Thanks for the suggestion. I did that, but no matter what value I used, the card kept crashing, so I decided to left it untouched.
That worked for a couple of days, and then GPU2 crashed. Then I decided to play with the voltages as you suggested.
I ended with this configuration:
Core clock 965
Core voltage 650
Memory controller voltage 680
Memory clock 1130
Memory voltage 1300
Fan 60
That worked fine for all card including GPU0, meaning no one crashed, but GPUs 1 to 5 had a Hash rate of 32+ and GPU0, that was around 28 without any change, dropped to 15-16. I removed the changes from that GPU and it went back to 28 (but is consuming 66w vs 53w for the other ones.
I guess the problem is because GPU0 has the screen required for the boot in the HDMI port. This motherboard does not boot w/o a monitor plugged there, and the onboard HDMI port is useless.
Of course I stopped all activity on the console and unplugged the monitor, but the problem remains.
I guess I have to wait till there is a AMD driver or BIOS update that address this condition.
Hopefully this is helpful for other people or maybe someone has something to say.