Misbehaving 3060ti and 3070

[turboraketti]

I frequently get errors from two of my cards in my rig of 12. It’s either GPU0, which is a 3060ti, or GPU11, which is a 3070. T-Rex Miner gives me these lines frequently:

WARN: NVML: can't get fan speed for GPU #0, error code 999

[FAIL] 39/43 - Low difficulty or invalid share, 34ms ... GPU #0

When I try to modify overclock settings in the web GUI I get this type of error:

=== GPU 11, 10:00.0 GeForce RTX 3070 7982 MB, PL: 100 W, 270 W, 300 W === 16:49:31 SET POWER LIMIT: 125.0 W [Unknown Error] (exicode=123) Max Perf mode: 4 (auto) ERROR: Error assigning value 95 to attribute 'GPUTargetFanSpeed' (NB1:0[fan:16]) as specified in assignment '[fan:16]/GPUTargetFanSpeed=95' (Unknown Error). ERROR: Error assigning value 95 to attribute 'GPUTargetFanSpeed' (NB1:0[fan:17]) as specified in assignment '[fan:17]/GPUTargetFanSpeed=95' (Unknown Error). ERROR: Error assigning value 0 to attribute 'GPUGraphicsClockOffset' (NB1:0[gpu:11]) as specified in assignment '[gpu:11]/GPUGraphicsClockOffset[4]=0' (Unknown Error). ERROR: Error assigning value 0 to attribute 'GPUMemoryTransferRateOffset' (NB1:0[gpu:11]) as specified in assignment '[gpu:11]/GPUMemoryTransferRateOffset[4]=0' (Unknown Error). Attribute 'GPUFanControlState' (NB1:0[gpu:11]) assigned value 1.

I run on the latest stable, but has also tried latest beta. A reboot seems to fix it, but it comes back quickly on either GPU #0 or #11.

Any ideas why this is?

[givo]

These kind of problems are often caused by bad risers, wiring, connections, … Try switching cabling with neighboring cards and see if the problem stays with the same card or has moved to neighboring card…

[turboraketti]

Thank you! It seems to have to do with the riser not recovering from OC out of GPU bounds. After a lot of experimenting it seems that once I happen to tweak the overclocking out of the working range, so that I’m getting errors, the riser has to be power-cycled for the unit to start working again. A simple reboot is not sufficient, the riser needs to be powered off. And, of course, before powering it off I need to restore a working configuration in HiveOS again, or the problem will be right back when the miner starts, even if I issued a power-off.

The remedy for me seems to be:

• Restore a known working configuration for the card
• Power off completely
• Boot up

It will probably work with replacing the last two steps with “Shutdown & reboot in 30 s” from the Hive web GUI. I also tried (as a proof of concept – not recommended!) to power off the riser while running, pulling its plug, and then soft-reboot, which also worked (but may be harmful to the hw).

[joejoecash]

Hi turboraketti,
I’m trying to do a little research on this fan bug. Would you mind telling me what risers you are using?

I’m actually trying to figure out a way to exploit this bug to get my 3060’s up to full hashrate as reported by a user in this thread:
3060 with risers : What to mine? - Nvidia Cards - Hive OS Forum

Thanks

[turboraketti]

Sure. I have two different kinds. This is what’s printed on them:

• PCE164P-N06 VER 008S
• PCE164P-N08 VER 0095

Some bags have “MAKKI Mining Riser” printed on it, but I’m not sure if that holds for both kinds.
Hope that helps.

[joejoecash]

turbo, thank you very much for your reply! I’ve taken note of your riser information & will try to line up a couple test risers.