Quote from: dragonmike on March 30, 2019, 05:54:10 PM

Quote from: seefatlow on March 30, 2019, 09:19:07 AM

OK. I have been trying over several days to replicate the results from pbfarmer on cnv8_trtl and I can say that I am unable to get the efficiency reported by him. Any attempt to reach close to his reported voltage levels of 805mv to 840mV resulted in random DEAD GPUs(stuck in enqueue) after 20-30 minutes hashing or significant hashrate drop per GPU

Even at 870mV, I am still unable to get a stable system running. Dead GPUs and hashrate drop to 15kH/s still occurs after several tens of minutes of mining. And the weird thing is hashrate drops happens to Hynix mem GPUs only. Huh

Somehow CN-TRTL algo is more taxing that CN_R? I am able to run CN_R without failures for a week and with lower voltage settings (850mV - 870mV). I can't seem to do this for CN_TRTL

My settings are as below. But the hashrates aren't sustainable

GPUs : Ref Vega 64 and Vega 56 reference bios
V64 cclk/memclk 1220/1100 @ 870mV L28+28 (Samsung mem) - 19.5kH/s
V56 cclk/memclk 1220/940 @ 870mV L24+24 (Samsung mem) - 19.3kH/s (Hynix mem) - 18.7 kH/s
ATW power draw 190W per GPU
Adrenalin Driver 18.6.1

Kerney/Todd,

Any ideas what could possibly be wrong? Unoptimized CN_TRTL code?

That's quite weird, out of my six reference Vega56's (flashed to 64), 4 are hashing rock solid at 1408@825 core / 1107 mem and the weaker two cards get 850mV.
I reckon I could potentially go even lower.
Have you fiddled with your power play tables? Is your SoC set higher? That's quite typical for instability, I can sing a song or two about it...

Where do I see the SOC voltage and where can I set them? My ppt tables starts with 800mv for P0 and moves up to 820mv for P3-P5. Do you see a problem with this?