I use stock cards for tests. Yes at some dcri value I can see memory controller load deviations in GPU-Z, it's about 90% but then it's 65% for a few seconds. For my Nitro+ 480 I see it at "-dcri 47". Try to increase -dcri value a bit, for me "-dcri 50" is stable.
Of course your values will be different if you use different card/bios/clocks.
Yes it is stable with dcri 50, however it decreases eth hashrate in greater amount.
Instead, if possible could you please double check miner code and see if miner works stable, without sacrificing ETH hashrate? For me "-dcri 30" is the most for ETH >> 29.3 ETH, 875 DCR but when using this value, gpus randomly have memory controller load deviations in GPU-Z, and this is seen through miner console. As a result lower effective hashrate is calculated on pool.
What if you use "-dcri 35"?
i tried that
with -dcri 30 i get 142 on eth and 4200 on dcr
with -dcri 35 i get 132 on eth and 4600 on dcr
it's stable but eth hashrate is much lower
What if you use "-dcri 33"?
Above I marked the key phrase ("try to increase -dcri value a bit"), but it seems I should say values directly...
I understand you'd like to see x2 hashrate and smooth work, but hardware has different opinion, memory controller goes crazy at some -dcri values and I don't see any workarounds for now (without losing hashrate).
For stock clocks/bios, increasing -dcri value +3 gave me -0.5% ETH of maximum but stable hashrate. May be it does not work for modded bios, I don't know.
That might be the case. Anyone else having speed issues with stock bios RX series?