Monero Mining Performance

mantis · Dec 30, 2017

Ok, it was my mistake.. the affinity needed to be inside "" marks.

Now its running perfectly, after i affined the idle cores for Cast XMR, i now get 600-612h/s on both CPUs. Flawless victory!

dwright1542 · Dec 30, 2017

mantis said:
Ok, it was my mistake.. the affinity needed to be inside "" marks.

Now its running perfectly, after i affined the idle cores for Cast XMR, i now get 600-612h/s on both CPUs. Flawless victory!

Nice! Yeah, memory optimization is everything for Cryptonight based calcs. NUMA specifically addresses that. Same reason 5 year old HD79XX series GPU's still are viable: it's not about the actual GPU speed...it's about the memory optimizations.

voodooFX · Jan 1, 2018

Does running the docker-mining-image inside a esxi VM (ubuntu) provides "impossible" results?
I mean, is this something well known?
Because I'm getting 674.5 H/s from a 4-Core VM on a IBM M4 host with 1xE5-2670V2, which looks very high/impossible, based on the numbers in the first post..

EDIT

getting similar performance also on bare metal (E3 1220V2)

Code:

dp@titan:~$ docker logs elegant_brown | tail -n 30
[ 78%] Building CXX object CMakeFiles/xmrig.dir/src/Platform_unix.cpp.o
[ 80%] Building CXX object CMakeFiles/xmrig.dir/src/Cpu.cpp.o
[ 82%] Building C object CMakeFiles/xmrig.dir/src/crypto/c_keccak.c.o
[ 85%] Building C object CMakeFiles/xmrig.dir/src/crypto/c_groestl.c.o
[ 87%] Building C object CMakeFiles/xmrig.dir/src/crypto/c_blake256.c.o
[ 89%] Building C object CMakeFiles/xmrig.dir/src/crypto/c_jh.c.o
[ 91%] Building C object CMakeFiles/xmrig.dir/src/crypto/c_skein.c.o
[ 93%] Building CXX object CMakeFiles/xmrig.dir/src/crypto/CryptoNight.cpp.o
[ 95%] Building CXX object CMakeFiles/xmrig.dir/src/log/SysLog.cpp.o
[ 97%] Building CXX object CMakeFiles/xmrig.dir/src/api/Httpd.cpp.o
[100%] Linking CXX executable xmrig
[100%] Built target xmrig
 * VERSIONS:     XMRig/2.4.3 libuv/1.9.1 gcc/6.3.0
 * HUGE PAGES:   available, disabled
 * CPU:          Intel(R) Xeon(R) CPU E3-1220 V2 @ 3.10GHz (1) x64 AES-NI
 * CPU L2/L3:    1.0 MB/8.0 MB
 * THREADS:      4, cryptonight-lite, av=1, donate=0%
 * POOL #1:      a.mwork.io:4334
 * COMMANDS:     hashrate, pause, resume
[2018-01-01 14:22:01] use pool a.mwork.io:4334 64.71.135.158
[2018-01-01 14:22:01] new job from a.mwork.io:4334 diff 20000
[2018-01-01 14:22:44] accepted (1/0) diff 20000 (192 ms)
[2018-01-01 14:22:59] new job from a.mwork.io:4334 diff 30000
[2018-01-01 14:23:04] speed 2.5s/60s/15m 641.7 640.8 n/a H/s max: 641.8 H/s
[2018-01-01 14:23:54] accepted (2/0) diff 30000 (190 ms)
[2018-01-01 14:24:04] speed 2.5s/60s/15m 641.6 641.5 n/a H/s max: 641.8 H/s
[2018-01-01 14:24:12] accepted (3/0) diff 30000 (201 ms)
[2018-01-01 14:24:29] new job from a.mwork.io:4334 diff 45000
[2018-01-01 14:25:04] speed 2.5s/60s/15m 641.6 641.6 n/a H/s max: 641.8 H/s
[2018-01-01 14:26:04] speed 2.5s/60s/15m 641.7 641.5 n/a H/s max: 641.8 H/s

Or I'm looking at the wrong numbers?

voodooFX · Jan 1, 2018

I found another Socket 2011 MB laying around.
I'm now indecise if put in 2x E5-2670V2 or 2xE5-2687W

That would be

Sandy Bridge 16c/32 thread 3.10-3.80GHz - 300W TDP
vs
Ivy Bridge 20c/40 thread 2.50-3.10Ghz - 230W TDP

Does AEON scale better with cores or frequency?

mantis · Jan 2, 2018

voodooFX said:

Does running the docker-mining-image inside a esxi VM (ubuntu) provides "impossible" results?
I mean, is this something well known?
Because I'm getting 674.5 H/s from a 4-Core VM on a IBM M4 host with 1xE5-2670V2, which looks very high/impossible, based on the numbers in the first post..

EDIT

getting similar performance also on bare metal (E3 1220V2)

Code:

dp@titan:~$ docker logs elegant_brown | tail -n 30
[ 78%] Building CXX object CMakeFiles/xmrig.dir/src/Platform_unix.cpp.o
[ 80%] Building CXX object CMakeFiles/xmrig.dir/src/Cpu.cpp.o
[ 82%] Building C object CMakeFiles/xmrig.dir/src/crypto/c_keccak.c.o
[ 85%] Building C object CMakeFiles/xmrig.dir/src/crypto/c_groestl.c.o
[ 87%] Building C object CMakeFiles/xmrig.dir/src/crypto/c_blake256.c.o
[ 89%] Building C object CMakeFiles/xmrig.dir/src/crypto/c_jh.c.o
[ 91%] Building C object CMakeFiles/xmrig.dir/src/crypto/c_skein.c.o
[ 93%] Building CXX object CMakeFiles/xmrig.dir/src/crypto/CryptoNight.cpp.o
[ 95%] Building CXX object CMakeFiles/xmrig.dir/src/log/SysLog.cpp.o
[ 97%] Building CXX object CMakeFiles/xmrig.dir/src/api/Httpd.cpp.o
[100%] Linking CXX executable xmrig
[100%] Built target xmrig
 * VERSIONS:     XMRig/2.4.3 libuv/1.9.1 gcc/6.3.0
 * HUGE PAGES:   available, disabled
 * CPU:          Intel(R) Xeon(R) CPU E3-1220 V2 @ 3.10GHz (1) x64 AES-NI
 * CPU L2/L3:    1.0 MB/8.0 MB
 * THREADS:      4, cryptonight-lite, av=1, donate=0%
 * POOL #1:      a.mwork.io:4334
 * COMMANDS:     hashrate, pause, resume
[2018-01-01 14:22:01] use pool a.mwork.io:4334 64.71.135.158
[2018-01-01 14:22:01] new job from a.mwork.io:4334 diff 20000
[2018-01-01 14:22:44] accepted (1/0) diff 20000 (192 ms)
[2018-01-01 14:22:59] new job from a.mwork.io:4334 diff 30000
[2018-01-01 14:23:04] speed 2.5s/60s/15m 641.7 640.8 n/a H/s max: 641.8 H/s
[2018-01-01 14:23:54] accepted (2/0) diff 30000 (190 ms)
[2018-01-01 14:24:04] speed 2.5s/60s/15m 641.6 641.5 n/a H/s max: 641.8 H/s
[2018-01-01 14:24:12] accepted (3/0) diff 30000 (201 ms)
[2018-01-01 14:24:29] new job from a.mwork.io:4334 diff 45000
[2018-01-01 14:25:04] speed 2.5s/60s/15m 641.6 641.6 n/a H/s max: 641.8 H/s
[2018-01-01 14:26:04] speed 2.5s/60s/15m 641.7 641.5 n/a H/s max: 641.8 H/s

Or I'm looking at the wrong numbers?

You are mining cryptonight-lite, its about 3.5x higher hashrate than the cryptonight algo. So just divide your hash rates by 3.5 and it will be approximately comparative to the results posted in this thread.

voodooFX · Jan 2, 2018

dwright1542 · Jan 2, 2018

voodooFX said:
I found another Socket 2011 MB laying around.
I'm now indecise if put in 2x E5-2670V2 or 2xE5-2687W

That would be

Sandy Bridge 16c/32 thread 3.10-3.80GHz - 300W TDP
vs
Ivy Bridge 20c/40 thread 2.50-3.10Ghz - 230W TDP

Does AEON scale better with cores or frequency?

Both?

Actually, the amount of L3 cache available on each CPU. A quick looks shows the 2670V2 has 25MB vs 20MB on the 2687. 2670 should be able to run 5 more full power aeon threads.

gigatexal · Jan 2, 2018

but AEON seems to also benefit from the IPC gains on new archs so the differences could be a wash?

mantis · Jan 4, 2018

So, any news about this Intel vulnerability... will the patches affect MY HASHRATES?

JustinH · Jan 4, 2018

For CPU hashing - yes. I did a quick test today and Saw about a 15% reduction in hash rates. Not going to upgrade my kernels anytime soon ( the machines are not exposed to the Net and I have my wallet on a different machine so little risk there)

Sent from my iPhone using Tapatalk

gigatexal · Jan 4, 2018

What kernel version got upgraded? I’ll be sure to avoid that

JustinH · Jan 4, 2018

It will be in the upcoming 4.15 release and I believe most of the distros will back port it.

I tested on Fedora with the patch backported to 4.14.

Sent from my iPhone using Tapatalk

gigatexal · Jan 4, 2018

Good to know.

Sent from my iPhone using Tapatalk

mantis · Jan 5, 2018

15% reduction, that is a lot... i'm running windows, probably need to find a way to block auto updating completely.

alex_stief · Jan 5, 2018

From what I understood so far how the patch works 15% performance drop seems pretty high. I would not have expected that for this kind of application

nfsden · Jan 5, 2018

Say hello, to new botnet. + almost 80Mh to monero nethash just in past few hours (I guess they are from botnets on js webminers)
Monero becomes abotnet coin.

Joel · Jan 5, 2018

alex_stief said:
From what I understood so far how the patch works 15% performance drop seems pretty high. I would not have expected that for this kind of application

I suspect this will be a temporary problem until the devs can optimize for the patch. Sucks nonetheless.

keybored · Jan 5, 2018

Wow... Network is at 630MH/s today. IIRC just 2 weeks ago it was at 500MH/s. At this rate it'll be at 1GH/s by Spring if not earlier...

nfsden · Jan 5, 2018

keybored said:
Wow... Network is at 630MH/s today. IIRC just 2 weeks ago it was at 500MH/s. At this rate it'll be at 1GH/s by Spring if not earlier...

it will be even bigger if developers don't take action to stop botnets.
Roughly 300Mh sitting on private pools (mainly they are botnets based on coinhive javascript web miner software)
Monero Mining Power Distribution

Joel · Jan 5, 2018

Devs probably don't care. After all, they're earning .1% of the botnets income. Unless you're talking about the devs of Monero itself.

Monero Mining Performance

Member

Active Member

Active Member

Active Member

Member

Active Member

Active Member

I'm here to learn

Member

Active Member

I'm here to learn

Active Member

I'm here to learn

Member

Well-Known Member

Member

Active Member

Active Member

Member

Active Member