reported vs effective eth hashrate

[nerdralph]

I've been tuning both my cards and mining software for over a year now, and until recently I've been focused on the hashing speed reported by the miners.  In the last few months, instead of my fork of Genoil's miner, I've been using sgminer-gm (kudos to Wolf).  Sgminer, like ethminer-nr, reports errors that occur when a share that is found by the GPU fails to verify on the CPU.  What I like about sgminer is that it shows the effective hashrate after accounting for these errors (WU).  I've confirmed the reliability of the worker utility by comparing to pool average hashrates over more than a week.  What I've found is that my well-tuned cards (mostly Tonga) get an effective hashrate that is 98-99% of the gross rate.  For example a Gigabyte R9 380 Windforce with Hynix RAM with custom straps (basically my public strapmod + tighter tRFC) clocked at 905/1625 has a gross hashrate of 23.61 and WU of 23.26.  My Asus strix Rx 470 however shows 26.37 for WU vs 27.43 gross hashrate with RAM clocked at 2Ghz.  That suggests I have some work to do on tuning the Strix...

[1715327335]

1715327335

[1715327335]

1715327335

[1715327335]

1715327335

[1715327335]

1715327335

[whitrzac]

Lower your OC a bit and see if it becomes more stable.

I have a hynix card that the memory will clock at 2200+ at 1500 strap, but isn't stable above ~1900. It would lock the system a few times a day until I figured it out.

[nerdralph]

Lower your OC a bit and see if it becomes more stable.

I have a hynix card that the memory will clock at 2200+ at 1500 strap, but isn't stable above ~1900. It would lock the system a few times a day until I figured it out.

It's unstable at 2050.  At 2025 I've run it for over a week, but the error rate is high.  I also might make some tweaks to the strap to see if it will reduce the error rate.