Monero Mining Performance

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

Marsh

Moderator
May 12, 2013
2,644
1,496
113
dual E5-2666 v3 QFSC

SuperMicro X10DAX , dual E5-2666 v3 QFSC , 8 x 16gb ram , L3 cache 25MB , 10 cores , 10 Threads
dual E5-2666 v3 ( 20 active cores, 20 threads ) , 1094 H/s 265w = 4.12 H/s per watt

xmr-stak-cpu config.txt: 4 threads = "low_power_mode" : true to use 4MB per thread L3 cache

Model name: Genuine Intel(R) CPU @ 2.90GHz
Stepping: 2
CPU MHz: 3099.941
CPU max MHz: 3300.0000
CPU min MHz: 1200.0000
 

modder man

Active Member
Jan 19, 2015
657
84
28
32
Do the numbers at the top of this post still hold true? Trying to determine what I should expect our of my RX470's at this point I was aiming for ~800H/s from each card. Thread says 550 h/s and I am already getting more than that.
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
Do the numbers at the top of this post still hold true? Trying to determine what I should expect our of my RX470's at this point I was aiming for ~800H/s from each card. Thread says 550 h/s and I am already getting more than that.
CPU numbers are good from a relative standpoint. On the GPU side, I think things have gotten much better.
 

Benten93

Member
Nov 16, 2015
48
7
8
Do the numbers at the top of this post still hold true? Trying to determine what I should expect our of my RX470's at this point I was aiming for ~800H/s from each card. Thread says 550 h/s and I am already getting more than that.
I get about 725H/s on my RX480s, so 800h/swill Be a Little out of range :)
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
Added EPYC 7351P preliminary numbers. Fairly awesome 177W results.

Almost all of the EPYC 7601 XMR mining performance since it has 64MB L3 as well, but at only around $800.
 

Klee

Well-Known Member
Jun 2, 2016
1,289
396
83
@Klee what is that?

It's xmr-stak-cpu-dev on my dual E5-2667 V3 es cpu's running in "low power mode"

It runs faster and cooler on 16 threads than xmr-stak-cpu on 32 threads.:D

Plus the desktop is still SUPER responsive because I still have 16 threads running at 2.9 GHz or greater to run what ever I want.

I just mine 24/7 with it and you can't even notice it using the desktop, I run all my regular programs with zero issues.




"cpu_threads_conf" :
[
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 0 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 1 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 2 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 3 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 4 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 5 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 6 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 7 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 8 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 9 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 10 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 11 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 12 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 13 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 14 },
{ "low_power_mode" : true, "no_prefetch" : true, "affine_to_cpu" : 15 },"
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
Interesting Intel Atom C3958
[2017-09-16 04:35:04] accepted: 1/1 (100.00%), 283.66 H/s at diff 100002 (yay!!!)
[2017-09-16 04:35:27] Stratum detected new block
[2017-09-16 04:39:58] accepted: 2/2 (100.00%), 284.98 H/s at diff 100002 (yay!!!)
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
Added 1P EPYC 7301 - very close to 2P 7251 results but with lower power consumption. ~7.5H/s/W
 

funkywizard

mmm.... bandwidth.
Jan 15, 2017
848
402
63
USA
ioflood.com
Have some confirmed stats from my most efficient server thus far:

floureon energy meter model ts-1500

supermicro motherboard:
Mobo model: X9DRW-3F PCB Rev 1.02 BIOS Version 3.2
Supermicro CSE815TQ-R700WB
removed existing power supplies. added 1x PWS-504P-1R (500 watt 80+ platinum)

dual e5-2660v2 cpus
6 cores enabled in bios
8x8gb pc3l-10600r
ssd boot drive

reported stats from "ipmitool sensor":

cpu1: 43c. cpu2: 47c
fans 1,2,3,4: varies from 5800-6300 rpm

reported power from "ipmitool dcmi power reading": 146w
reported power from power meter:

watts: fluctuated from 132-144w. usually below 140w
volts: 203-204v measured
amps: usually 0.695a measured (fluctuated up/down a small amount)
volt-amps: typically 142va measured

last 15 minutes: 868h/s -- this seems quite low, going to restart miner and see if it improves.

edit: restarted stak. 1 minute average 902 h/s. still below expectations.
 
  • Like
Reactions: Marsh

funkywizard

mmm.... bandwidth.
Jan 15, 2017
848
402
63
USA
ioflood.com
All cores seem to be at the full turbo of 2.6ghz. 24 threads active which is ideal for the l3 cache. Just not seeing the performance I saw when I first fired this up with these settings 4 days ago.

Either way, easily the best power use of any dual e5 I've set up thus far.
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
Put the EPYC 7601 into a 1U 1P system that is fairly barebones only the BMC + a 10Gb SFP+ port connected. DDR4-2400 to lower IF speed.

1786H/s at 179W using Wolf's in Docker. @Klee and @Marsh - first 10H/s/W CPU?
 
  • Like
Reactions: Marsh and Klee

funkywizard

mmm.... bandwidth.
Jan 15, 2017
848
402
63
USA
ioflood.com
Restarted the stak miner a couple more times until I got the results I expected. Did not change any configurations. Did not reboot server. Did not edit bios settings.

New 15 minute average performance: 970h/s.

ipmitool dcmi power reading: 145w.

so probably the same 142va measured on the power meter previously (sorry, already had it disconnected at this point)
 

funkywizard

mmm.... bandwidth.
Jan 15, 2017
848
402
63
USA
ioflood.com
Put the EPYC 7601 into a 1U 1P system that is fairly barebones only the BMC + a 10Gb SFP+ port connected. DDR4-2400 to lower IF speed.

1786H/s at 179W using Wolf's in Docker. @Klee and @Marsh - first 10H/s/W CPU?
great results : )

thanks for sharing.

have two more data points:

====

Another X9DRW board with e5-2660v2
this one has 2x8gb low voltage ram
it also has one fan removed (3 fans total not 4)
pretty sure it's the same chassis as my last reported result

on the more efficient 500w platinum psu, "ipmitool dcmi power reading" shows 133w and 989h/s (1 minute average)

from recent tests, at least with this motherboard, the ipmitool readings report very close to the "volts * amps" readings of my 208v power meter.

i.e. they seem to be spot on if you care about your circuit amps usage, but strictly speaking seem to report a bit higher in watts than my meter shows.

anyway, I'm happy to accept (at least for this chassis + motherboard) the ipmitool watt readings are not erroneously low.

so... 989 / 133 = 7.4 H/W pretty good

====

second system. same 500w platinum psu, same motherboard, 4 fans this time, 8x8gb lv ram, dual e5-2680v2, 6 cores per cpu enabled in bios, mining on 24 threads

power meter shows 0.81 - 0.83 A @ 205v (166-170 VA) and 0.95 - 1.00 power factor.

Power meter shows 162-165w. ipmitool reports 177w.

1184h/s. Using the ipmitool number of 177w, that's 6.7 H/w -- not bad.

this same system was at 0.88a @ 205v (180 VA), ipmitool reporting 181w, power meter reporting 172w, when using a different psu: PWS-704P-1R (750w 80+ gold)

looking just at volt-amps (the more relevant metric in the datacenter), you drop by about 12 VA going to the better PSU (more efficient and more correctly sized to the task at hand). In a datacenter this psu upgrade should pay for itself inside of a year, even if you threw the old one in the trash.
 
  • Like
Reactions: Marsh

Marsh

Moderator
May 12, 2013
2,644
1,496
113
E5 v2 H/s per watt is pretty good and H/s per dollar value is better than E5 v3.
For E5 v2 , we don't need to invest in higher cost DDR4 ram and new system board.

I like E5 v2 CPU. Just hoping price starting dropping soon.
 

funkywizard

mmm.... bandwidth.
Jan 15, 2017
848
402
63
USA
ioflood.com
the command is:

"ipmitool dcmi power reading"

works on some motherboards when the smbus cable is connected from the psu to the motherboard.

may not be accurate on all systems -- the fat twin I tested for example excludes chassis-level power usage (such as fans) from the measurement.

best to verify that a particular chassis / power supply / motherboard gives accurate numbers before relying on these. aside from the fat twin however, the numbers it gives out have been pretty close.

for logging -- probably, I haven't looked into it. the command gives an instantaneous power reading, a minimum, a maximum, an average, and a time span. I'm not sure how to reset the time span. Personally, I run the command a few times and look at the instantaneous reading, and either pick the value it stays at most of the time, or a range that it fluctuates within.