Monero Mining Performance
- Thread starter Patrick
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
I think my question got list in the rapid conversations yesterday, so bumping this over to page 10 to see if anyone can provide me some pointers.
My answer:
1. Setup a Ubuntu VM on each - give it a low priority so it can use cores but low priority on them (e.g. schedule everything else ahead of these VMs)
2. Install Docker
Logout/ back in.
3. Create a Docker swarm
4. Install Portainer.io on the manager swarm node:
5. Navigate to the manager swarm node's IP address:9000 with a web brower, login and go to services.
6. In Portainer services Add Service:
Image: servethehome/monero_cpu_minergate
Scheduling mode: global
Environment variables (add one here): name=username value=example@example.com (your minergate username)
That will allow you to run a service on each of the swarm nodes that will mine using leftover CPU time and have a GUI.
1. Setup a Ubuntu VM on each - give it a low priority so it can use cores but low priority on them (e.g. schedule everything else ahead of these VMs)
2. Install Docker
Code:
wget -qO- https://get.docker.com/ | sh
sudo usermod -aG docker $your_user3. Create a Docker swarm
Code:
docker swarm init --advertise-addr (LAN IP address)
Code:
docker run -d -p 9000:9000 -v "/var/run/docker.sock:/var/run/docker.sock" portainer/portainer6. In Portainer services Add Service:
Image: servethehome/monero_cpu_minergate
Scheduling mode: global
Environment variables (add one here): name=username value=example@example.com (your minergate username)
That will allow you to run a service on each of the swarm nodes that will mine using leftover CPU time and have a GUI.
One lone CPU is E5-2650 v3 QS , the other lone E5-2650 v3 CPU is production stepping.
I have not attempt to see if I could mixed the two cpu together.
Calif power $$$ cost is based on usage,
baseline usage tier 1 is $0.183 per kwh ,
101%-200% tier 2 is $0.243 per kwh
over 200% tier 3 is $0.401 per kwh
I am breaking even (excluding equipment cost ) if my power usage stay below tier 2
Losing money if the extra power usage by the mining rigs goes above tier 2 cost.
When the weather gets hot, I'll need to stop mining or cut back to 1 or 2 rigs to conserve power.
@Marsh - may be worth trying both of them in the same node. I think @MiniKnight is right that if they do work, you will save a good chunk of power.
I guess heating in the CA "winter" is saved with mining at home.
BTW I setup the 2nd 2U 4N. 3832H/s in 617w. Actually profitable excluding hardware costs but the hardware payback is like 13 months.
I guess heating in the CA "winter" is saved with mining at home.
BTW I setup the 2nd 2U 4N. 3832H/s in 617w. Actually profitable excluding hardware costs but the hardware payback is like 13 months.
My ASUS 4nodes with 7 x E5-2650 v3 QS CPU produced about 2.7KH/s using 600w.
Until I see this month's PG&E electricity bill (due Feb 11 ), ASUS 4 nodes are mining 12-14 hours per day,
The two X99 single cpu rigs are running 24hours .
I hate to have my power usage cross over to tier 3 $.40 kwh.
Currently , I want to get a sense what is the mining rig power usage and profit before spending more $$$.
My estimate $$ profit is about $1 per 12hours ( excluding hardware cost ).
Until I see this month's PG&E electricity bill (due Feb 11 ), ASUS 4 nodes are mining 12-14 hours per day,
The two X99 single cpu rigs are running 24hours .
I hate to have my power usage cross over to tier 3 $.40 kwh.
Currently , I want to get a sense what is the mining rig power usage and profit before spending more $$$.
My estimate $$ profit is about $1 per 12hours ( excluding hardware cost ).
I feel the need to join in this insanity...Though it looks like the bulk of my hardware is older than would be useful. Most of my current hardware is 1366. How are you guys setting up e5 v3 nodes so cheap that payback is 13 months?
Patrick , we love to see the math .
My math $$$:
ASUS 4 nodes hardware cost:
$550 + $50 tax = $600 for the ASUS 4 nodes ( almost firesale price )
$110 x 7 =$770 E5-2650 v3 QS ( 1 single ebay listing for 7 cpu with BO , happened the night before ASUS sale ending, I considered luck )
$60 per 16gb ram stick x 7 = $420 (min ram to run the miner software , Ubuntu OS )
$5 x 4 =$20 16gb M2 SSD
ASUS Hardware cost = $1,810
If Monero mining produce $2 per day net , then 3 years payback for the ASUS 4 nodes.
One way to see faster payback is to hope XMR $$$ shot up like a hot IPO stock.
BTW, MSI X99 and Asrock X99 board are like $150 each from Ebay.
Although, the original intent was to build a workstation.
I am doing this for fun, I keep telling myself .
My math $$$:
ASUS 4 nodes hardware cost:
$550 + $50 tax = $600 for the ASUS 4 nodes ( almost firesale price )
$110 x 7 =$770 E5-2650 v3 QS ( 1 single ebay listing for 7 cpu with BO , happened the night before ASUS sale ending, I considered luck )
$60 per 16gb ram stick x 7 = $420 (min ram to run the miner software , Ubuntu OS )
$5 x 4 =$20 16gb M2 SSD
ASUS Hardware cost = $1,810
If Monero mining produce $2 per day net , then 3 years payback for the ASUS 4 nodes.
One way to see faster payback is to hope XMR $$$ shot up like a hot IPO stock.
BTW, MSI X99 and Asrock X99 board are like $150 each from Ebay.
Although, the original intent was to build a workstation.
I am doing this for fun, I keep telling myself .
My differences are that I am at 3.8kH/s. I also am using 1x 8GB per CPU to get that figure. Otherwise similar pricing to what you have. The other side is that I am using "free" excess data center power.
I bought 10x of those 16GB m.2's which has been awesome for these chassis.
I bought 10x of those 16GB m.2's which has been awesome for these chassis.
So I was getting really annoyed why I was getting such low numbers with 2x E5-2670's at 669H/s on Server 2016. Then I realized Windows goes by Nodes or NUMA (Says Node in Task Manager), I already knew it went by NUMA, but hell.. Windows has to be different. So, gotta what I did before, I was getting my best results by using 24 threads, which didn't make much sense since the math doesn't work out with L3 cache.
Then I looked, and realized Windows does core assigning totally different from Linux, which I should've figured out. So manually, I was launching the program and counting how many threads I could get on each node per Windows how it was assigning the threads. You can't just go 0-24 or w/e you want in the assigning of the config like I did
tl;dr, some quick assigning and testing, dirty math. I'm now at 20 threads (10 threads per NODE(CPU)) and I'm getting 820H/s on Windows, which is insane. I thought this was only a linux thing
http://i.imgur.com/Iycw9vd.png
Once I run it longer, it runs even higher: http://i.imgur.com/ExnAaP3.png
Then I looked, and realized Windows does core assigning totally different from Linux, which I should've figured out. So manually, I was launching the program and counting how many threads I could get on each node per Windows how it was assigning the threads. You can't just go 0-24 or w/e you want in the assigning of the config like I did
tl;dr, some quick assigning and testing, dirty math. I'm now at 20 threads (10 threads per NODE(CPU)) and I'm getting 820H/s on Windows, which is insane. I thought this was only a linux thing
http://i.imgur.com/Iycw9vd.png
Once I run it longer, it runs even higher: http://i.imgur.com/ExnAaP3.png
Last edited:
OK 10k pulls in a week... who put this on a big cluster? Monero CPU Mining in Docker
Also, just checked my other chassis @Marsh
1284w - 5083H/s
Interestingly, that is using 3 of 4 nodes as E5 V4's and one is V3. That is also using 8 DIMMs/ node. One is a E5-2699 V4.
The E7 4P machine is 2.6KH/s and 1kW.
So the net conclusion here is that the higher core count CPUs, even as V4's are not necessarily better for mining.
One other project I am working on is adding Power as a label for each node in Docker/ Rancher. That way I know how much power each node uses
Also, just checked my other chassis @Marsh
1284w - 5083H/s
Interestingly, that is using 3 of 4 nodes as E5 V4's and one is V3. That is also using 8 DIMMs/ node. One is a E5-2699 V4.
The E7 4P machine is 2.6KH/s and 1kW.
So the net conclusion here is that the higher core count CPUs, even as V4's are not necessarily better for mining.
One other project I am working on is adding Power as a label for each node in Docker/ Rancher. That way I know how much power each node uses
Last edited:
I may be slightly drinking due to my results, but maybe its because the L3 cache of scaling with core count. (Not counting AVX vs AVX 2.0)OK 10k pulls in a week... who put this on a big cluster? Monero CPU Mining in Docker
Also, just checked my other chassis @Marsh
1284w - 5083H/s
Interestingly, that is using 3 of 4 nodes as E5 V4's and one is V3. That is also using 8 DIMMs/ node. One is a E5-2699 V4.
The E7 4P machine is 2.6KH/s and 1kW.
So the net conclusion here is that the higher core count CPUs, even as V4's are not necessarily better for mining.
One other project I am working on is adding Power as a label for each node in Docker/ Rancher. That way I know how much power each node uses
If it scaled with core count as I am no engineer, I'm basing my numbers off of my E5-2670 V1's
8 core 20MB
16 Core 40MB
24 Core 60MB
and so on, maybe that extra 5MB of L3 cache would be running a lot better and use the same amount of power.
I'll just say this, these are drunk thoughts.
So it turns out I was sitting on 10kh/s of capacity doing nothing when idle... I fixed that.
now i tried the docker image, but its literally the first time I even installed docker. I didn't see any stats output and I didn't know how to control niceness etc properly so I shut it down and compiled wolf's. I am getting only 1% better going from # cores to cache /2MB (over provision cores), so I'm sticking to core count because I would rather not eat extra cache.
That being said all this talk about sysctl -w vm.nr_hugepages=# did absolutely nothing for me. recompiled and all. I would love to get a 10% boost that allegedly comes from it but I dunno where I messed up.
Patrick, can I get stats for docker performance and can I made sure it is running true idle only (I set wolf's to nice -19 and it seems to have minimal impact on things once I disabled cpu usergroups)?
lastly, it seems people report much better performance from the windows client by claymore. Has anyone tested this. I tested claymore vs gui miner on windows and yeah claymore was like +45%, but then a 5% mining tax.
Also, my r9 290s are putting out 820 h/s each, which makes them pretty good as long as I don't run during weekday daytime (variable rate electricity (12 hours profit, 6 hours of break even, and then 6 hours of losses).
now i tried the docker image, but its literally the first time I even installed docker. I didn't see any stats output and I didn't know how to control niceness etc properly so I shut it down and compiled wolf's. I am getting only 1% better going from # cores to cache /2MB (over provision cores), so I'm sticking to core count because I would rather not eat extra cache.
That being said all this talk about sysctl -w vm.nr_hugepages=# did absolutely nothing for me. recompiled and all. I would love to get a 10% boost that allegedly comes from it but I dunno where I messed up.
Patrick, can I get stats for docker performance and can I made sure it is running true idle only (I set wolf's to nice -19 and it seems to have minimal impact on things once I disabled cpu usergroups)?
lastly, it seems people report much better performance from the windows client by claymore. Has anyone tested this. I tested claymore vs gui miner on windows and yeah claymore was like +45%, but then a 5% mining tax.
Also, my r9 290s are putting out 820 h/s each, which makes them pretty good as long as I don't run during weekday daytime (variable rate electricity (12 hours profit, 6 hours of break even, and then 6 hours of losses).
Try docker ps to get the container name. Then docker attach containername
That will give you console output. Close the SSH terminal to get out if ctrl p q does not work.
Alternatively instead of using -d to launch you can use -it. With the docker management tools you can just open logs and see via GUI.
That will give you console output. Close the SSH terminal to get out if ctrl p q does not work.
Alternatively instead of using -d to launch you can use -it. With the docker management tools you can just open logs and see via GUI.
Inspired by @SavageWS6
I complied xmr-stak-cpu bianry from GitHub - fireice-uk/xmr-stak-cpu: Monero CPU miner
Installed into a ASUS 4nodes RS720Q-E8-RS8-P chassis with 7 x E5-2650 v3 QS ( QFSB ) cpu.
Most optimal threads number is 22 for a dual E5-2650 v3 ( 10 core , 25MB cache ) = 858 H/s
Most optimal threads number is 12 for a single E5-2650 v3 = 465 H/s
Enable Hugepages support in Ubuntu 16.0.4
added: sudo sysctl -w vm.nr_hugepages=128
# /etc/security/limits.conf
* soft memlock 626688
* hard memlock 626688
Config.txt change: "no_prefetch" : true, ( from false )
Overall, power usage dropped from 600w (Wolf minerd )to 576w (xmr-stak-cpu )
there is a 1% donation to the developer of xmr-stak-cpu, should subtract 1% from xmr mined
10% H/s increase from Wolf miner to xmr-stak-cpu with single CPU
5-6% H/s increase from Wolf miner to xmr-stak-cpu with dual CPU
There is a docker version of xmr-stak-cpu, will try it later .
Data point:
Dual E5-2650 v3 QS ( 10 core , 25MB cache )
20 threads = 805 H/s
21 threads = 837 H/s
22 threads = 858 H/s
23 threads = 832 H/s
24 threads = 791 H/s
Single E5-2650 v3 QS ( 10 core , 25MB cache )
10 threads = 406 H/s
11 threads = 439 H/s
12 threads = 465 H/s
13 threads = 440 H/s
14 threads = 407 H/s
I complied xmr-stak-cpu bianry from GitHub - fireice-uk/xmr-stak-cpu: Monero CPU miner
Installed into a ASUS 4nodes RS720Q-E8-RS8-P chassis with 7 x E5-2650 v3 QS ( QFSB ) cpu.
Most optimal threads number is 22 for a dual E5-2650 v3 ( 10 core , 25MB cache ) = 858 H/s
Most optimal threads number is 12 for a single E5-2650 v3 = 465 H/s
Enable Hugepages support in Ubuntu 16.0.4
added: sudo sysctl -w vm.nr_hugepages=128
# /etc/security/limits.conf
* soft memlock 626688
* hard memlock 626688
Config.txt change: "no_prefetch" : true, ( from false )
Overall, power usage dropped from 600w (Wolf minerd )to 576w (xmr-stak-cpu )
there is a 1% donation to the developer of xmr-stak-cpu, should subtract 1% from xmr mined
10% H/s increase from Wolf miner to xmr-stak-cpu with single CPU
5-6% H/s increase from Wolf miner to xmr-stak-cpu with dual CPU
There is a docker version of xmr-stak-cpu, will try it later .
Data point:
Dual E5-2650 v3 QS ( 10 core , 25MB cache )
20 threads = 805 H/s
21 threads = 837 H/s
22 threads = 858 H/s
23 threads = 832 H/s
24 threads = 791 H/s
Single E5-2650 v3 QS ( 10 core , 25MB cache )
10 threads = 406 H/s
11 threads = 439 H/s
12 threads = 465 H/s
13 threads = 440 H/s
14 threads = 407 H/s
Seems a bit low. I am getting about the same/ higher numbers on the dual E5-2650L V3 which should be slower.
BTW @Marsh - Rancher is like cheating for managing this.
The issue is how hard it is to script the Stak setup. I asked him and he said he likes the control of doing it manually. If you can, happy to make a Docker image with it.
BTW @Marsh - Rancher is like cheating for managing this.
The issue is how hard it is to script the Stak setup. I asked him and he said he likes the control of doing it manually. If you can, happy to make a Docker image with it.
Well I did the same thing today without docker. I spent a few hours trying to get Stak to compile on Centos 7. then gave up, used an arch machine, and had it compiled in 2 minutes. Then I moved library after library to the working directly on the centos machine to try and get the damn thing to run. I failed. Either needs a container or someone better than me. I'm really trying to avoid containers though.
If someone does make a "Stak" container please incorporate a "nice" setting to launch it with low or adjustable priority. TYVM
If someone does make a "Stak" container please incorporate a "nice" setting to launch it with low or adjustable priority. TYVM