Monero XMR Mining on 64 core AMD EPYC

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
7742 with the old xmrig container was ~5400 per CPU
7702p was ~5050
 

Klee

Well-Known Member
Jun 2, 2016
1,289
396
83
Monero is changing the algo the next fork to a new algo that is VERY gpu unfriendly and VERY favorable to cpu's and especially the new AMD cpu's.

They should be spectacular mining Monero after the change to the randomX algo in October.

My new Ryzen 3700x does 7700 H/s on randomx's benchmark, and the beta xmrig miner, with all 16 thread's totally stock with zero overclocking.

Just to compare, my dual E5-2667 V3 setup does 6600 H/s on 16 threads and 7600 H/s on 32 threads, but running 32 threads puts out alot more heat.

I have not tested with my kilowatt meter yet.

My guess an Eypc 7742 should do ~45,000 H/s and with a dual 7742 box hitting ~90,000 H/s mining Monero after the fork.:eek:

That is a rough guess based on my 3700x by calculating the speed per thread at the slower speed of a 7742 and multiplying by the huge number of threads it has.

Should be interesting.
 
Last edited:

Klee

Well-Known Member
Jun 2, 2016
1,289
396
83
This is the persons github who created the RandomX algo.

His github:
tevador/RandomX

Releases: I used the binarys since they work fine.
tevador/RandomX

Some basic info plus a link to a randomx mineable coin in the link below. The posting is a bit unorganized but it will give you the basics. I have not tried solo mining only running the benchmark at first then only running the beta version of xmrig now.

An Introdction to RandomX, Benchmarking tool and pool/solo mining

Link to the latest xmrig version 2.99.5 beta.
xmrig/xmrig

The beta xmrig works fine with my dual Xeon pc but is buggy with my Ryzen 3700x in Windows 10 and Ubuntu 18.04.2.

But it runs long enough on AMD cpu's to give you an idea what to expect with the new algo.
 
Last edited:

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
@MiniKnight about 1900

@Klee Using an AMD EPYC 7702P

Code:
$ ./randomx-benchmark --mine --jit --largepages --threads 127 --init 64 --nonces 800000
RandomX benchmark v1.0.4
 - full memory mode (2080 MiB)
 - JIT compiled mode
 - hardware AES mode
 - small pages mode
Initializing (64 threads) ...
Memory initialized in 1.32075 s
Initializing 127 virtual machine(s) ...
Running benchmark (800000 nonces) ...
Calculated result: 81751c8a378b4a8716599822d4d33e97573307e5acfad1dcf3528766c51942f4
Performance: 24401.8 hashes per second
 

Klee

Well-Known Member
Jun 2, 2016
1,289
396
83
Oh and just for fun the profit of a dual 7742 box if it can acually do 90,000 H/s at the current rate of the network hash and difficulty.

This is just a wild ass guess since I do not know what the change will be with the cpu friendly algo randomx.
 

Attachments

Klee

Well-Known Member
Jun 2, 2016
1,289
396
83
The benchmark is kind of inconsistent, that's why I use the beta xmrig for testing. I'm mining one of the few randomx coins,
txchange, not that I expect it to be worth anything only because its a randomx coin and only to test.
 

Klee

Well-Known Member
Jun 2, 2016
1,289
396
83
24401.8 seems too slow compared to my 3700x at ~7700 H/s, plus the 3900x does ~13,000.

Patrick there is a command you have to use with the benchmark to run one instance per numa node, I can't remember what it is tho.

Edit: here is a link to a benchmark posting site that shows some of of the benchmark commands used for various setups including some quad opteron servers.
RANDOMX BENCHMARKS FOR MONERO MINING!

The beta version of xmrig is numa aware tho not fully optimized yet.
 
Last edited:

Klee

Well-Known Member
Jun 2, 2016
1,289
396
83
Whatever the hash rates on monero with the new algo will end up being it's clear that the new AMD cpu's are MASSIVELY better than cpu mining the older algo's .

Plus its fun to play around with new hardware.;)
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
So @Klee, the default for the new CPUs is NPS=1 or a single NUMA node per socket. On the AMD EPYC 7702P that is a single NUMA node. I could try NPS=4, or potentially try splitting into four iterations.

This is probably a bad use of a $4.4K CPU, but I am excited to see how this turns out.
 
  • Like
Reactions: funkywizard

funkywizard

mmm.... bandwidth.
Jan 15, 2017
848
402
63
USA
ioflood.com
7302 / 7302P would be interesting based on being under $1000 with 128mb cache.

However, with only 16 cores and 32 threads, that might be insufficient for maximum performance as it gives you 4mb cache per thread.

Does monero mining have a mode that uses twice the normal cache per thread? I seem to recall that it did, but not totally sure.
 
Last edited:

funkywizard

mmm.... bandwidth.
Jan 15, 2017
848
402
63
USA
ioflood.com
7742 with the old xmrig container was ~5400 per CPU
7702p was ~5050
Sounds in the right ballpark. Dual 2680v2, if memory serves, was something like 1200h/s, with 50mb total cache between the two cpus.

With 256mb cache, you'd hope for something on the order of 5x the performance -- 6000 h/s, ignoring differences in IPC and clockspeed.

So a little lower than I'd hoped but roughly in the right ballpark.
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
EPYC 7402P is running at about 2450H/s and uses about 150W more than at idle doing so.
 
  • Like
Reactions: funkywizard

jims2321

Active Member
Jul 7, 2013
184
44
28
My Numbers on 3900x 3600 CL16 with tuned memory (Ryzen Dram Calculator) and a mild overclock on the cpu (4Ghz) is over 13K

Imgur

Running on my Dell R815 (6378 w/126GB) and this cli " seq 0 7 | xargs -P 0 -I node numactl -N node ./randomx-benchmark --mine --largePages --jit --nonces 1000000 --init 8 --threads 7"

returns

Initializing 7 virtual machine(s) ...
Running benchmark (1000000 nonces) ...
Calculated result: 36d7bb6b4ca582590247e7f3b86d08ae406c0636a09339c9e43175a4ed7c654d
Performance: 1276.38 hashes per second
Calculated result: 36d7bb6b4ca582590247e7f3b86d08ae406c0636a09339c9e43175a4ed7c654d
Performance: 1276.13 hashes per second
Calculated result: 36d7bb6b4ca582590247e7f3b86d08ae406c0636a09339c9e43175a4ed7c654d
Performance: 1276.05 hashes per second
Calculated result: 36d7bb6b4ca582590247e7f3b86d08ae406c0636a09339c9e43175a4ed7c654d
Performance: 1274.22 hashes per second
Calculated result: 36d7bb6b4ca582590247e7f3b86d08ae406c0636a09339c9e43175a4ed7c654d
Performance: 1272 hashes per second
Calculated result: 36d7bb6b4ca582590247e7f3b86d08ae406c0636a09339c9e43175a4ed7c654d
Performance: 1271.49 hashes per second
Calculated result: 36d7bb6b4ca582590247e7f3b86d08ae406c0636a09339c9e43175a4ed7c654d
Performance: 1271.45 hashes per second
Calculated result: 36d7bb6b4ca582590247e7f3b86d08ae406c0636a09339c9e43175a4ed7c654d
Performance: 1260.65 hashes per second

= 10,178.37 h/s @ 630w Per iDrac monitoring
 

Klee

Well-Known Member
Jun 2, 2016
1,289
396
83
As soon as the 16 core Ryzen 9 hits in September i'm getting one. ;)

So 13,000 @ 24 threads puts the 3950x Ryzen over 17,000.
 

LuckyX2

New Member
Aug 15, 2019
2
0
1
7402P is a poor fit for RandomX since it requires 2MB of L3 per thread and the 7402P is 96t with 128MB of L3.
The 7742 is a perfect fit with 256MB and 128t. The 32c/64t models with 128MB L3 are also a perfect fit.

The recently leaked 32c/64t Threadripper with 128MB of cache will likely be the best option if quad channel memory doesn't bottleneck it. High core count, high clocks (presumably ~4.2GHz all core) and the exact optimal amount of L3.

FYI here's a 7742 getting 38kh/s. I believe there are still some optimizations to be made.
AMD EPYC 7742 RandomX Hashrate : MoneroMining

Patrick, thanks for your testing. I think you may be having the same issue as above. I see in your output it says "
- small pages mode" even though you included the large pages flag. You may see an improvement enabling huge pages in the OS.
Mathkamy goes over setting that in his comment here:
Share your RandomX benchmarks! : Monero
 

Klee

Well-Known Member
Jun 2, 2016
1,289
396
83
On the current Monero algo my Ryzen 3700x mines at ~952 H/s on all 16 threads with Xmrig 3.0.0. in Windows 10.

Xmrig 3.0 crashes mining RandomX on both Windows 10 and Ubuntu 18.04.2 still, hard crash that resets the pc.

So i'm running the old algo just to test to see if its a issue with xmrig 3.0 or xmrig 3.0 and randomx.

HWMonitor shows ~60 w and 44c cpu temp.

Definitely the Ryzen runs MUCH cooler than the dual Xeon box, putting my hand on the rear fan exhaust fan it is barely warm.

I think for cpu mining the the Ryzen cpu's are the best option as far as price/performance/power usage.
 
Last edited:

jims2321

Active Member
Jul 7, 2013
184
44
28
I can test it out on my 3900x tomorrow when I get into my office. I am running Ubuntu 18.04.3 LTS on it.