Skip to content

Latest commit

 

History

History
159 lines (117 loc) · 8.09 KB

README.md

File metadata and controls

159 lines (117 loc) · 8.09 KB
Raw

nvidia-docker image for Tpruvot CCMiner (CUDA)

This image uses Tpruvot CCMiner from my own Debian/Ubuntu mining packages repository. It requires a CUDA compatible docker implementation so you should probably go for nvidia-docker. It has also been tested successfully on Mesos 1.2.1.

compute52 - cuda9 variants - klaust

According to djm34 Lyra2Z (at least) is way faster if CUDA compute level is limited to 5.2 instead of enable all features for recents CPUs.

So in my mining packages repository, I provide two flavors of ccminer-tpruvot packages. The second one is ccminer-tpruvot-compute52 that can be turned into a CUDA docker image by building Dockerfile.compute52 (see below).

For instance, here are a few benchmarks when mining Zcoin (XZC) zith cuda-ccminer or cuda-ccminer-compute52 images:

cuda-ccminer cuda-ccminer-compute52
GTX1070 1275 kH/s 1496 kH/s
GTX1080 1593 kH/s 1871 kH/s
GTX1080 (another) 1651 kH/s 1932 kH/s
GTX1080Ti (PwrLimit 200) 2352 kH/s 2746 kH/s

Speaking about Lyra2Z, please run ccminer with --submit-stale otherwise you will loose nearly half of your hashrate !

Also, if you are running NVIDIA driver >= 384.81, you can use CUDA 9 variants :-) It's usually faster:

GTX1080 / XZC
Regular build (CUDA 8) 1634 kH/s
CUDA level 5.2 only (CUDA 8) 1915 kH/s
Regular build (CUDA 9) 2151 kH/s
UDA level 5.2 only (CUDA 9) 2153 kH/s

This is probably highly related to the cypher, in the example above you can see there was clearly something wrong for this crypto with compute level > 5.2. The "bug" has been fixed in CUDA 9 which, by the way, give an additional hashrate boost.

I'd be very happy to add your own hashrates here if you do some testing !

I also provide a package (and a docker image) for KlausT fork which is actively developped as well. CUDA >= 9.0 because he's dropping previous support anyway (but again, you don't need to have CUDA 9 on the host, the only requirement is to have driver >= 384).

-api variants

A new variant is being introduced, only for tpruvot / tpruvot-cuda9 (I'll provide it for all images as soon as I do a full cleanup of Dockerfile + meta script for generating them).

It provides the PHP files connecting the TCP api and embed Apache2+PHP as well as a Python wrapper I made myself to start both Apache2 and CCMiner.

nvidia-docker run -dt --restart=unless-stopped -p 4068:4068 -p 4069:4069 --name cuda-xvg-ccminer acecile/cuda-ccminer-cuda9-api /root/multi-process-launcher/multi-process-launcher.py --cmd '/root/start-apache2.sh 4068 4069' --cmd '/usr/bin/ccminer -a lyra2v2 -o stratum+tcp://xvg-lyra.suprnova.cc:2595 -u greenie.rig1 -p x --api-bind=0.0.0.0:4068 --api-allow=0/0'

You can now access status page at http://you.host:4069 and live hashrate graphs at http://you.host:4069/websocket.htm (yes, .htm it's not a typo). It's currently using modified version of the existing examples but I hope to get them merged soon in tpruvot's repository.

Build images

git clone https://github.com/eLvErDe/docker-cuda-ccminer
cd docker-cuda-ccminer
docker build -t cuda-ccminer .
docker build . -t cuda-ccminer-api -f Dockerfile.api
docker build . -t cuda-ccminer-compute52 -f Dockerfile.compute52
docker build . -t cuda-ccminer-cuda9 -f Dockerfile.cuda9
docker build . -t cuda-ccminer-cuda9-api -f Dockerfile.cuda9.api
docker build . -t cuda-ccminer-cuda9-compute52 -f Dockerfile.cuda9.compute52
docker build . -t cuda-ccminer-cuda9-klaust -f Dockerfile.cuda9.klaust
docker build . -t cuda-ccminer-cuda9-klaust-api -f Dockerfile.cuda9.klaust.api

Publish it somewhere

docker tag cuda-ccminer docker.domain.com/mining/cuda-ccminer
docker push docker.domain.com/mining/cuda-ccminer

Test it (using dockerhub published image)

nvidia-docker pull acecile/cuda-ccminer:latest
nvidia-docker run -it --rm acecile/cuda-ccminer /usr/bin/ccminer --help

An example command line to mine Groestl on MiningPoolHub (ccminer supports nearly all algorythm so check its documentation and picks what you want):

nvidia-docker run -it --rm -p 4068:4068 --name cuda-grs-ccminer acecile/cuda-ccminer /usr/bin/ccminer -a groestl -o stratum+tcp://europe1.groestlcoin.miningpoolhub.com:20486 -u acecile.catch-all -p x --api-bind=0.0.0.0:4068 --api-allow=0/0

Ouput will looks like:

*** ccminer 2.2.3 for nVidia GPUs by tpruvot@github ***
    Built with the nVidia CUDA Toolkit 8.0 64-bits

  Originally based on Christian Buchner and Christian H. project
  Include some kernels from alexis78, djm34, djEzo, tsiv and krnlx.

BTC donation address: 1AJdfCpLWPNoAMDfHF1wD5y8VgKSSTHxPo (tpruvot)

[2017-12-05 23:15:44] 1 miner thread started, using 'groestl' algorithm.
[2017-12-05 23:15:44] Starting on stratum+tcp://europe1.groestlcoin.miningpoolhub.com:20486
[2017-12-05 23:15:44] Stratum difficulty set to 5 (0.01953)
[2017-12-05 23:15:44] GPU #0: Intensity set to 19, 524288 cuda threads
[2017-12-05 23:15:45] API open in full access mode to 0/0 on port 4068
[2017-12-05 23:15:49] GPU #0: GeForce GTX 1080 Ti, 26.08 MH/s
[2017-12-05 23:15:51] accepted: 1/1 (diff 0.036), 26.51 MH/s yes!
[2017-12-05 23:16:03] GPU #0: GeForce GTX 1080 Ti, 26.60 MH/s
[2017-12-05 23:16:03] accepted: 2/2 (diff 0.065), 26.55 MH/s yes!
[2017-12-05 23:16:04] accepted: 3/3 (diff 0.069), 26.58 MH/s yes!

Background job running forever

nvidia-docker run -dt --restart=unless-stopped -p 4068:4068 --name cuda-grs-ccminer acecile/cuda-ccminer /usr/bin/ccminer -a groestl -o stratum+tcp://europe1.groestlcoin.miningpoolhub.com:20486 -u acecile.catch-all -p x --api-bind=0.0.0.0:4068 --api-allow=0/0

You can check the output using docker logs cuda-grs-ccminer -f

Use it with Mesos/Marathon

Edit mesos_marathon.json to replace miner parameter, change application path as well as docker image address (if you dont want to use public docker image provided). Then simply run (adapt application name here too):

curl -X PUT -u marathon\_username:marathon\_password --header 'Content-Type: application/json' "http://marathon.domain.com:8080/v2/apps/mining/cuda-ccminer?force=true" -d@./mesos\_marathon.json

You can check CUDA usage on the mesos slave (executor host) by running nvidia-smi there:

Wed Dec  6 00:21:09 2017       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 375.82                 Driver Version: 375.82                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 108...  On   | 0000:82:00.0     Off |                  N/A |
| 51%   67C    P2   197W / 200W |   1729MiB / 11172MiB |     98%      Default |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID  Type  Process name                               Usage      |
|=============================================================================|
|    0     21474    C   /usr/bin/ccminer                              1729MiB |
+-----------------------------------------------------------------------------+