Great info here. I've had my R815s turned totally off for a bit... all Quad 6380 machines.
I am using XMRig for testing RandomX now and will use it for production when the fork happens November 30th. I am so done with XMR-STK with the abusive nature of the developer.
XMRig is top notch. You can also compile from the source and set the donation level to 0% if desired.
xmrig 4.5.0-beta (unified 3 in 1 miner, CPU+OpenCL+CUDA)
https://github.com/xmrig/xmrig/releases/tag/v4.5.0-betaHere is what I do for testing right now:
RandomX testing
wget https://github.com/xmrig/xmrig/releases/download/v4.5.0-beta/xmrig-4.5.0-beta-xenial-x64.tar.gz
tar xvf xmrig-4.5.0-beta-xenial-x64.tar.gz
cd xmrig-4.5.0-beta
edit config.json:
"donate-level": 1,
"algo": "rx/0",
"pools": "rx.minexmr.com:4444",
"user": "47wcnDjCDdjATivqH9GjC92jH9Vng7LCBMMxFmTV1Ybf5227MXhyD2gXynLUa9zrh5aPMAnu5npeQ2tLy8Z4pH7461vk6uo",
./xmrig
One thing to note that I have seen on the quad 16-core 6378 Opteron's is that disabling one core in each of the eight NUMA nodes increases the hash rate slightly while reducing power and heat slightly. The same reduction of one core per NUMA node may also be beneficial on your quad 16-core 6380 Opteron's.
Also note that you will need at least 8GB per Opteron or 32GB minimum for each R815 with quad Opterons. RandomX needs 2GB+256MB per NUMA node. Each Opteron has two NUMA Nodes so that equals 4.5GB per Opteron. That means that 4x 1GB sticks per Opteron will not work. You will either need 8x 1GB sticks or 4x 2GB sticks per Opteron at the minimum.
THANKS. Did you use anything else on the command line/config file or CPU set up? Or is it optimal to let XMRIG figure it out?