With the TRM 4.4 release (cn_config 14*14) and timings set to: --CL 20 --RP 11 --RC 44 --RCDRD 12 --RCDWR 8 --RFC 250 --FAW 20 --RRDS 4 --RRDL 4 --RAS 32 --REF 7800
Quad Vega 56 flashed with V64 BIOS 1447/1050MHz (1340MHz actual) at 850mv I am getting 8670H/s on CNR pulling 740W at the wall.
This is actual hashrate that shows up also poolside. I can run much tighter timings at much higher hashrate without any apparant errors, but poolside the hashrate drops because of the errors induced. It would be nice to have a benchmarking tool to see how stable your settings are before you start mining.
I wrote a long recipe-based tutorial included in the release for v0.4.4 (CN_MAX_YOUR_VEGA.txt). The last step in that document describes how to run tests to assess poolside hashrate. There is no simple way to run a quick benchmark test and assess the quality of hashing under a specific set of modded timings/clocks, although having a tool with known result sets would help to estimate the nr of hw errs faster. In the general case, it is way too easy to be fooled by randomness here though. So, my recommendation is to run the recommended test in that document using xmrig-proxy for 50k shares and then check the poolside hashrate reported in TRM. It might very well be that you need to dial down your modded timings to something less aggressive in the end, just like your previous tests indicate, but then I'd know you've ran a sufficient test to zoom in on the true poolside hashrate.
Cheers, K