CUDA miner

1161719212269

Comments

  • DavBEDavBE Member Posts: 26
    Hello,

    I'm registering/signing in to report :

    My rig (6 x MSI Geforce GTX 750 Ti) is doing 51Mh/s constantly for about 3 days thanks to genoil. I'm running Ubuntu 15.04.

    I read the whole thread and noticed someone said he couldn't compile with 4Gb of RAM (don't remember who and I'm not feeling like browsing 18 pages to find out). I have only 4Gb of RAM and I could compile ethminer without any issue.

    Regards,

    David
  • EfterlysTEfterlysT Member Posts: 99
    I dont know if i am blind or something but i dont see a download link to this Cuda miner ? I am trying to get some 750Ti cards up mining ETH on win 8.1 machines
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @EfterlysT if you want to mine ether with GTX750Ti, you have to use Linux. Windows works but has WDDM driver issues causing low hash rate.
  • davidrentaodavidrentao Member Posts: 60
    edited September 2015
    i thought you are better mining on Liunx and using ssh to control miner. local hash around 8.7M, i test watt mining ETH in afternoon, it's more higher than QUARK mining,ETC mining higher around 10% watt to using at same Freq.

    i got some problem is 750ti connect to pool. pool monitor cannot show full hash on the website table, it's awaly show "0" hash at sometime. i have other AMD card mining, it's different with 750ti miner. problem come from liunx' ethminer? or connect line? Genoil any good idea on it?
  • EfterlysTEfterlysT Member Posts: 99
    Someone posted here in this thread that he ran a 750Ti @ 10Mhs in Windows 7. Is that low ?
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @EfterlysT it's unlikely that such speeds can be achieved on win/gtx750 after block 30,000.
  • EfterlysTEfterlysT Member Posts: 99
    @Genoil Ah ok. well i would be glad to get 8Mhs from my 750Ti:s. Anyway i am going to try to get Ubuntu running, i never used Linux before in my life so that will be interesting :)
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @davidrentao what build/version of ethminer CUDA are you running?
  • kenshirothefistkenshirothefist Member Posts: 73
    Anybody had a chance to test Nvidia Tesla K40 GPUs? What is expected hashrate?
  • EfterlysTEfterlysT Member Posts: 99
    I have some really stupid questions.
    Do i have to install ethereum + get if i am going for pool mining at sprnova ?
    Is the cudaminer a standalone miner or do i have to have geth running in the background ?

    I just installed Ubuntu on a rig and just sitting here feeling totally lost. It would be great if someone could point me in the right direction :)
  • GeodesixGeodesix Member Posts: 3
    Genoil said:

    @sot173 hmm mac..totally out of my comfort zone. Regarding your GTX760 (Compute 3.0), I think you should try to compile the cudaminer-uint64 branch for Windows. It may be a lot faster. The reason I'm saying this is because earlier this week, I found that that branch ran a lot faster than the frontier branch on another Compute 3.0 device (Grid K520 found on Amazon AWS G2 instances). They share the same GK104 core.

    @Genoil As you mentioned above, cudaminer-uint64 is the branch preferred for use with Grid k520 devices. I can't for the life of me figure out how to install that specific branch in a AWS G2 instance. Is it only available in windows format or is it possible to install it to an Ubuntu AWS G2...? I've installed the master and had it running with no issues but I'd like to figure out the differences in hash rates between the master and the uint64 branch if possible. Thanks...
  • PhantomPhantom Member Posts: 46
    @Geodesix: my GTX660 runs fine with Ethminer in master branch. It does about 8.3 Mh/s

    Then, I compile a new ethminer in cudamine-unit64. My GTX960 can do about 9.5 Mhs.
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    edited September 2015
    @Geodesix the AWS Grid K520 should be able to do 11.5MH/s (each). I'm not very good at remembering git on the command line, but I think it was git checkout origin/cudaminer-uint64. Of course if you already have 11.5 on the master branch, it's not needed any longer.

    Never tried it on AWS Win, just Ubuntu.
  • GeodesixGeodesix Member Posts: 3
    @Genoil Thank you for the response, I came across a glitch while running on the larger g8. instance yesterday, curious if there would be a way to exploit it. What happened was I was starting each core(?) one at a time with the --opencl-device # command (0 1 2 3 4 5 6 7) and I was able at one point to have 6 instances running concurrently according to "watch nvidia-smi" I noticed that the first 3(0 1 2) were using just over 1gb of ram for each of them but that the 4th(3) was using close to 3.5gb of ram. I get no errors if trying to start a --opencl-device from 0 - 7 but it doesn't work above that. Is it possible that we are only using 4 of 8 cores avail...? Excuse me if this is not the right terminology I'm not too familiar with cuda. I have tried to reproduce the 6 running concurrently but cannot, of note though it did run successfully for approx 16hrs before the spot price spiked and dropped the instance. It reminded me somewhat of a 4 core hyper-threaded cpu having 8 threads. Would it be possible to split it up into 8 X 2gb instances instead of 4 X 4gb ones...? Again sorry if the (cores) is not correct terminology, I know nvidia cards have lots of cuda cores but does one g.8 instance have 4 cores or 4 cards, etc...?
  • GeodesixGeodesix Member Posts: 3
    @Genoil I have just been able to reproduce the result, I grabbed pic, this is with a smaller single g.2 instance. I can run three ethminer instances within a single g.2 instance, If I try 4 it errors out due to only 4gb of memory.


  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    ...but does it give a higher total hash rate?
  • davidrentaodavidrentao Member Posts: 60
    genoil, i use of (from:https://github.com/ethereum/cpp-ethereum/wiki/Building-on-Linux), i test 750ti again in afternoon, i using VPN connect with pool, website report better than before. maybe local internet environment not very good, i put wallet on the VPS running at sometime. any newest version coming out of ethminer?
  • huangyaxuehuangyaxue Member Posts: 3
    edited September 2015
    Xinhai Group founded in 1993 has grown into the industry group involving mineral processing, wearable rubber, mining investment and LFC Machinery

    after twenty years development. We have been always committed to providing customers with professional, perfect, efficient equipment and solutions.

    The whole solution we provide for you with the core of reliable & efficient mineral processing schemes and environmental-protection & energy-saving

    tailings dry stacking, makes us stand up in every EPC bid.
    At Present, Xinhai business includes Mineral Processing EPC, Wear-Resistant Rubber, LFC Machinery, and Mining Investment.
    Xinhai mineral processing EPC provides mine owners with one-stop services including mineral beneficiation test, mine design, etc.. Xinhai is

    committed to solving the problems of mineral processing plant such as budget overspending, schedule delaying, disputes of the manufacturers, etc.
    http://www.xinhaimineral.com/products/submerged-spiral-classifier.html
    www.miningfacture.com
    www.xinhaiminingfacture.com
    www.miningmanufacturer.com
  • kenshirothefistkenshirothefist Member Posts: 73

    Anybody had a chance to test Nvidia Tesla K40 GPUs? What is expected hashrate?

    I just got a chance to test Nvidia Tesla K80 (Keppler based, GK210, CUDA capability 3.7). This is a beast for CUDA memory intensive apps, but obviously ethminer is not able to utilize it fully. K80 has two GPU processors, each is hashing at approx. 15 MHs, so both together for one card is approx. 30 MHs (this is on Win2k12R2) ... not much for a $5k card :smiley:

    @Genoil, do you think that ethminer-cuda could be optimized for those cards since they have 24GB of mem (12GB for each GPU processor)? These cards also have lots of CUDA cores, so maybe code should be explicitly changed to match the layout of blocks/grids K80's. Or there simply isn't room for such optimizations and Dagger algo can't make it much more out of these cards?
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @kenshirothefirst oh interesting. Are you building from source? Then you should maybe add compute 3.7 capability in the CMakelists.txt and play with the __launch_bounds directive for a bit. It has double the register size so it should be able to make some use of that. 15MH/s is too low, but perhaps that;s because the clocks are a bit on the low side vs gaming cards. Or was it just the OpenCL miner that you tried?
  • kenshirothefistkenshirothefist Member Posts: 73
    @Genoil this test was made with CUDA miner from here: http://cryptomining-blog.com/5684-updated-ethminer-0-9-41-nvidia-cuda-version-for-windows-2/

    I also tested OpenCL miner from here: http://cryptomining-blog.com/5612-updated-ethminer-0-9-41-opencl-and-cuda-for-windows/, more or less same speed, a bit slower than CUDA.

    I could try to build from sources ... do you have some step-by-step instructions for building from your source with VS2013? And regarding __launch_bounds ... what values do you suggest?
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @kenshirothefist I don't remember exactly at which point I removed Compute 3.7 support, but I did remove it because I figured people with Tesla K80 have more profitable calculations to run ;). Anyway if it ran, then it probably had support for Compute 3.7, unless 3.7 is backwards compatible with 3.5.

    VS2013 build instructions should be somewhere in this thread.

    With double the register size per SM you could probably achieve a higher occupancy in the dagger stage, but I'm afraid you'd still be limited by memory bandwidth. You could probably take some advantage of it, but it's not easy. If there would be high demand, then I could take a look at it, provided I had access to a K80. But I'm afarid there isn't and there will never be :)
  • DavBEDavBE Member Posts: 26
    I'm a bit lost regarding parameters. Could somebody help me with these ?

    --cl-local-work
    --cl-global-work
    --cuda-turbo
    --cuda-grid-size
    --cuda-block-size
    --cuda-schedule

    I would like to set ideal values for GTX 750 Ti GPUs.

    @Genoil : any help on this ?

    Regards,

    David
  • BripsBrips Member Posts: 52
    Hello, it's seems that the cuda miner give some bad results, so pay attention to that miner if you want to really use it.

    example :
    ℹ 14:33:26|cudaminer0 workLoop 1 #ec9d541d… #ec9d541d…
    ℹ 14:33:27|ethminer Solution found; Submitting to http://oneether.com:8080?miner=0xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx ...
    ℹ 14:33:27|ethminer Nonce: 8d0507fcf33c79e5
    ℹ 14:33:27|ethminer Mixhash: fb32378b1fd6590ab5f5c3305f50ec86fc88acacef3a42a8406219daede0ea01
    ℹ 14:33:27|ethminer Header-hash: 9396ccfe054cd728a312ae49c005ec2e8c0330ea715a6bfc093fd06f386b9825
    ℹ 14:33:27|ethminer Seedhash: ec9d541d1c32bec7debceb8103337e6ad3b1140e80ae7773a1013fd6f3014c46
    ℹ 14:33:27|ethminer Target: 00000002b7fd4fe528cf3c594e2a471559a72d54047280e724ec916d94676ed4
    ℹ 14:33:27|ethminer Ethash: 501a313bc8546f2021e994ceadb3126164c9374375175397df4d8754991c28b6
    ✘ 14:33:27|ethminer FAILURE: GPU gave incorrect result!
    ℹ 14:33:27|ethminer Getting work package...
    ℹ 14:33:27|ethminer Got work package:
    ℹ 14:33:27|ethminer Header-hash: 9396ccfe054cd728a312ae49c005ec2e8c0330ea715a6bfc093fd06f386b9825
    ℹ 14:33:27|ethminer Seedhash: ec9d541d1c32bec7debceb8103337e6ad3b1140e80ae7773a1013fd6f3014c46
    ℹ 14:33:27|ethminer Target: 00000002b7fd4fe528cf3c594e2a471559a72d54047280e724ec916d94676ed4
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @Brips I've found two references on this board of opencl miner users that post the same error message, so I tend to beleive this is not related to the CUDA miner. But yes, it happens from time to time. My guess is that the GPU thread has come up with a solution while the farming thread has already moved on to the next block.
  • RoBiKRoBiK Member Posts: 6
    I have already filed a bug two weeks ago about this: https://github.com/ethereum/cpp-ethereum/issues/2967
  • BripsBrips Member Posts: 52
    @Genoil ok ! thanks, i will return to cuda miner so ;)
  • ol92ol92 Member Posts: 4
    @Genoil
    Your miner is very good on GTX 750 ti but performance on a GTX 980 ti is about the same as a GTX 980 (22mh/s-23Mh/s) even with an overclocked memory (3960 mhz in P2 state) and gpu core (1400mhz). This gpu has similar bandwith than a R9 290X and should have a better hashrate : do you have some idea how to improve this figures ?
    Thanks
  • kenshirothefistkenshirothefist Member Posts: 73

    I just got a chance to test Nvidia Tesla K80 (Keppler based, GK210, CUDA capability 3.7). This is a beast for CUDA memory intensive apps, but obviously ethminer is not able to utilize it fully. K80 has two GPU processors, each is hashing at approx. 15 MHs, so both together for one card is approx. 30 MHs (this is on Win2k12R2) ... not much for a $5k card :smiley:

    Just FYI: Also tried Ubuntu, manually compiled for compute 3.7, same speed.

    BTW: this is also a proof, that http://ethercloud.info/ is total scam since they can never fulfill "business" and "enterprise" packages with K40 and K80 with these speed and prices ...

    OK, enough playing with Ethereum on K80, lets go do something more usefull ;)
  • MrYukonCMrYukonC Member Posts: 627 ✭✭✭
    Brips said:

    Hello, it's seems that the cuda miner give some bad results, so pay attention to that miner if you want to really use it.

    example :
    ℹ 14:33:26|cudaminer0 workLoop 1 #ec9d541d… #ec9d541d…
    ℹ 14:33:27|ethminer Solution found; Submitting to http://oneether.com:8080?miner=0xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx ...
    ℹ 14:33:27|ethminer Nonce: 8d0507fcf33c79e5
    ℹ 14:33:27|ethminer Mixhash: fb32378b1fd6590ab5f5c3305f50ec86fc88acacef3a42a8406219daede0ea01
    ℹ 14:33:27|ethminer Header-hash: 9396ccfe054cd728a312ae49c005ec2e8c0330ea715a6bfc093fd06f386b9825
    ℹ 14:33:27|ethminer Seedhash: ec9d541d1c32bec7debceb8103337e6ad3b1140e80ae7773a1013fd6f3014c46
    ℹ 14:33:27|ethminer Target: 00000002b7fd4fe528cf3c594e2a471559a72d54047280e724ec916d94676ed4
    ℹ 14:33:27|ethminer Ethash: 501a313bc8546f2021e994ceadb3126164c9374375175397df4d8754991c28b6
    ✘ 14:33:27|ethminer FAILURE: GPU gave incorrect result!
    ℹ 14:33:27|ethminer Getting work package...
    ℹ 14:33:27|ethminer Got work package:
    ℹ 14:33:27|ethminer Header-hash: 9396ccfe054cd728a312ae49c005ec2e8c0330ea715a6bfc093fd06f386b9825
    ℹ 14:33:27|ethminer Seedhash: ec9d541d1c32bec7debceb8103337e6ad3b1140e80ae7773a1013fd6f3014c46
    ℹ 14:33:27|ethminer Target: 00000002b7fd4fe528cf3c594e2a471559a72d54047280e724ec916d94676ed4

    This is a very common error. It happens just as frequently on AMD cards. I don't think it has anything at all to do with the hardware or the CUDA/OpenCL compute implementation, but rather the interaction with the pool.
Sign In or Register to comment.