STH AEON xmrig 1MB L3 cache Docker miner testing

Joel

Active Member
Jan 30, 2015
816
166
43
39
Also one windows 7 pc running my GTX 1060 p106 dedicated mining gpu's to test them, still need risers and decide on a power supply to run all 6 of the mining gpu's in my mining case.
I have a $60 Lenovo D30 1120w PSU that I've just installed in a Rosewill chassis; waiting on the install kit and some PCIe power expanders but the intent is to eventually house 7x 1060s plus a 2680v2. Once I confirm it's working I'll let you know. :)

https://forums.servethehome.com/index.php?threads/lenovo-1100w-psu.18163/
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,051
5,042
113
Added Xeon Gold 5119T results... perhaps the fastest 85W TDP CPU there is.
 

Joel

Active Member
Jan 30, 2015
816
166
43
39

Patrick

Administrator
Staff member
Dec 21, 2010
12,051
5,042
113
Added Intel Xeon D-2183IT 16 core numbers. Very impressive results. Almost 2x the Aeon AV1 hashrate of the Xeon D-1587. Then again, it is using over 140W to achieve that number.

Just making a note for myself, cryptonight I seem to be maxing out at around 766H/s using 16 threads.
 
Last edited:

MiniKnight

Well-Known Member
Mar 30, 2012
3,001
910
113
NYC
Does anyone have another Skylake system handy? Are they all like that with such a big Monero to Aeon AV1 delta? That's more like 4x than 3x.
 

Joel

Active Member
Jan 30, 2015
816
166
43
39
Does anyone have another Skylake system handy? Are they all like that with such a big Monero to Aeon AV1 delta? That's more like 4x than 3x.
Even Ivy has similar deltas. 1170 XMR vs. 4100 AEON for 2x 2680v2. Not quite 4x, but close.
 

ServeTheSam

Member
Dec 10, 2017
38
14
8
Does anyone have another Skylake system handy? Are they all like that with such a big Monero to Aeon AV1 delta? That's more like 4x than 3x.
Running 16.04 LTS with hardware access in the cloud....yes. Almost 4x multiple.

I only have one Skylake build on-prem and I don’t have the chip yet.
 

Nero24

Member
Jul 26, 2017
32
5
8
44
www.planet3dnw.de
There is another CPU affected by this circumstance; former AMD architectures like Kaveri, Piledriver or Bulldozer have 2 MB L2 per module. So they were perfect for Monero. But the successors Carrizo and Bristol Ridge only have 1 MB L2 per module. Unable to use this CPU for Monero mining! But Aeon perfectly fits into 1 MB L2 cache if configured right. :D
 

IamSpartacus

Well-Known Member
Mar 14, 2016
2,321
562
113
Has anyone tested Threadripper 1950x mining Aeon? I'm getting about 2.3KH/s using 16 threads right now but wondering if that's the right thread count. I tried going higher than 16 threads and my hashrate dropped.
 

Nero24

Member
Jul 26, 2017
32
5
8
44
www.planet3dnw.de
Has anyone tested Threadripper 1950x mining Aeon? I'm getting about 2.3KH/s using 16 threads right now but wondering if that's the right thread count. I tried going higher than 16 threads and my hashrate dropped.
It depends on the miner's config. The regular scratchpad size for Aeon is 1 MB per thread. Since Threadripper has 32 MB L3 cache and 16 cores plus SMT, you can run 32 threads and use the whole L3 cache and all compute ressources.

But if the miner uses double hash, the scratchpad size is 2 MB per thread. In this case 16 threads are perfect to not flood the L3.
 

IamSpartacus

Well-Known Member
Mar 14, 2016
2,321
562
113
It depends on the miner's config. The regular scratchpad size for Aeon is 1 MB per thread. Since Threadripper has 32 MB L3 cache and 16 cores plus SMT, you can run 32 threads and use the whole L3 cache and all compute ressources.

But if the miner uses double hash, the scratchpad size is 2 MB per thread. In this case 16 threads are perfect to not flood the L3.
I'm using the STH docker image so I guess that is using double hash.
 

nsiderTalon

New Member
Jan 7, 2018
5
2
3
29
I'm currently getting 5053.2 H/s with AV2 and 32 threads, and 4819.7 H/s with AV1 and 16 threads. Remember to set "sudo sysctl vm.nr_hugepages=128" in your bash script, I was getting around 2.3 KH/s before doing this.
 

IamSpartacus

Well-Known Member
Mar 14, 2016
2,321
562
113
AV1 and 32 threads for max performance or AV2 and 16 threads for max efficiency :) It's not much slower but reduces power consumption significantly :D
I see. Thank you.


I'm currently getting 5053.2 H/s with AV2 and 32 threads, and 4819.7 H/s with AV1 and 16 threads. Remember to set "sudo sysctl vm.nr_hugepages=128" in your bash script, I was getting around 2.3 KH/s before doing this.
I'm not running a bash script, I'm running the servethehome/universal_cryptonight:latest docker image. That's a major difference. I wonder if those parameters can be set inside the container?