CUDA miner

1434446484969

Comments

  • dreambottledreambottle Member Posts: 7
    I have installed Linux Mint to achieve better results with a spare GTX750Ti that I have (currently unusable under Windows) and also compare the results of GTX770 with the Windows 7 setup.

    And I currently struggle to compile the cuda miner.

    I have installed CUDA tooolkit 7.5 from the .deb package and followed the guidelines at cpp-ethereum docs, but the compiled ethminer does not have most cuda-specific commands. Do I need to set any additional env variables or flags while compiling?
  • blueboxbluebox ✭✭ Member Posts: 181 ✭✭

    @Genoil Thanks a lot for your answer. I am currently using your Ethminer version on Windows too. Because it allows me to set different DAG folder and have different cards to mine in different pools. With MacOSx I am still struggling to build it. If you need access to a mac, I can let you use mine throught TeamViewer for a couple of minutes of hours. I think it is probably something silly that needs to be done for the build to work. But I have no experience in programming for that.

    @robinminer If you really, really feel the need to mine under OS X you can do it, but I wouldn't recommend compiling on your own. I bashed my head on it for a few days, that's all I'm giving it.

    Two strikes against: One, the etherminer fork has issues compiling under various X versions, mostly related to clang/gcc/dep issues. I got @Genoil's cudaminer to 99% one time, where it crapped out on the "ethstratum" objects, even though I had ETHSTRATUM de-selected. Thought about tearing it out of the code but I'd just be making it a bigger headache than it's worth... Two, if you went opencl only (if you could get that to compile), you're beholden to Apple's lesser-than-optimum opencl implementation for AMD chips, and opencl mining on nvidia isn't quite as good as cuda at the moment. Double whammy.

    Best bet is to go with the homebrew cpp-ethereum tap, which works (opencl only), but you'll get 65% of the equivalent hashrate on an AMD chip, probably 50% or less on nVidia.

    That's why I just went with a bootcamp win7 install on my Pro tower with a pair of cheap R7-370's to get the best out of the situation (31MH).

    My understanding of compiling code written on windows VB/.Net/C++ and porting to any modern unix flavor is that Microsoft has yet — after what, 10-15 years? — to support modern C++ code (C99?) or better. C# was their answer I guess, but I don't know/don't care about Windows. I only hear the windows devs in my group half-acknowledging it, but most of them are still in denial about Window's decline and OS X's rise to #2 over linux in most-developed platform. (Interesting that Microsoft is plugging bash into future win10...)
  • dreambottledreambottle Member Posts: 7
    edited April 2016
    Ok, I looked into cmakelists and figured out the flag for cmake: -DETHASHCUDA=ON. (Also enabled -DETHSTRATUM=ON for stratum.)

    It's working and I'm having amazing results for GTX770.

    But I need a little bit more help: to overclock the card on linux, I need to enable the corresponding NVidia coolbits. I've run `sudo nvidia-xconfig --cool-bits=28` (28 == 4 + 8 + 16 for fan control and full OC capabilities). But it seems to have no effect even after rebooting.
    I'm using 352.79 nvidia driver. Here's the generated /etc/X11/xorg.conf: http://pastebin.com/W0yib5dW
    Any ideas why it doesn't work?
  • workwork ✭✭✭✭ Member Posts: 2,084 ✭✭✭✭
    @dreambottle pretty sure coolbits option should be under the device section, not screen.

    Also, isn't it 1 = OC, 4 = fan control; so 5 = both? and 16 = voltage control so 21.

    ref: https://www.gpugrid.net/forum_thread.php?id=3525&nowrap=true#38973

    28 definitely isn't going to do what you want. Bit 0 is kinda important.
  • dominusdominus Member Posts: 36
    coolbits 28 works for sure with 750ti, 960, 970 and 980.

    sudo nvidia-xconfig -a --cool-bits=28 --allow-empty-initial-configuration
    It should make the xorg.conf for You.


  • miningpoolhubminingpoolhub ✭✭ Member Posts: 308 ✭✭
    Genoil said:

    I pushed a few commits to the "107" branch, including the option to immediately bail after a lost connection, thx @miningpoolhub fro the suggestion. Only works on getwork protocol for now.

    Add -FF exit --failover-retries 0 to command line options to achieve this.

    Additionally there's the OpenCL tweaks for NVidia cards that I was talking about earlier. This is useful for Nvidia people on Linux/OSX that have trouble compiling the cudaminer. With this change, OpenCL on NVidia is almost as fast as CUDA. Will try to get it just as fast. Not faster ;)

    Not releasing binaries of this yet. Still want to add failover support for stratum add apply a fix that some guy on bitcointalk gave to possibly improve stability of the stratum client. Will hopefully get to that tomorrow.

    @Genoil

    Great works. Seems like more useful when --failover-retries 0 works with Stratum too.

    Thanks.
  • thelosers106thelosers106 Member Posts: 6
    Hi,

    I'm having some trouble with the miner. I tried the command --erase-dags old but it keeps deleting the current DAG file and then it recreates them. This is my full command when I ran it:
    ethminer.exe --cuda -F http://127.0.0.1:4052/miner/10/minername --erase-dags old --dag-dir . --cuda-devices 0

    I even tried supplying it with the current block number (--current-block) but it still delete the current DAG file and then recreate it.. I am sure the DAG file is not corrupted as after it finished creating the DAG file, I let it run for a while and there's some valid submits too. Then I stopped it and re run it again the same thing happens.. Any help would be much appreciated.

    Thanks,
    theLosers106.
  • GenoilGenoil ✭✭✭ 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @thelosers106 --dag-dir . is probably the issue. have you tried --dag-dir ./dag for instance?
  • thelosers106thelosers106 Member Posts: 6
    @Genoil well its the same outcome (see attached pic).. just to make sure I got the right binaries, the latest is in the release directory am I correct?

  • GenoilGenoil ✭✭✭ 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @thelosers106 do you by any chance have future dags already stored in the DAG directory?
  • thelosers106thelosers106 Member Posts: 6
    edited April 2016
    @Genoil Yes I do have future DAGs.. I just pre generated a few future DAGs just to make sure it did not delete them.. when I deleted the future DAGs that I have precomputed, it is now working perfectly.. thanks for the help (: .. not sure why this is causing such problem.. did not have time to look at your latest code yet..

    Edit: I had 2 future pre computed DAG files, basically the next DAG and the consequent DAG..
  • GenoilGenoil ✭✭✭ 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @thelosers106 the auto deletion feature isn't very sophisticated. It just reads the directory contents, skips the two largest files and deletes the rest. I should change the logic in such a way that it compares the seedhash of current DAG with a DAG filename in the directory and then deletes all DAGs that are smaller than that one. Thanks for the feedback.
  • thelosers106thelosers106 Member Posts: 6
    @Genoil Is it easy to estimate or get the exact size of what a DAG file will be without generating it? If yes, then just estimate the current DAG file size and delete which ever files that are smaller than the current one.. your method of comparing the seedhash is better to a certain degree as what IF someone creates a DAG file too far ahead.. IMO both ways have its pros and cons.. just a matter of which is easier/you like to implement..
  • CubirezCubirez Member Posts: 71
    Hey guys,
    I have some questions about the –cuda-devices command : is it possible to run two instances with 3cards each on the same 6 GPU rig?
    I want to do some testing as I have a problem with one of my rigs : I’ve got 2 GTX970 rigs with exactly the same setup, both mining on ethpool, the only difference being the GPU brand, and the risers (1 with ribbon/ 1 with USB). However, while they both report the same hashrate in ethminer, around 125MH/s, one of them is getting effective average hashrate at 125MH/s, where the second only gets 103 MH/s (this is based on over one week observations).
    So here are two suppositions i want to test: either there is a problem with the risers on this second rig (USB risers) or one/some of the cards aren’t working well.
    To test that I wanted to run two instances of Genoil’s miner : one with –cuda-devices 0 1 2, and the other with –cuda-devices 3 4 5. But I can’t make the second one to work (gpu will always launch no matter what cuda devices I declare).
    Any ideas?
    Here what’s the command looks like

    cpp-ethereum/build/ethminer/./ethminer --farm-recheck 200 -F http://127.0.0.1:8080/GPU345 -U --cuda-devices 3 4 5 --cuda-extragpu-mem 0  --cuda-block-size 128 --cuda-grid-size 8192
  • hozehoze Member Posts: 27
    Cubirez said:

    Hey guys,
    I have some questions about the –cuda-devices command : is it possible to run two instances with 3cards each on the same 6 GPU rig?
    I want to do some testing as I have a problem with one of my rigs : I’ve got 2 GTX970 rigs with exactly the same setup, both mining on ethpool, the only difference being the GPU brand, and the risers (1 with ribbon/ 1 with USB). However, while they both report the same hashrate in ethminer, around 125MH/s, one of them is getting effective average hashrate at 125MH/s, where the second only gets 103 MH/s (this is based on over one week observations).
    So here are two suppositions i want to test: either there is a problem with the risers on this second rig (USB risers) or one/some of the cards aren’t working well.
    To test that I wanted to run two instances of Genoil’s miner : one with –cuda-devices 0 1 2, and the other with –cuda-devices 3 4 5. But I can’t make the second one to work (gpu will always launch no matter what cuda devices I declare).
    Any ideas?
    Here what’s the command looks like



    cpp-ethereum/build/ethminer/./ethminer --farm-recheck 200 -F http://127.0.0.1:8080/GPU345 -U --cuda-devices 3 4 5 --cuda-extragpu-mem 0  --cuda-block-size 128 --cuda-grid-size 8192


    Hi. Just use custom DAG folder for second instances. I use custom for both:
    1. ethminer.exe --farm-recheck 400 -G -F http://127.0.0.1:8080/i5amd --dag-dir c:\eth\amddag\
    2. ethminer2.exe -U --farm-recheck 400 -F http://127.0.0.1:8080/i5cuda --dag-dir c:\eth\cudadag\
  • CubirezCubirez Member Posts: 71
    @hoze My problem is that i'm running only nvidia cards. So i have to specify the devices i'm using.
  • blueboxbluebox ✭✭ Member Posts: 181 ✭✭
    @Cubirez This isn't card type/brand specific. Run your first 3-card miner with your normal command; it will use the existing dag file in the default location. Run the second miner (3 4 5) with whatever custom --dag-dir directory you want, path doesn't matter so long as it exists. It should download a second dag file there and then run.
  • BCRAMR41006BCRAMR41006 Member Posts: 1
    Hey Genoil. Can u help me do this on my rig? I have a GTX730 4gb and would love to mine. Been working at it for a week and cant get it to with the ethminer setup.
  • excavateexcavate Member Posts: 7
    @BCRAMR41006 what exactly are you struggling with? What OS, type or error, error messages, etc?
  • CybullCybull Member Posts: 17
    I was wondering would decreasing the grid size on ethminer also decrease the amount of ether mined. Currently mining on win 7 on a gtx 770 and would like my PC to be less laggy but I don't want that to come at the cost of a major performance downgrade.
  • workwork ✭✭✭✭ Member Posts: 2,084 ✭✭✭✭
    Cybull said:

    I was wondering would decreasing the grid size on ethminer also decrease the amount of ether mined. Currently mining on win 7 on a gtx 770 and would like my PC to be less laggy but I don't want that to come at the cost of a major performance downgrade.

    Yes, it will almost certain decrease hashrate.
  • CubirezCubirez Member Posts: 71
    @bluebox I didn't express myself properly.
    I think there is a problem with --cuda-devices as it's impossible for me to separate my 6 nvidia cards into two instances of 3 cards each. Maybe this isn't an issue when using opencl devices and cuda devices.

    In my case here what's happening :
    When using --cuda-devices and calling 1 or 2 cards, GPU 0 will automatically be called
    For instance :
    --cuda-devices 1 will call gpus 0 and 1
    --cuda-devices 1 2 will call gpus 0, 1, 2
    --cuda-devices 0 1 will call gpus 0, 1, and 0 again, leading to seg fault

    When using --cuda-devices and calling 3 cards, GPU 5 will automatically be called :
    --cuda-devices 0 1 2 ==> gpu 0, 1, 2, 5 called
    --cuda-devices 3 4 5 ==> gpu 3, 4 , 5 and 5 again => seg fault

    To overpass this issue and be able to investigate on my rig problems, i called 2 ethminer instances using different DAG files, but one has 2 GPUs while there other has 4 :
    I called GPU 0 and 1 using --cuda-devices 1
    I called GPU 2, 3, 4, 5 using --cuda-devices 2 3 4

    Maybe there's a problem with the way --cuda-devices works, or i am missing something here :open_mouth:
  • dominusdominus Member Posts: 36
    ethminer.exe -U --farm-recheck 200 -F http://... --cuda-devices 0 1 -R C:\DAG\0
    ethminer.exe -U --farm-recheck 200 -F http://... --cuda-devices 2 3 -R C:\DAG\1

    I'm using those 2 batch files without problems. + Minning another coin with the 5th gpu.
    The --cuda-devices works as supposed in my rig. I can select the gpus as I wish.
  • GenoilGenoil ✭✭✭ 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    Thx for the feedback. Looks like a bug indeed. How's --opencl-devices working for you, despite the lower performance ?
  • CubirezCubirez Member Posts: 71
    @Genoil Just tried with --opencl-devices, it worked perfectly well.
    So there might be a little bug with the --cuda-devices allocation :)
  • modilekamodileka Member Posts: 1
    Hi. Anyone want to help me to get started mining with my gtx 970 and i7?
    I'm some what familiar with mining but it's been atleast year since last time so little guide would be right on time.
    Thanks
  • GenoilGenoil ✭✭✭ 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @Cubirez it's not going to be simple to fix, as @dominus is reporting correct behavior. Can you show me some useful output from nvidia-smi when running different variations of --cuda-devices?
  • CubirezCubirez Member Posts: 71
    @Genoil What do you mean with useful output from nvidia-smi?

    Here's the standard output i get when running cards 2, 3, 4 and 5, with the following command
    cpp-ethereum/build/ethminer/./ethminer --farm-recheck 200 -F http://127.0.0.1:8080/GPU2345 -U --cuda-devices 2 3 4  --dag-dir /home/me/.ethash2 --cuda-extragpu-mem 0 --cuda-block-size 128 --cuda-grid-size 8192
    
    [email protected]:~$ nvidia-smi
    Sun Apr 10 16:18:01 2016
    +------------------------------------------------------+
    | NVIDIA-SMI 352.79     Driver Version: 352.79         |
    |-------------------------------+----------------------+----------------------+
    | 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 970     Off  | 0000:01:00.0     Off |                  N/A |
    | 81%   59C    P8    25W / 172W |     15MiB /  4095MiB |      0%      Default |
    +-------------------------------+----------------------+----------------------+
    |   1  GeForce GTX 970     Off  | 0000:02:00.0     Off |                  N/A |
    | 80%   58C    P8    20W / 172W |     15MiB /  4095MiB |      0%      Default |
    +-------------------------------+----------------------+----------------------+
    |   2  GeForce GTX 970     Off  | 0000:03:00.0     Off |                  N/A |
    | 84%   71C    P0   177W / 172W |   1448MiB /  4095MiB |     99%      Default |
    +-------------------------------+----------------------+----------------------+
    |   3  GeForce GTX 970     Off  | 0000:06:00.0     Off |                  N/A |
    | 89%   79C    P0   174W / 172W |   1448MiB /  4095MiB |     99%      Default |
    +-------------------------------+----------------------+----------------------+
    |   4  GeForce GTX 970     Off  | 0000:07:00.0     Off |                  N/A |
    | 89%   78C    P0   179W / 172W |   1448MiB /  4095MiB |     99%      Default |
    +-------------------------------+----------------------+----------------------+
    |   5  GeForce GTX 970     Off  | 0000:08:00.0     Off |                  N/A |
    | 89%   80C    P0   149W / 172W |   1448MiB /  4095MiB |    100%      Default |
    +-------------------------------+----------------------+----------------------+
    
    +-----------------------------------------------------------------------------+
    | Processes:                                                       GPU Memory |
    |  GPU       PID  Type  Process name                               Usage      |
    |=============================================================================|
    |    2      8812    C   cpp-ethereum/build/ethminer/./ethminer        1431MiB |
    |    3      8812    C   cpp-ethereum/build/ethminer/./ethminer        1431MiB |
    |    4      8812    C   cpp-ethereum/build/ethminer/./ethminer        1431MiB |
    |    5      8812    C   cpp-ethereum/build/ethminer/./ethminer        1431MiB |
    +-----------------------------------------------------------------------------+
    
    
  • blueboxbluebox ✭✭ Member Posts: 181 ✭✭
    @Cubirez Didn't catch that you were running a linux compile. @dominus and I are both running on windows, where --cuda-devices works as expected, sorry...
  • CybullCybull Member Posts: 17
    Can anyone tell me what "sulotion found submitted to ..." means on ethminer?
Sign In or Register to comment.