CUDA miner

1373840424369

Comments

  • newkidONdablocknewkidONdablock interwebzMember Posts: 121
    I'd say 3-5 months
  • feeleepfeeleep Member Posts: 31
    Hi,

    @Genoil - I am the op of coinmine.pl pool and i created coinotron compatible stratum server. It works fine with miner proxy however when I use your stratum miner it does not send anything to the pool. Not sure where the problem is but maybe you can help to debug it.

    feeleep
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @feeleep sure I'll have a look
  • feeleepfeeleep Member Posts: 31
    edited March 2016
    thanks - when i mine with getwork is works ok - just checked

    EDIT: to clarify this - miner receives notify messages from pool, it just does not generate shares and not try to send anything. At least this is how it looks in command window. I use CPU for testing
  • Bensam123Bensam123 Member Posts: 62
    edited March 2016
    -E old causes ethminer to crash on launch.
  • cgladuecgladue Member Posts: 2

    I'd say 3-5 months

    thank you
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @feeleep there's a small difference between your implementation and coinotron's. You send the header hash, seed hash and target with "0x" prepending the hex value and coinotron doesn't. I can check for that.

    @Bensam123 thanks for your feedback. I am able to reproduce the behavior. It happens when you are using the default DAG storage directory (which I wasn't). A temporary fix would be to specify that directory using the -R flag. Will fix this tonight or tomorrow. How about --cuda-devices / --opencl-devices? Does that work as advertised?
  • fredeqfredeq Member Posts: 42
    Heya

    First of all I would like to share my config (1x280x + 2x380):
    setx GPU_FORCE_64BIT_PTR 0
    setx GPU_MAX_HEAP_SIZE 100
    setx GPU_USE_SYNC_OBJECTS 1
    setx GPU_MAX_ALLOC_PERCENT 100
    setx GPU_SINGLE_ALLOC_PERCENT 100
    ethminer.exe --farm-recheck 200 -G -F http://127.0.0.1:8080/f1 --cl-local-work 256 --cl-global-work 16384
    Runs all 3 cards at 20,7Mhash. Quite surprised that 380 is able to match 280x speeds.
    Will have to sell after ETH mining is done though.

    @Genoil Maybe I am doing something wrong, but setting "--opencl-devices" to either "0" or "1" or "2" always starts my 280x ("0" card).
    If I submit two digits, it always chooses "0" and "1" cards (even If I do "--opencl-devices 3 5")
    Maybe something is wrong with my rig though? Doing "--list-devices" returns miner title + donate and then exits.

    On my 2x750ti rig starting without "--cuda-devices" indeed starts both cards.
    Sadly 750ti under win7 are crap...
    Any chance you know why 750tis cant hash efficiently under win7? Any way to fix the miner to support it?
    So far I could get my monitor card to 200Khash and second one to 800Khash ;)
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @fredeq thanks for your feedback. I guess I'll need to spend some time on a 4 GPU rig to fix that.
  • Bensam123Bensam123 Member Posts: 62
    Genoil said:

    @Bensam123 thanks for your feedback. I am able to reproduce the behavior. It happens when you are using the default DAG storage directory (which I wasn't). A temporary fix would be to specify that directory using the -R flag. Will fix this tonight or tomorrow. How about --cuda-devices / --opencl-devices? Does that work as advertised?

    Yup, you don't need to use cuda-devices anymore to have all the devices selected.

  • antonio8antonio8 Member Posts: 44
    @Genoil

    Tried the 1.0.6b for about 15 minutes to test it out. Avg is about 50-51 Mh/s for my rig. I could get more as I did not OC the memory. I just left it at factory OC setting.

    If I remember correctly I was only getting about 48 Mh/s before. I used the exact same bat.

    Nice.
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @antonio8 that must be magic. not a single bit was changed in the kernel or mining code :).
  • LeChuckDELeChuckDE Member Posts: 46
    @antonio8 , @Genoil seems normal varrity in hashresults. same here got on both little bit more 5,6 MH/s cuda 5,5 Mh/s before and 4,0 MH/s openCL 1.1 3,6 MH/s before. Start a benchmark few seconds ago and now 5,4 MH/s Cuda and 3,7MH/s openCL 1.1
  • antonio8antonio8 Member Posts: 44
    Genoil said:

    @antonio8 that must be magic. not a single bit was changed in the kernel or mining code :).

    Should have mentioned that was the Sept release. 48.4 Mh/s just reran it.
  • feeleepfeeleep Member Posts: 31
    @Genoil - it must be something different - coinotron sends 0x as well and also did testing on my side removing 0x from values and still no luck. It must be something different but dont know what... yet...
  • thelosers106thelosers106 Member Posts: 6
    @Genoil hey.. I am having some issue mining ethereum on a 750ti.. I keep getting this when I start it up and then the driver will crash..

    I have tried using --cuda and also --opencl .. both produces the same error.. I manage to get it working with this additional command (--cuda-block-size 32 --cuda-grid-size 4096 --cuda-streams 1) but then the speed is showing up like this

    Any ideas to why this is happening?

    Thanks,
    theLosers106.
    3.png 85.9K
    4.png 38.9K
  • dexkzdexkz Member Posts: 3
    Hello,
    i have two same config PC ( GTX980 ), one running win 8.1 other win 10. The first one gives 20-22 mh\s the other one, barely 5-6 mh\s
    What can I do, without changing the OS to win 8.1? Thank you!
  • LeChuckDELeChuckDE Member Posts: 46
    Install the old driver 347 in Windows 10
  • Bensam123Bensam123 Member Posts: 62
    @Genoil What can you do about recovery? When one of the GPUs in a rig fails right now the miner shuts down. You can manually restart the miner every couple hours with a batchfile, but that's not the optimal solution and often times you get penalized pool side for a interruption.

    Would it be possible to have the miner automatically restart and if the card is problematic shut it down until the program is manually restarted?
  • dexkzdexkz Member Posts: 3
    LeChuckDE said:

    Install the old driver 347 in Windows 10

    Yeah, I did. The speed is now normal, but it keeps falling off and make 0 hashrate some times. Like @thelosers106

    prntscr.com/ajbtfc
  • dreambottledreambottle Member Posts: 7
    @Genoil, I'm curious, is there still any room for improvement in mining speed, or are current speeds the maximum we can expect?
  • LeChuckDELeChuckDE Member Posts: 46
    @dexkz same by my setup, the WDMA 2.0 from Windows 8.1/10 is the problem, the Driver crashes in background (0 H/s) comes up again and crashes .... changed to Ubuntu 15.10 and compile my binaries by my self.
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    edited March 2016
    @dreambottle don't expect hashrate improvements. The current kernel has been public for about 9 months already. If it could be faster, somebody would have made it. Or somebody has and is keeping it quiet :smile:

    That said, I found somebody yesterday that apparently has brought the VGPRS usage on AMD down to 67. That's 9 less than my kernel, 17 less than official. If he could bring it down to 64, AMD might be in for some extra hash.
    Post edited by Genoil on
  • thelosers106thelosers106 Member Posts: 6
    LeChuckDE said:

    @dexkz same by my setup, the WDMA 2.0 from Windows 8.1/10 is the problem, the Driver crashes in background (0 H/s) comes up again and crashes .... changed to Ubuntu 15.10 and compile my binaries by my self.

    But I am using Windows 7 though.. @Genoil any ideas?
  • hozehoze Member Posts: 27
    Genoil said:

    1.0.6b is out. It appears to work fine, but I'm just leaving it in beta for a bit because I couldn't test the new opencl-devices flag and fix of the cuda-devices flag yet. So please test and let me know if it is working or not:

    Changes:
    --opencl-devices: flag allows you to specify which opencl devices you want to use. 0 based, space delimited.

    -FF/--farm-failover flag allows you to specify failover farm URL. this was requested by one of my donators with one specific goals in mind: the ability to run a failover stratum proxy for added stability. clever. this is also the main reason i haven't added a native stratum failover.

    --failover-retries. the name speaks for itself

    -when mining with multiple CUDA cards, you shouldn't have to specify all cards any longer with --cuda-devices. Instead it will try to use them all. Please test on multi-GPU rig.

    -E/--erase-dags : Erase unneeded DAG files. Default is 'none'. Possible values are:
    none - don't erase DAG files (default)"

    For me the version is crash on gtx 950. Amd working fine.


    Windows 10 64bit, driver 347.52. 0.9.4.1 genoil 1.0.5 version working fine.
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    If you use 347 drivers, you have to compile with cuda 6.5. The 1.0.6b Win64 binary crashes when using the -E flag combined with not using the -R flag. It's fixed in src. Before I release anew binary, I have another issue to fix.
  • newkidONdablocknewkidONdablock interwebzMember Posts: 121
    so where are we standing now? newest drivers or 347 drivers? which are better? also win 7 or 8/10 recommended?
  • blueboxbluebox Member Posts: 181 ✭✭
    @newkidONdablock If you're using win7/8.0 you should be able to use the cuda 7.5 installer which comes with 352.xx and should work fine. I'm using the latest 36x.xx driver with the latest cuda7.5 toolkit and having no problems with Genoil's 1.0.5 on both AMD R7-370 and nVidia 970 cards... fwiw
  • newkidONdablocknewkidONdablock interwebzMember Posts: 121
    im using win 7 and 364 latest driver but I didnt install cuda toolkit, should I ?
  • hozehoze Member Posts: 27
    Genoil said:

    If you use 347 drivers, you have to compile with cuda 6.5. The 1.0.6b Win64 binary crashes when using the -E flag combined with not using the -R flag. It's fixed in src. Before I release anew binary, I have another issue to fix.

    OK. Thanks.
Sign In or Register to comment.