Quote from: Uaciuganadu on December 05, 2018, 02:16:09 PM

Hello Dok,

I am having strange issues with Vega 64 and the Heavy algos.

I have discovered the issue during testing and it seems like the Vega cards only hash at full potential immediately after they have been reset. If i run the benchmark again after running another algo, it drops down in hash.

First run after device reset:

Code:

Thread[0]: 0.0 H/s <- RX580 8GB
Thread[1]: 0.0 H/s <- RX580 8GB
Thread[2]: 1033.0 H/s <- Vega 64
Thread[3]: 1046.0 H/s <- Vega 64
Thread[4]: 973.0 H/s <- Vega 64 low clock
Thread[5]: 980.0 H/s <- Vega 64 low clock
Total: 4032.0 H/S
Thread[0]: 555.0 H/s <- RX580 8GB
Thread[1]: 555.0 H/s <- RX580 8GB
Thread[2]: 1041.0 H/s <- Vega 64
Thread[3]: 1042.0 H/s <- Vega 64
Thread[4]: 980.0 H/s <- Vega 64 low clock
Thread[5]: 979.0 H/s <- Vega 64 low clock
Total: 5152.0 H/S

Second run with no reset after running a Bittube bench:

Code:

Thread[0]: 0.0 H/s <- RX580 8GB
Thread[1]: 0.0 H/s <- RX580 8GB
Thread[2]: 1035.0 H/s <- Vega 64
Thread[3]: 1027.0 H/s <- Vega 64
Thread[4]: 973.0 H/s <- Vega 64 low clock
Thread[5]: 970.0 H/s <- Vega 64 low clock
Total: 4005.0 H/S
Thread[0]: 555.0 H/s <- RX580 8GB
Thread[1]: 555.0 H/s <- RX580 8GB
Thread[2]: 1032.0 H/s <- Vega 64
Thread[3]: 1030.0 H/s <- Vega 64
Thread[4]: 971.0 H/s <- Vega 64 low clock
Thread[5]: 970.0 H/s <- Vega 64 low clock
Total: 5113.0 H/S

It seems like no mater what algo change happens after the first run, the Heavy algos ca not reach maximum potential again till the Vega`s are reset. I have also tested "heavy_mode" : 1 and 2 but has no effect on the issue.

Can confirm this behaviour on even on CN v7 algos, but --resetvega fixes these
Just waiting for th CN lite code rewrites