukezen]

With TBM you need to back off the overclocks compared to all other miners

example: my 3070s

TBM
core 1100
MEM 2000

All other miners
core 1100
MEM 2400

TBM won’t even start with MEN as low as 2100

[joethebirder]

Tried lower clock, thanks the advice! Now its not crashing, just the ratio of rejected shares a bit more than expected. I should try other configuration.

[joethebirder]

Thanks your answer!
Have you better hashrate with TBM (lower clock) than others with the normal?

[joethebirder]

Crashed again with CORE 1800, MEM 2100, PL80.
In the log:

Stratum thread exited.
source/sp_cuda/cudaminer.cpp:cuda_search():607: A cuda runtime error occured: unknown error.
Shutting down threads.
terminate called after throwing an instance of 'cuda_runtime_error'
  what():  CUDA error in func cuda_work_thread at line 583 unknown error

[Lukezen]

Yeah you end up with about the same for some reason lol

Core 1100
MEN 2000

No PL, not required once cores locked

[joethebirder]

Yes, you are right, I have much better power consumption with locked core and without PL settings.
But now I still have very bad efficiency ratio (around 93%). Accepted 56 and the rejected is 5.

[mdrodge]

Invalid or Rejected?
Rejected = problems like submitted share to slow or bad internet. (not on the best pool for your location? Poor WiFi or bad LAN Cable? A router/modem that is melting or just made of crap?)

Invalid = to much overclock.

%93 is not grate but it’s not insane so my guess is try a closer pool.
watch a Son_of_a_tech video about choosing the best pool

Also 60 shares is not a lot to build an average from
((sometimes on boot I get a bunch of bad shares, then it doesn’t miss a beat for days witch I think is because of my less than ideal network config or a time server issue))

[joethebirder]

Invalid!
Thanks your advise!
Finally I switched to other miner, I cant set the overclock setting fine.
Only worked with low MEM settings and with this cant produce more than others, or may be a bit worse.

[mdrodge]

On my 16/20 series nV cards with GDDR6 run anyware between 2133 & 2366
most of the slower ones are Samsung and the faster ones seem to be Micron.
“Low mem settings” are better than bad settings.
(if your over 29mh on a 1660ti then your not doing to bad)

33mh from a 1660ti is insanely fast, on my 3 msi 1660ti’s I get 30.8mh stable for months with Micron mem at 2366

My 3060ti with Hynix GDDR6 only does 1666 so count your self lucky.

[mdrodge]

tough one thing to take a look at if you feel the memory isn’t performing is memory Temps (I know, I know, we are hassling the nvidia support staff in the dev support forum about allowing temps in the linux driver there response so far is it isn’t even in the windows driver, we pointed out that it’s a bad idea to lie about code to a forum full of devs and sys admins but anyway)

Maybe your thermal pads are letting you down or maybe it’s poor cooler design.
Though I believe you probably just need to be more realistic and expect the performance that is normal for the device if you had 1660 Supers you might be getting 33mh but 1660ti’s are more like 30-31mh
in my experience

I get best performance with nbminer

[joethebirder]

Thank you very much your advises, I really appreciate!
At the moment I try with different miner with different settings.
Its not fine tuned and run only about since a hour, but this configuration looks very hopeful.

Sometimes the same brand with same memory gives different results…

[mdrodge]

An overclock can take a few days to dial in.
Get it stable for a few hours but don’t be surprised if you check back 2 days later and find 1 invalid share.
If that happens just drop it a little more and wait.
If this doesn’t happen you might even be able to go up a couple more mhz.
Eventually it’ll settle down and you’ll have found max stable clock for 24/7 OC.