Why the hashrate on NEXA with a monitor connected to the gpu is so low?

[MolochBaal]

As an example, with a 3070 at core 1750 MHz I only get 55 MH/s, but on my rig with 6 3070 with the igpu connected to a monitor I get +90 MH/s at the same core frequency. Also it doesn't get better if I turn off the monitor.

And this behaviour don't happen with BzMiner (I get like 3 MH/s less with the monitor plugged, so like 88 MH/s), also tried with --screen -1 (wont run) and 0 (less hashrate)

--

Another problem that I have on both my pc and my rig from long ago is that if I set the memory clock (--mclk) to 5001 it keep on that unless I reboot the system, I thing is something related to power state. It doesn't happen with any other miner. So my solution is use --moff -1000. I mine different algorythms, so some need more memory and can't stay on 5001 neither 810. MHz

I use windows 10 22H2, nvidia drivers now 535.98, but I've tried with like every WHQL.

[jgonzis]

The problem is that it can memory is used and cannot load the full LUT, that is why it happens. I also recommend to use drivers 528.xx to allow the proper OC

[MolochBaal]

The problem is that it can memory is used and cannot load the full LUT, that is why it happens. I also recommend to use drivers 528.xx to allow the proper OC

I solved the problem using --keepfree it gets the propper hashrate (~90 MH/s) using values from 768 (this is memory that windows reserve) to 5120, higher values shows an error in allocating memory, and lower values gets low hashrate.

--

Rolled back to 528.49, but the problem with memory stuck in power state 3 (5001 MHz) or 5 (810 MHz) is still going.

My other issue is that I've never been able to use --coff, but it doesn't bother me too much.

--

Edit: Careful with windows auto updating nvidia drivers, seems like it doesn't like 528.xx, so disable auto updates for drivers.