CUDA miner

1262729313269

Comments

  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    edited February 2016
    I replaced the 1.0.4 beta with a beta2 zip. Added automatic reconnecting on failures. Still on my todo list for the release version is to actually shut down the miners until a connection is available to prevent waste of energy when the pool stays down.

    @bluebox the increase you see has nothing to do with actual better performance. The way hashrate is calculated in ethminer is rather crude, causing significant rounding errors depending on the --farm-recheck value. In the stratum version, the only thing --farm-recheck still does is set the interval between hashrate calculations. Low values tend to produce higher hashrates, higher values tend to produce more realistic values. The GPUs meanwhile are just hashing away in different threads.
  • JerkyJerky Member Posts: 16
    So I am building a new rig. Is there anything wrong with running this miner for R9 280x? I like the dag location option because I am going to mount a ramdisk for the dags on a USB drive Linux machine. Is there a major issue with this?
  • ArtikoArtiko Member Posts: 4
    Hello @Genoil

    First, thanks for your work on this cuda miner, it is more CPU friendly than openCL...
    Stratum support is a very good idea, so i've tried to build 1.0.4 last night en ubuntu.

    Did you build only on windows ? I needed to fixes somes error with jsoncpp lib and one thing in a cmakefile. I'm programmer but not really fine with cpp and cmake, i will put modified on github and you may integrate them if interesting

    This morning i got a infinity loop with message "Read response failed: End of file", your reconnect on beta2 may fix that i will try it.

    I was looking for mining to other pool like ethpool et supranova but they seem (by network sniffing) to have another "stratum" protocol than ur miner et coinotron... They seems only to stream jsson works.
    There is differents stratum protocols ?
  • ArtikoArtiko Member Posts: 4
    I did a pull request on github.
  • Mo35Mo35 Member Posts: 43
    edited February 2016
    its hopeless, im giving up, i cant get to build this on windows (nor win10 , nor 7). Anyone around is able compile latest master (1.0.3) with 6.5 cuda support ?
  • ArtikoArtiko Member Posts: 4
    edited February 2016
    Mo35 said:

    its hopeless, im giving up, i cant get to build this on windows (nor win10 , nor 7). Anyone around is able compile latest master (1.0.3) with 6.5 cuda support ?

    Binaries 1.0.3 seems to be available here : https://github.com/Genoil/cpp-ethereum/tree/master/releases
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @Artiko thanks for your PR! I will test it first tomorrow but it looks fine. I'm developing on Windows, so any Linux issues mostly escape my attention. The 1.04b2 does auto reconnect, but it isn't exactly issue free yet, got an issue reported with ethpool already, which I will have to look into.

    ethpool should work (orher than the reconnect issue) as @dr_pra has already contacted me and is testing as well. Other pools do indeed use some custom stratumish protocol, but I'm not planning to support those. Stratum is a protocol with a spec, not some kind of loose agreement you can change however you like and label "stratum" ;).

    @Mo35 what are you actually running into?
  • Mo35Mo35 Member Posts: 43
    edited February 2016
    @Genoil not sure what im missed , or what im doing wrong , but i cant get past to cmake part , output & error logs doesnt realy help me much to understand whats up , there is some spam about
    error C2065: 'CL_VERSION_2_0' : undeclared identifier and fatal error C1083: Cannot open include file: 'pthread.h'

    cmakeerror log : http://expirebox.com/download/469be5f2ea70dbf9f31c443050fd98fb.html
    cmakeoutput log : http://expirebox.com/download/101b3c5d4515bfee98e2b1b63738da24.html

    what i did is installed latest stable Cmake , VS12 2013 c++ part , installed Win7SDK , C++ redist. and CUDA toolkit 7.0 . I get trough insanely slow external dep. part , making build dir. and the failing miserably with Cmake

    tried same setup on win10 and 7 , with identical results
  • blueboxbluebox Member Posts: 181 ✭✭
    1.0.4 beta1 failed after 12 hours (windows 7), not sure if it was a problem with @coinotron connectivity or not. Went back to standard mining_proxy, but will try beta2 tonight.

    Network hash rate is down, difficulty starting to turn down as well... .00875btc/eth, wow
  • dr_pradr_pra Member Posts: 445 ✭✭✭
    The ethpool.org support for the stratum connection of cudaminer is currently still in private testing. We will release it as soon as Genoils stratum implementation is out of beta.
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @Mo35 I recognize these errors. I'll try to replicate them and remember how I fixed them.
  • Mo35Mo35 Member Posts: 43
    @Genoil
    Well im making some progress , initial issue solved , dont use Cmake higher than 3.4.3 ...

    now im kinda gliching with that botched buildinfo.h part
    1. opening ethereum.sln
    2. setting ethminer as startup project
    3. setting up as release x64
    4. building and failing on buildinfo.h
    at this step what should i do with file modifications reload/all , ignore/all on external changes ?
    im reloading all .
    output looks following :
    Error 1 error C1083: Cannot open include file: 'BuildInfo.h': No such file or directory C:\ethtest\libdevcore\Common.cpp 25 1 devcore
    Error 34 error LNK1181: cannot open input file '..\libdevcore\Release\devcore.lib' C:\ethtest\ethminer\LINK ethminer

    Then from Buildinfo.h im deleting header file and adding existing buildinfo.h from build dir.
    when im retrying build output is following

    1>------ Build started: Project: ZERO_CHECK, Configuration: Release x64 ------
    2>------ Build started: Project: devcore, Configuration: Release x64 ------
    3>------ Build started: Project: BuildInfo.h, Configuration: Release x64 ------
    2> Common.cpp
    3> ETH_FATDB: 0
    3> Files "C:/ethtest/BuildInfo.h.tmp" to "C:/ethtest/BuildInfo.h" are different.
    3>
    2> devcore.vcxproj -> C:\ethtest\libdevcore\Release\devcore.lib
    4>------ Build started: Project: ethminer, Configuration: Release x64 ------
    4> main.cpp
    4>LINK : fatal error LNK1181: cannot open input file 'pthread.lib'
    ========== Build: 3 succeeded, 1 failed, 6 up-to-date, 0 skipped ==========
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @Mo35 this should be fixed. I don't know where that pthread comes from. But I do know how you can make it build now.

    1. right click ethminer in Solution Explorer
    2. Choose Release (or All) Configuration
    3. Open Configuration Properties > Linker > Input
    4. Edit additional dependencies, remove pthread.lib from the list

    Rebuild

    I'll have a look into fixing all this crap today so it's easier for people to build.
  • Mo35Mo35 Member Posts: 43
    @Genoil Finaly after 3 days of headbanging , success ... 1.0.3 with older cuda support ;)
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @Mo35 cool. And..does it work as expected with the older drivers?

    I applied a small hack to make it compile on MSVC without the BuildInfo thing. It still isn't exactly how it should be because the MSVC project still can't find the file in the IDE, but at least it compiles without having to fiddle around. I didn't run into the pthread issue again.
  • Mo35Mo35 Member Posts: 43
    edited February 2016
    @Genoil basicaly i got what i wanted , im running 347.52 under win10 , with my previosly posted hacks i forced win to eat up thos older drivers and prevented windows from trying to update them via GP, additionaly with https://support.microsoft.com/en-us/kb/3073930 i made Windows Update to stfu and stop offering me newly released nvidia drivers , by banning them (rest of updates shows up nicely) , i gained ~2 Mhz over 1.0.1 build and i got ability to move DAGs to other disk. So finaly have fully functional Win10 eth mining setup , with expected hash power and functionality i wanted.

    GTX970 + 2xGTX950 nets almost @40Mhz, and using --cuda-streams 1 , when i want to watch movie or just do something with less graph.lag , sets me @ ~38Mhz

    Thanks for the help and advices!

    p.s. btw pthread was actualy listed twice in dependacies for me , i built from latest master branch 1.0.3
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @Mo35 that's great to hear! I bet others would want to know how you did it. If you could prepare a simple README or al least send me the binaries, I will host them along the other releases.
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    1.0.4 beta3 is up.
    - fixes some Linux build issues
    - stops miners when disconnected from stratum
    - reconnection issues on ethpool (still in closed beta) have been resolved server-side

    If I don't get too many issues reported with the stratum miner, I guess I'll label it "release" after the weekend and I can start thinking about 1.0.5. Or maybe I'll label it 1.1 if my plan works out well ;)


  • crsminercrsminer Member Posts: 75
    @ all

    Hey guys please donate some ETH to Genoil, he is the only dev who shares his work on ethminer for free!

    @Genoil maybe you advertise also your BTC address since making transfers with ETH it's not so simple :wink:
  • Mo35Mo35 Member Posts: 43
    @Genoil run into issue , again ;)
    my 1st build was with cuda toolkit 7.0 , now im trying to rebuild with 6.5 , and ending up with following
    ------ Rebuild All started: Project: ethash-cuda, Configuration: Release x64 ------
    6> Building Custom Rule C:/ethtest/libethash-cl/CMakeLists.txt
    6> CMake does not need to re-run because C:\ethtest\libethash-cl\CMakeFiles\generate.stamp is up-to-date.
    7> Building NVCC (Device) object libethash-cuda/CMakeFiles/ethash-cuda.dir/Release/ethash-cuda_generated_ethash_cuda_miner_kernel.cu.obj
    6> Generating OpenCL Kernel Byte Array
    7> nvcc fatal : Unsupported gpu architecture 'compute_52'
    7>
    7> CMake Error at ethash-cuda_generated_ethash_cuda_miner_kernel.cu.obj.cmake:207 (message):
    7> Error generating
    7> C:/ethtest/libethash-cuda/CMakeFiles/ethash-cuda.dir//Release/ethash-cuda_generated_ethash_cuda_miner_kernel.cu.obj

    any tips how to fix that ?

    p.s. noobish how to readme , including windows hacks and newcomer tips how to compile is in progress
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    You should install CUDA 6.5 with Compute 5.2 support. Sorry don't have a link.
  • Mo35Mo35 Member Posts: 43
    edited February 2016
    @Genoil Yeah , figured myself :) , Cuda 6.5.14 wich is available dosnt support compute 5.2 , there is patch wich upgrades toolkit to 6.5.19 with compute 5.2 support.
    https://developer.nvidia.com/cuda-downloads-geforce-gtx9xx

    succesfully compiled 1.0.3 master branch and 1.0.4b3 for cuda 6.5 with compute 5.2 card support
    feel free to use if you need , but preferably let Genoil check them and aprove release beforehand
    http://wikisend.com/download/361278/ethminer1.0.3_master_6.5cuda.zip
    http://wikisend.com/download/960168/ethminer1.0.4b3_stratum_6.5cuda.zip

    "how to" will eventualy follow when i write up my adventures as newbie in compiling stuff
    Post edited by Mo35 on
  • simplysimply Member Posts: 10
    @Mo35 Did you compiled with Cuda 6.5 Toolkit for GTX 970 and GTX 980 or with regular Cuda 6.5?
  • Mo35Mo35 Member Posts: 43
    edited February 2016
    @simply with cuda 6.5.19 gtx9xx (compute 5.2) support included , that was the only possible way , using it myself with 970 and 950 cards
  • ArtikoArtiko Member Posts: 4
    Mining on coinotron stratum since 48hours using 1.0.4b2 on linux with GTX 750Ti & GTX 960
    Work very well !
  • saiphirsaiphir The NetherlandsMember Posts: 7
    edited February 2016

    I got it to work, thank you @Genoil ! 40 MH/s with 2x GTX 970s with CUDA 7.5 on linux 17.3.

    Does anyone possibly know how to unlock the voltages/power on these cards under linux 17.3 352.39? Also they seem to be stuck on performance level 2, reverting back from 3 when starting ethminer :/ Tehy should be able to run on 7000 Mhz right? now on 1441 Mhz (which I am able to change in powermizer) but the memory transfer rate seems to be stuck on 6000 Mhz on both cards.
  • simplysimply Member Posts: 10
    @saiphir I'm also stuck on 6000 MHz with my GTX 970, but on windows
  • Mo35Mo35 Member Posts: 43
    it could be driver releated , after i tinkered with my win10 and running it on 347.52 win7 driver , im unable to alter mem clock at all on 970 strix , tho on 950 strix im able to overclock both gpu and mem clock ,
  • simplysimply Member Posts: 10
    I can boost the memory clock, but it does not matter. When I start ethminer it goes to 6000
  • happytreefriendshappytreefriends Member Posts: 537 ✭✭✭
    edited February 2016
    Genoil, do you want donations in BTC or ETH? Thank you for all your work and I want to donate a bit to ya.

    I only use your 1.0.3 ethminer in windows 7 with Radeon GPU;s but still getting about 3-4% extra from your miner. *PS If you have anything special, send it my way* B)
Sign In or Register to comment.