I tried to reproduce the issue with clocks and low speed until miner restart.
Windows 10, latest blockchain drivers, stock RX580 Nitro+.
1. No overclock: If I start miner in ETH-only mode, it shows about 16MH/s only (and low memory controller load in GPU-Z) and I see this speed for about 20 seconds, then it shows 24.6MH and works fine. For dual mode it shows full speed immediately. Unfortunately, I have no idea why GPU works slowly within first 20 seconds, it is something related to drivers.
2. I close the miner and apply some changes in clocks: memory 2000 -> 2100, core 1300 -> 1100. When I start miner, I see that it applies new clocks in about 10 seconds, but still shows 18MH/s only for first 20 seconds. Then it shows about 25.5MH.
Note that I use default ROM so these speeds are not high.
3. I close miner and apply new clocks again: -cclock 1000, -mclock 2050. Same behavior: it shows low speed at start, then normal speed for these clocks. GPU-Z shows changes in clocks in about 10 seconds after start.
So right now I cannot reproduce the issue with new clocks: in my tests new clocks are applied immediately, I don't need to restart the miner or reboot the system.
Windows 10, latest blockchain drivers, stock RX580 Nitro+.
1. No overclock: If I start miner in ETH-only mode, it shows about 16MH/s only (and low memory controller load in GPU-Z) and I see this speed for about 20 seconds, then it shows 24.6MH and works fine. For dual mode it shows full speed immediately. Unfortunately, I have no idea why GPU works slowly within first 20 seconds, it is something related to drivers.
2. I close the miner and apply some changes in clocks: memory 2000 -> 2100, core 1300 -> 1100. When I start miner, I see that it applies new clocks in about 10 seconds, but still shows 18MH/s only for first 20 seconds. Then it shows about 25.5MH.
Note that I use default ROM so these speeds are not high.
3. I close miner and apply new clocks again: -cclock 1000, -mclock 2050. Same behavior: it shows low speed at start, then normal speed for these clocks. GPU-Z shows changes in clocks in about 10 seconds after start.
So right now I cannot reproduce the issue with new clocks: in my tests new clocks are applied immediately, I don't need to restart the miner or reboot the system.
I have an update regarding this issue:
- First I told you it happened on my 480 ASUS DUAL 4GB that was having an OCL thread hang, then at restart the GPU would stay in a lower performance mode, albeit having the clocks applied at miner start.
The problem is that this issue with non-rounded milivolt values with the miner makes the GPU have an OCL thread hang AND reset it's clocks. Then with the clocks reset when you open the miner back after the restart changing the default clocks/voltages makes the GPU enter in this lower performance mode.
With this ASUS DUAL 480 4GB, 1050core 1755mem and custom strap it hashes 25,35mh/s normally, 21,9mh/s on this lower performance mode. Restarting the PC makes it go back to the higher hashing value.
I have this same problem. It's like my rx 480 4gb drops into limp mode and hashes at half the speed until I restart the miner.
