Dedicated Open Compute miner build- Part 2.The RandomX Edition.

Klee

Well-Known Member
Jun 2, 2016
1,285
393
83
did some tweaking on my Dell R815. disable HT assist to free up 2MB of ram per core. And installed TurionPowerControl (that was a bitch). reran the randomX benchmark and now I am up to 11253 h/s on the box. but the power draw jumps to 815w

11253 h/s / 815w = 13.807 h/w

At that h/w level, not sure it will be worth it mining on my Dell except during cold weather so I can head my basement!

I plan to transition slowly to Ryzen 36XX boxes, Ryzen's use a lot less power to run.
 

jims2321

Active Member
Jul 7, 2013
185
44
28
Absolutely the Ryzen 3000 series (3600 and upwards) are going to be your best bang for the buck. somewhere around 40h/w or higher if the benchmarks are valid. As I mention I am primarily running my servers for the heat generation in the basement as the weather get cold. I will more than likely turn them off come late spring.
 

Raice

Member
Jul 12, 2017
43
6
8
38
Guangzhou, China
HP DL360 Gen8 2x2660v2 (Static High Performance, Max Performance), 128Gb RAM, Quad-channel (4*16GB DIMM each CPU)

Oracle Linux 7.7 4.14.35-1902.4.8.el7uek.x86_64

Keep in mind that this benchmark is not NUMA-friendly

Code:
[root@localhost build]# seq 0 1| xargs -P 0 -I node numactl -N node ./randomx-benchmark --mine --jit --largePages --threads 10 --init 10 --nonces 2000000

RandomX benchmark v1.1.0
- full memory mode (2080 MiB)
- JIT compiled mode
- hardware AES mode
- large pages mode
Initializing (10 threads) ...
RandomX benchmark v1.1.0
- full memory mode (2080 MiB)
- JIT compiled mode
- hardware AES mode
- large pages mode
Initializing (10 threads) ...
Memory initialized in 4.07401 s
Initializing 10 virtual machine(s) ...
Running benchmark (2000000 nonces) ...
Memory initialized in 4.18038 s
Initializing 10 virtual machine(s) ...
Running benchmark (2000000 nonces) ...
Calculated result: 2d29fa19e233225c7ac29e4be0d84d6cffeede27d52916e645590b0ec03a6621
Performance: 3493.21 hashes per second
Calculated result: 2d29fa19e233225c7ac29e4be0d84d6cffeede27d52916e645590b0ec03a6621
Performance: 3439.8 hashes per second
Total hashrate is 6933 h/s, average power is 304W, 23h/W
 

Raice

Member
Jul 12, 2017
43
6
8
38
Guangzhou, China
We also have about 150 OCP nodes with 2x2660v2 each, mining monero now. I've done some tests with dual-channel memory. HR was around 6000h/s, did't measure power though. Want to test it again later with proper quad channel memory and power measurements using DCMI interface
 

jims2321

Active Member
Jul 7, 2013
185
44
28
Just got my R710 back online. Running the 1.10 version of the randomx benchmark I am getting arond 2431 h/s.

miner@dellr710:~/RandomX$ seq 0 1| xargs -P 0 -I node numactl -N node ./randomx-benchmark --mine --largePages --jit --nonces 100000 --init 6 --threads 6
RandomX benchmark v1.1.0
- full memory mode (2080 MiB)
- JIT compiled mode
- hardware AES mode
- large pages mode
Initializing (6 threads) ...
RandomX benchmark v1.1.0
- full memory mode (2080 MiB)
- JIT compiled mode
- hardware AES mode
- large pages mode
Initializing (6 threads) ...
Memory initialized in 7.20128 s
Initializing 6 virtual machine(s) ...
Running benchmark (100000 nonces) ...
Memory initialized in 7.21398 s
Initializing 6 virtual machine(s) ...
Running benchmark (100000 nonces) ...
Calculated result: f35a269c763e9ef175708a5099c2c30ea25f6cb26fe79b0049d96846f5fbfd47
Performance: 1219.11 hashes per second
Calculated result: f35a269c763e9ef175708a5099c2c30ea25f6cb26fe79b0049d96846f5fbfd47
Performance: 1212.28 hashes per second
 

Klee

Well-Known Member
Jun 2, 2016
1,285
393
83
We also have about 150 OCP nodes with 2x2660v2 each, mining monero now. I've done some tests with dual-channel memory. HR was around 6000h/s, did't measure power though. Want to test it again later with proper quad channel memory and power measurements using DCMI interface
I would hate to have that electricity bill on 150 nodes. I have six and have only test mined with them for a day or two at a time in the last year and a half.

I have them in my dining room and at %100 they do put out some heat except with the 2603 v1's installed so during the summer they are mostly off to save on the electric bill.
 

Klee

Well-Known Member
Jun 2, 2016
1,285
393
83
Testing the nodes, not the latest version of xmrig.

 
Last edited:
  • Like
Reactions: Patrick

Klee

Well-Known Member
Jun 2, 2016
1,285
393
83
Tried xmrig 4.0.1- beta on my Ryzen 3700x pc with a single RX580.

The opencl mining did not work, I have AMD-GPU-Pro and legacy opencl installed and working.

Version 3.1.3 mined a tiny bit faster than 4.0.1-beta with just the cpu cores.
 

hifijames

Member
Dec 26, 2017
32
7
8
60
Tried xmrig 4.0.1- beta on my Ryzen 3700x pc with a single RX580.

The opencl mining did not work, I have AMD-GPU-Pro and legacy opencl installed and working.

Version 3.1.3 mined a tiny bit faster than 4.0.1-beta with just the cpu cores.

How does rx/loki hashrate translate to cn/r?
 

Klee

Well-Known Member
Jun 2, 2016
1,285
393
83
How does rx/loki hashrate translate to cn/r?

That's really a apple vs orange comparison.

Cpu's mine MUCH faster on randomx vs cn/r and gpu's mine probably about 2/3's of the rate on randomx.

At first I thought that there would be a massive drop in the network hash rate because the gpu's would no longer be dominating or even competitive and they all would leave.

Now I'm not so sure, there is so many factors involved that affects the network hash rate.

I think at first there will be a huge drop just because of the change but that will probably swing back and get higher simply because everyone that mines has the hardware to mine with cpu's plus the bot nets i'm sure will be updated.

Also quite a few will update their gpu rigs because thats what they have the most compute power invested in and i'm sure they will switch to the new algo and mine at least for a while just to see before they switch to another algo if it proves to be unprofitable for them.

So this is my forcast:
Short term a big drop in network hash rate.
Long term will increase to probably halfway between the current hash rate and the peak when the asics were hashing on the network.
 
Last edited:

Klee

Well-Known Member
Jun 2, 2016
1,285
393
83
UPDATE: Network upgrade no longer in mid October, now its November ~30th.

From the reddit Monero page:

Preliminary information thread regarding the scheduled protocol upgrade of November 30 : Monero

"General information about the scheduled network upgrade
Approximately the 30th of November there will be a scheduled network upgrade on the Monero network. To be sufficiently prepared, a user, service, merchant, pool operator, or exchange should run CLI v0.15 or GUI v0.15. The scheduled network upgrade introduces a few major changes. First and foremost, a new long-term Proof-of-Work algorithm, namely RandomX, will be introduced. Miners therefore ought to upgrade their mining software as well. Second, long payment IDs will be phased out in order to improve privacy and user experience as well as reduce support work for services and exchanges. Third, transactions will now require at least two outputs. Fourth, the ten block (approximately twenty minutes) lock time for incoming transactions will be enforced on the protocol level. Both these changes will improve privacy for the user as well as for the whole network.

A few notes:

 

ssian

New Member
Sep 29, 2019
1
0
1
What ram timings are best for RandomX? Reading that timing and latency can have a noticeable difference on hashing performance.
 

ari2asem

Active Member
Dec 26, 2018
536
85
28
The Netherlands, Groningen
ok, guys...maybe stupid question....I am trying to run randomx-benchmark on windows 10 64bit from PowerShell, nothing happens. I downloaded the compiled version (for windows) from GitHub its self.

the folder I see only BENCHMARK.EXE....tried ./benchmark, ./benchmark.exe, .\benchmark, .\benchmark.exe....all of this with --help or -help....nothing happens....even tried PowerShell as admin....and I have dotNET version 4.7.2 on windows 10....any help??
 

Raice

Member
Jul 12, 2017
43
6
8
38
Guangzhou, China
ok, guys...maybe stupid question....I am trying to run randomx-benchmark on windows 10 64bit from PowerShell, nothing happens. I downloaded the compiled version (for windows) from GitHub its self.

the folder I see only BENCHMARK.EXE....tried ./benchmark, ./benchmark.exe, .\benchmark, .\benchmark.exe....all of this with --help or -help....nothing happens....even tried PowerShell as admin....and I have dotNET version 4.7.2 on windows 10....any help??
you need to pass some arguments to it. RTFM.
 

JSchuricht

Active Member
Apr 4, 2011
164
53
28
Single Quanta node with dual 2650L V2 is doing 31.65 hash per watt with XMRIG 4.4.0 with dual channel 1333 RAM. Quad channel is an extra 20h/s and 9 watts. The node uses 20 watts more with random x vs Monero/cryptonight. Not keeping my hopes up on seeing similar numbers with the Wiwynn nodes running dual 2660 V1's whenever I get time to test them.
 

ari2asem

Active Member
Dec 26, 2018
536
85
28
The Netherlands, Groningen
does anyone knows how to change this value in windows 10?
Code:
vm.nr_hugepages = 2560
thus not asking how to enable large pages in windows 10, because it is already enabled. i want to change just that number...how to do it windows 10?
 

bash

Member
Dec 14, 2015
91
21
8
38
scottsdale
Now that the network update is here I will be doing some testing on the following systems this weekend.

EPYC 7401P 24-Core
Ryzen 3900
Dual E5-2697 v2

Will update with results and headaches.
 
  • Like
Reactions: Klee