T rex dual (Eth + Ergo) issue, and NO OVERCLOCK

[Mardreck]

Hello team,

I am mining with a 3070 ti LHR Since yesterday suddently I am having an issue with trex dual ergo eth mining (I changed nothing, it worked well). The error in red says: TREX Can't find nonce with device (ID=0, GPU=O), Cuda Exception: CUDA_ERROR_ILLEGAL_ADRESS, try to reduce overclock to stabilize GPU state. Of course I set GPU to default state, no OC and it is bringing me the same error. When I mine ETH Alone the system works properly even when I do OC (I always do low aggressive OC, never crashed).

I tried to increase VRAM through windows alocating min 2000 and max 8000 (I have a 3070 ti LHR)

Thank you all for your help.

[i1itione]

I guess that Eth+Ergo dual mining needs above 8GB VRAM, because I use 3080 Ti and it uses above 9GB.

[Mardreck]

Hi thanks for the answer. But i waa mining quite happy until 2 days ago. I saw on another topic that this is happening to more people. Thanks

El 19 mar 2022, a las 4:42, i1itione @.***> escribió:

 I guess that Eth+Ergo dual mining needs above 8GB VRAM, because I use 3080 Ti and it uses above 9GB.

— Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android. You are receiving this because you authored the thread.

[grenelt]

@Mardreck You are aware that the DAGs are constantly growing? And when mining at a vram limit stops isn't only a matter of DAG size, but also of vram allocation from the driver. Depends on OS, driver version, display resolution (if one is connected...). So not all users face the problem at the same time, but the reason is the same.

[coryb13]

@Mardreck You are aware that the DAGs are constantly growing? And when mining at a vram limit stops isn't only a matter of DAG size, but also of vram allocation from the driver. Depends on OS, driver version, display resolution (if one is connected...). So not all users face the problem at the same time, but the reason is the same.

See more context in my post on the following thread, but it's working fine on a 3060 Ti, but not a 3070 Ti. Both have 8gb ram.

https://github.com/trexminer/T-Rex/issues/1162

Note that I am having the same problem as OP, but only with the 3070 Ti cards.

Edit: Any reason gddr6x memory or cards might take more memory than gddr6? I'm trying to figure out why this might be happening with the 70 Ti, but not the 60 Ti... I just checked, and my 80 only has 1.93 gb left of 10gb, which does go with what you said, @grenelt, that the dag moved beyond the 70 Ti's 8gb memory.

[i1itione]

I found this parameter " --dataset-mode" and its default is double which means use double size, so I guess if you turn off this feature, 8GB VRAM is still enough for ETH+Ergo currently.

[storopoli]

--dataset-mode 1 solves the problem for me with 3070Ti and t-rex 0.25.9. Thanks!

[PrinceOfParallax]

--dataset-mode 1 solves the problem for me with 3070Ti and t-rex 0.25.9. Thanks!

Did not work for me.

[mayronguevara]

a workaround is use T-rex v. 0.24.2 , works for dual mining. for that i think the DAG in the network is not the problem, if not, fail with all version of the software. i don't know.

[trexminer]

Please try 0.25.12. CUDA_ERROR_ILLEGAL_ADRESS is a misleading error. In reality the miner failed to allocate memory for the dataset, i.e. GPU doesn't have enough VRAM.

[storopoli]

Yes! I just tested 0.25.12 works like a charm on 3070 Ti. Thank you @trexminer! Let other users report back and maybe we can close the issue.