Vega 64 memory clock stuck dual mining ETH-TON

[n392K8Z7]

3 out of 4 vega cards are working as expected, but the 4th one has an issue with running on the set clocks. HiveOS settings:

It does not seem to run on the set clocks (maybe stuck in some p-state?):

If I set the SOC frequency manually, it seems to go a bit higher, but still not the same memory clock as the one that has been set:

Here is the used flightsheet:

The same card is running fine on single-algo mining ETH, this setting is running without restarts for many months now:

Here is the flightsheet for single-algo mining ETH:

If I stop mining and change the memory frequency on the card, it seems to get "unstuck" and works again, but if I start dual-mining again, the same card gets stuck again. It is always the same card, the other cards are working fine. I have tried dropping the extra config arguments for ETH and SSL to TCP for the eth pool, but does not seem to make any difference.

Do you have any ideas on what could be the issue? Maybe a bug in the dual mining feature?

[lian2468]

I have the same problem in all 4 of my Vega rigs, from some reason there is always one card that is with reduced ETH hashrate no matter how many cards there is in the rig. is there any command that we can put just to neutralized the specific card from running dual mining ?

[n392K8Z7]

Yes, from the dual-mining eth txt:

If you don't to dual mine on all devices, you can add a device selection argument inside the --ton .. --ton_end clause. If you only want to dual mine on gpu 0 and gpu 3 you would use:

--ton -d 0,3 -o stratum+tcp://ton.hashrate.to:4002 -u . -p x --ton_end

https://github.com/todxx/teamredminer/blob/master/doc/DUAL_ETH_MINING.txt

[lian2468]

I think this can help us, I need to play with it : --dual_intensity=X,Y,Z,... The Vega and Polaris implementation can adjust the intensity of the dual mining with a factor between 0.0 and 1.0, where 1.0 means running as much dual algo as possible. This option is recommended for forcing down power draw without lowering core clk, which will have an impact on the ethash hashrate as well. For now, this option is ignored for Navi and Big Navi, support will be added in a future release.

                          The argument is provided either as a single global value or a list of values per gpu.  Examples:

                            --dual_intensity=0.75            All gpus use 75% of max possible TON gpu availability.
                            --dual_intensity=0.75,1.0,0.5    Separate values for the first three gpus in the rig, remaining gpus
                                                             use the first value for 75%.

  --dual_tuner_weights=X:Y:Z  The dual mining tuner needs scoring weights for how much 1 MH/s of the primary and dual algo
                          are worth.  There are defaults built into the miner that were valid when each algo was implemented
                          and added, but as network hashrates and market prices move, new values may need to be provided. 
                          This argument provides three values, all expected to be positive numbers:
                            X    Score for 1 MH/s of the primary algo (typically ethash). 
                            Y    Score for 1 MH/s of the secondary algo (typically TON). 
                            Z    Cost per 1 W of measured power. We recommend setting this to zero for now. 

[lian2468]

Yes, from the dual-mining eth txt:

If you don't to dual mine on all devices, you can add a device selection argument inside the --ton .. --ton_end clause. If you only want to dual mine on gpu 0 and gpu 3 you would use:

--ton -d 0,3 -o stratum+tcp://ton.hashrate.to:4002 -u . -p x --ton_end

https://github.com/todxx/teamredminer/blob/master/doc/DUAL_ETH_MINING.txt

I have tried this command, it is not mining TON, but the ETH Rate is still stuck on 8 MH

[n392K8Z7]

1. Stop mining
2. Change the memory frequency to like 1000 MHz
3. save the settings
4. change the memory MHz back to the desired value.
5. save settings
6. start miner

It will "unstuck" the memory frequency, and it will work with full performance mining only ETH afterward if you set the device selection argument inside the --ton.

I have tried the intensity setting as well on the affected card, it increased ETH performance a little bit, but reduces TON performance by a large amount.

The default SOC was 960 MHz, which resulted in about 8MH/s for ETH and 1,3GH/s for TON for the affected card.

Setting the SOC to 1060 MHz manually, resulted in 41MH/s for ETH and 930MH/s for TON

Setting SOC to 1060 MHz manually and changing the intensity (with --dual_intensity) to 0.3 (30%) on the affected card, it increased the ETH further while reducing TON: 43MH/s for ETH 377MH/s for TON

Changing the intensity without setting SOC MHz did not increase ETH by any noticeable amount, but decreased TON performance significantly (ETH stayed the same around 8MH/s, but TON reduced from 1,3MH/s to 370MH/s)

[lian2468]

Thank you for the instructions !!! it solved the issue

[n392K8Z7]

No problem :) It is just a "workaround", so the affected card that has reduced performance on dual mining can single-mine ETH with full performance. Hopefully, we can find out/fix the issue so the cards can dual-mine with full performance.