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?
No command line options or changes for the Dell R815 Server with the quad 6348 Opteron's.
As I stated on the Dell R815 Server with the quad 6378 Opteron's it is best to disable one core per NUMA node to gain an additional 1% in hash rate while very slightly lowering power consumption and processor heat. This is done by first running XMRig to that it fills in the CPU core options in the config.json file. Then exiting XMRig and now editing the RX section to remove one core from each of the eight NUMA nodes.
I reformat the very long RX CPU section [0, 1, 2, ... 61, 62, 63] to this:
[0, 1, 2, 3, 4, 5, 6, 7,
8, 9, 10, 11, 12, 13, 14, 15,
...
48, 49, 50, 51, 52, 53, 54, 55,
56, 57, 58, 59, 60, 61, 62, 63]
that way I can easily see each NUMA node. I then delete the last entry on each line. Be sure to leave the trailing comma in the first seven lines and on the last line remove the ", 63" but leave the "]" bracket.
Now when you run XMRig it will use 56 cores when mining.
THANKS. Do you think perhaps the 52 MH/S of Loki hashrate is latent XMR hashrate? i.e. machines that are commissioned and ready to move to XMR? The benefit of doing this to know the machines work and to make tweaks and test airflow etc