CUDA miner

1171820222369

Comments

  • taylor_luktaylor_luk Member Posts: 16
    @MrYukonC that is related to some timing problem. I think there is a github issue tracking this problem.
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    edited September 2015
    @ol92 all I know is that the speed doesn't seem to be limited by the memory bandwidth but by something else, GPU side. At this point I simply don't know what to do about it, if anything.

    You can verify that indeed by under/ over clocking your memory. Doesn't have any effect (only tested on 780), while changing the GPU clock has an immediate effect.
    Post edited by Genoil on
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    Question to GTX 950/960 2GB Windows users: Do you see a big difference between benchmark & real mining modes? I'm talking about 30% degradation or more.

    In other words, is the GTX750 Windows bug also present on Maxwell 5.2?

    And, is there, by any chance, any user of a GTX750 4GB card? Yes, they do exist...
  • crsminercrsminer Member Posts: 75
    @Genoil I have a GTX 960 2GB, an I see no difference between benchmark and mining.

    In fact I posted my findings regarding difference between benchmark and mining some pages back.
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @crsminer but you're on Linux right?
  • crsminercrsminer Member Posts: 75
  • visuallausivvisuallausiv Member Posts: 5
    Hi. I tried to use --list-devices and the end results were.

    C:\Users\Admin\Desktop\ethminer-cuda-0.9.41-new2>ethminer --list-devices

    C:\Users\Admin\Desktop\ethminer-cuda-0.9.41-new2>pause
    Press any key to continue . . .

    I need to find the cpu.

    I found how to assign a gpu or gpus. --cuda-devices 0 1 2 etc.

    But then again I still need to find gpu devices as I have onboard and pciex16gen3 gfx card.

    Now I need to assign the cpu as well if possible.

    Any suggestions?
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @visuallausiv you have to choose a mining mode (CPU, OpenCL or CUDA) first, otherwise it won't list anything
  • MrYukonCMrYukonC Member Posts: 627 ✭✭✭
    Genoil said:

    @visuallausiv you have to choose a mining mode (CPU, OpenCL or CUDA) first, otherwise it won't list anything

    @Genoil Is that specific to your fork? Because that's not how the stock ethminer works.
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @MrYukonC could be, I'm a bit behind. I'll have a look when I find the time for it.
  • EfterlysTEfterlysT Member Posts: 99
    edited September 2015
    How do i get ethminer to show the hashrate ?


    I also having alot of -32003 errors on all my rigs, not sure what that mean or why i get them.
    I am mining at sprnova
    Post edited by EfterlysT on
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    edited September 2015
    Attention Linux users using GTX750(Ti). During my research into why the hashrate on GTX750Ti / Windows is so bad, I also looked into the behavior on Linux. Below is graphic of the memory bandwidth when running a kernel that mimics the dagger algorhythm. It's not exactly the same, but the memory access pattern is very similar. On the X-axis you see the size of the DAG file in MB, on Y-axis the memory throughput in GB/s.



    If you are using GTX750Ti on Linux and already observe a slight decrease in hashrate, you'll know why...

    I need to do some more research into if this can be avoided in any way. BTW this was also tested at a GTX960 and there the problem was not apparent.
  • davidrentaodavidrentao Member Posts: 60
    thanks Genoil, let us know what problem from 750ti mining, great job...
  • numisnumis Member Posts: 2
    EfterlysT said:

    -32003 error:


    Me too, Did you solve the problem?
  • DavBEDavBE Member Posts: 26
    My miner worked great for a long time but I'm now getting this error :
    ./ethminer: symbol lookup error: ./ethminer: undefined symbol: _ZN3dev3eth15EthashCUDAMiner14s_numInstancesE

    Could someone help me resolv this ?
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @DavBE which build (commit hash) are you using?
  • DavBEDavBE Member Posts: 26
    @Genoil : 76108834965f5b8d398d9ba00e400779eb5c4c0c

    It used to work. Some thing must have changed on mysystem, I'm running Ubuntu 15.04.
  • DavBEDavBE Member Posts: 26
    @Genoil : I found out what the problem was : I moved the directory to another location. I now moved it back to where it was originally and it seems ok.
  • skunkskunk Member Posts: 13
    edited October 2015
    sot173 said:

    Hello and thank you for the great work on this project, using the latest binaries on Win 8.1 Pro my GTX-760 is hashing at 9.2MH/s.

    However, I am still having trouble to compile under OSX 10.9.5, CUDA 7.0, getting the following error:

    Scanning dependencies of target ethash-cl
    [ 25%] Building CXX object libethash-cl/CMakeFiles/ethash-cl.dir/ethash_cl_miner.cpp.o
    nvcc fatal : redefinition of argument 'std'
    CMake Error at ethash-cuda_generated_ethash_cuda_miner_kernel.cu.o.cmake:207 (message):
    Error generating
    /Volumes/OSX-STORAGE/Users/honeydive/Applications/Ethereum/cpp-ethereum-cudaminer-frontier-3/libethash-cuda/CMakeFiles/ethash-cuda.dir//./ethash-cuda_generated_ethash_cuda_miner_kernel.cu.o


    make[2]: *** [libethash-cuda/CMakeFiles/ethash-cuda.dir/ethash-cuda_generated_ethash_cuda_miner_kernel.cu.o] Error 1
    make[1]: *** [libethash-cuda/CMakeFiles/ethash-cuda.dir/all] Error 2

    Any help will be much appreciated.

    got the very same issue (but under linux) after upgrading to cmake-3.3.1, downgrading to cmake-3.2.2 fixed it...
    can't tell if it's a cmake or a makefile bug though...
  • TheFiendishOneTheFiendishOne Member Posts: 18
    I've only noticed the -32003 error if and when I have a browser open or certain software. If I close those it goes away, however my hashrate still falls under 9 but averages at about 8.7-8.8
  • davidrentaodavidrentao Member Posts: 60
    @Genoil, my 6 cards rig hash going down 51M to 45M, what happen with my 750ti? i never change any setting and hardware. wish you reply, thx..



    002.png 45.2K
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    edited November 2015
    @davidrentao time to move on to another algo for your GTX750Ti rig I'm afraid. I warned about this on Sept 28 already :). This is essentially the same issue that has plagued the GTX750Ti on Windows from block 30,000 onwards, but now it's Linux' turn. The problem is that NVidia cards can't handle random accesses to large (1GB+ on Maxwell 1, 2GB+ on Kepler and Maxwell 2) memory allocations very well. The TLB (translation lookaside buffer) that translates virtual addresses into physical ones gets completely filled up, causing page table trashing, leading to a significant bandwidth drop. See the graph I posted a few months back on sept 28.

    You can verify this by running the benchmark (-M flag), it should give you the full ~8.7 MH/s. This is because it uses an 1GB DAG instead of 1.something it is currently.

    There may be a fix for this in the future with a driver update, but it might just be hardware limits.
  • davidrentaodavidrentao Member Posts: 60
    edited November 2015
    @Genoil, I am come back my space last week,i went to holiday last month. so i don using more time to mange my rigs long time ago, what algo you running now of your 750ti? i am not sure other algo make profit in this period. haha. can you propose to me something?
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    i don't mine. but you could have a look at some quark and lyra2v2 coins. very well optimized for Nvidia with private kernels.
  • davidrentaodavidrentao Member Posts: 60
    Genoil said:

    i don't mine. but you could have a look at some quark and lyra2v2 coins. very well optimized for Nvidia with private kernels.

    I don have private kenels to 750TI, do you got that?
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    No but there are quite a few devs (djm34, sp_ to name a few) on bitcointalk that do. Not for ETH as far as I know, but for most other algo's there are.
  • davidrentaodavidrentao Member Posts: 60
    Genoil said:

    No but there are quite a few devs (djm34, sp_ to name a few) on bitcointalk that do. Not for ETH as far as I know, but for most other algo's there are.

    i know the people, thx very much!
  • crsminercrsminer Member Posts: 75
    numis said:

    EfterlysT said:

    -32003 error:


    Me too, Did you solve the problem?
    Any fix for 32003 error?
    Thanks!
  • vigmarvigmar Member Posts: 114 ✭✭
    Win10, GTX670

    result:

    Found suitable CUDA device [GeForce GTX 670] with 2147483648 bytes of GPU memory
    Benchmarking on platform: CUDA
    Preparing DAG...
    Warming up...
    i 20:57:09|cudaminer0 workLoop 0 #00000000тАж #00000000тАж
    i 20:57:09|cudaminer0 Initialising miner...
    Using device: GeForce GTX 670(3.0)
    Trial 1... 12661852
    Trial 2... 12661852
    Trial 3... 12757674
    Trial 4... 12574528
    Trial 5... 12653422
    min/mean/max: 12574528/12661865/12757674 H/s
    inner mean: 12690982 H/s
    _________________________________________________________

    but at the time of mining in the pool shows:

    Mining on PoWhash #295eacf7тАж : 7363595 H/s = 5242880 hashes / 0.712 s
    Mining on PoWhash #295eacf7тАж : 6562714 H/s = 4718592 hashes / 0.719 s
    Mining on PoWhash #295eacf7тАж : 7133170 H/s = 5242880 hashes / 0.735 s
    Mining on PoWhash #295eacf7тАж : 6446163 H/s = 4718592 hashes / 0.732 s
    Mining on PoWhash #295eacf7тАж : 7123478 H/s = 5242880 hashes / 0.736 s
    Mining on PoWhash #295eacf7тАж : 7201758 H/s = 5242880 hashes / 0.728 s
    Mining on PoWhash #295eacf7тАж : 6393756 H/s = 4718592 hashes / 0.738 s
    Mining on PoWhash #295eacf7тАж : 7211664 H/s = 5242880 hashes / 0.727 s
    ---------------------------------------------------

    it should be?
    any ideas?
Sign In or Register to comment.