CUDA miner

1313234363769

Comments

  • CubirezCubirez Member Posts: 71
    I can see another difference with the cmake options that have changed with the lastest ethminer versions.

    @Marvell9 : there's an updated guide by Phantom here https://forum.ethereum.org/discussion/comment/14274/#Comment_14274
    I hope this can help you.

    However i'm stuck myself while compiling with this (using xubuntu 14.04) :'(
    [ 68%] Building NVCC (Device) object libethash-cuda/CMakeFiles/ethash-cuda.dir/ethash-cuda_generated_ethash_cuda_miner_kernel.cu.o
    /usr/lib/gcc/x86_64-linux-gnu/4.8/include/stddef.h(432): error: identifier "nullptr" is undefined
    
    /usr/lib/gcc/x86_64-linux-gnu/4.8/include/stddef.h(432): error: expected a ";"

  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    edited March 2016
    @Cubirez yes ran into that myself too. This was caused by somebody who tried to make copatibility with Fedora work. I'll fix it in the rep later, but for now, add --std=c++11 to the NVCC flags in CMakelists.txt in libethash-cuda folder
  • GodmodeGodmode AustriaMember Posts: 12
    Hi guys!

    I have a Titan X, running at 1,5 GHz core and 4,0 GHz memory. If I run the benchmark (-M option), I get about 23 MHash/s. The odd thing is, if I start mining, I only get about 7 MHash/s. Is this normal behaviour, I'm new to ETH mining.

    System: [email protected],5 GHz, 16 GB DDR4 2666, Titan X @ 1,5 GHz, ASUS Rampage V Extreme all water cooled, Win 10 x64, Geforce driver 364.47, 0.9.41-genoil-1.0.4b3 for Windows

  • CubirezCubirez Member Posts: 71
    edited March 2016
    @Genoil Thanks this made it work ! Getting 113MHs out of my 6 GTX970 rig. Now it's time to start playing around with the settings to get a few MHs more :)

    I also updated the guide Phantom's guide with some informations i gathered on the thread/ the latest packages / your tip.

    Edit : reloaded my docx file, first one was corrupt
    Post edited by Cubirez on
  • CubirezCubirez Member Posts: 71
    By the way i managed to compile without a problem with only 4GB RAM.
    Did maybe take 3 to 4 minutes though
  • GodmodeGodmode AustriaMember Posts: 12
    edited March 2016
    Godmode said:

    Hi guys!

    I have a Titan X, running at 1,5 GHz core and 4,0 GHz memory. If I run the benchmark (-M option), I get about 23 MHash/s. The odd thing is, if I start mining, I only get about 7 MHash/s. Is this normal behaviour, I'm new to ETH mining.

    System: [email protected],5 GHz, 16 GB DDR4 2666, Titan X @ 1,5 GHz, ASUS Rampage V Extreme all water cooled, Win 10 x64, Geforce driver 364.47, 0.9.41-genoil-1.0.4b3 for Windows

    Ok, I have managed to solve the problem. I've installed Geforce driver version 347.88, because this driver uses WDDM 1.0. With this driver I get my 23MHash/s at real mining in Windows 10. :)
  • Marvell9Marvell9 Member Posts: 593 ✭✭✭
    Cubirez said:

    @Genoil Thanks this made it work ! Getting 113MHs out of my 6 GTX970 rig. Now it's time to start playing around with the settings to get a few MHs more :)

    I also updated the guide Phantom's guide with some informations i gathered on the thread/ the latest packages / your tip.


    I cant open this file can you post as a zip or something

    I also cant download this Nivdia package

    wget http://developer.nvidia.com/compute/cuda/7.5/RC/local_installers/cuda-repo-ubuntu1404-7-5-rc_7.5-7_amd64.deb

    keep getting a 404

  • CubirezCubirez Member Posts: 71
    Marvell9 said:

    Cubirez said:

    @Genoil Thanks this made it work ! Getting 113MHs out of my 6 GTX970 rig. Now it's time to start playing around with the settings to get a few MHs more :)

    I also updated the guide Phantom's guide with some informations i gathered on the thread/ the latest packages / your tip.


    I cant open this file can you post as a zip or something

    I also cant download this Nivdia package

    wget http://developer.nvidia.com/compute/cuda/7.5/RC/local_installers/cuda-repo-ubuntu1404-7-5-rc_7.5-7_amd64.deb

    keep getting a 404

    @Marvell9 My bad the first file i uploaded was corrupted. I edited my post, you should be able to open it now.

    As for your specific problem, the link you use isn't good anymore, try this one :

    wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_7.5-18_amd64.deb
  • Marvell9Marvell9 Member Posts: 593 ✭✭✭
    Cubirez said:

    Marvell9 said:

    Cubirez said:

    @Genoil Thanks this made it work ! Getting 113MHs out of my 6 GTX970 rig. Now it's time to start playing around with the settings to get a few MHs more :)

    I also updated the guide Phantom's guide with some informations i gathered on the thread/ the latest packages / your tip.


    I cant open this file can you post as a zip or something

    I also cant download this Nivdia package

    wget http://developer.nvidia.com/compute/cuda/7.5/RC/local_installers/cuda-repo-ubuntu1404-7-5-rc_7.5-7_amd64.deb

    keep getting a 404

    @Marvell9 My bad the first file i uploaded was corrupted. I edited my post, you should be able to open it now.

    As for your specific problem, the link you use isn't good anymore, try this one :

    wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_7.5-18_amd64.deb
    @Cubirez thanks a lot man I was able to get further by a lot now im trying to build the branched version of ethminer but I'm an Linux and cmake noob what should this line

    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS};--disable-warnings;--ptxas-options=-v;-use_fast_math;-lineinfo)

    look like after I apply this change -


    and add --std=c++11 to the NVCC flags (on line 8)

    this is like my sixth install of Ubuntu need to get it right this time lol
  • CubirezCubirez Member Posts: 71
    @Marvell9 it should loook like this :
    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS};--std=c++11;--disable-warnings;--ptxas-options=-v;-use_fast_math;-lineinfo)


  • Marvell9Marvell9 Member Posts: 593 ✭✭✭
    @Cubirez awesome once I get it running I'll send you and genoil some eth
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    Cubirez said:

    @Marvell9 it should loook like this :
    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS};--std=c++11;--disable-warnings;--ptxas-options=-v;-use_fast_math;-lineinfo)


    this is now rolled back. sorry for the inconvenience.
  • sikuntsikunt Member Posts: 13
    Godmode said:

    Godmode said:

    Hi guys!

    I have a Titan X, running at 1,5 GHz core and 4,0 GHz memory. If I run the benchmark (-M option), I get about 23 MHash/s. The odd thing is, if I start mining, I only get about 7 MHash/s. Is this normal behaviour, I'm new to ETH mining.

    System: [email protected],5 GHz, 16 GB DDR4 2666, Titan X @ 1,5 GHz, ASUS Rampage V Extreme all water cooled, Win 10 x64, Geforce driver 364.47, 0.9.41-genoil-1.0.4b3 for Windows

    Ok, I have managed to solve the problem. I've installed Geforce driver version 347.88, because this driver uses WDDM 1.0. With this driver I get my 23MHash/s at real mining in Windows 10. :)
    i tried downloading thee driver now when i start the miner i get "insufficient cuda runtime version"
  • CubirezCubirez Member Posts: 71
    @Marvell9 I hope it will work for you. I had a hard time getting it working as well. Now everything runs smoothly and at 127MHs for the 6 GTX970 after some tweaking. Btw save your donation for Genoil only. I didn't do anything but changing a few lines in the guide I found in this thread ;)
  • sikuntsikunt Member Posts: 13
    sikunt said:

    Godmode said:

    Godmode said:

    Hi guys!

    I have a Titan X, running at 1,5 GHz core and 4,0 GHz memory. If I run the benchmark (-M option), I get about 23 MHash/s. The odd thing is, if I start mining, I only get about 7 MHash/s. Is this normal behaviour, I'm new to ETH mining.

    System: [email protected],5 GHz, 16 GB DDR4 2666, Titan X @ 1,5 GHz, ASUS Rampage V Extreme all water cooled, Win 10 x64, Geforce driver 364.47, 0.9.41-genoil-1.0.4b3 for Windows

    Ok, I have managed to solve the problem. I've installed Geforce driver version 347.88, because this driver uses WDDM 1.0. With this driver I get my 23MHash/s at real mining in Windows 10. :)
    i tried downloading thee driver now when i start the miner i get "insufficient cuda runtime version"
    Also I'm using 3 Titans but only getting a total of 16 to 20mhs when I benchmark it shows about 45mh
  • Marvell9Marvell9 Member Posts: 593 ✭✭✭
    Cubirez said:

    @Marvell9 I hope it will work for you. I had a hard time getting it working as well. Now everything runs smoothly and at 127MHs for the 6 GTX970 after some tweaking. Btw save your donation for Genoil only. I didn't do anything but changing a few lines in the guide I found in this thread ;)

    @Cubirez thanks again you were right the instructions worked like a charm my six card box now gets around 125mhs at the pool using the SMI setting to set max clocks.

    The cards a bit more power hungry than I though pulling a total of 800watts or so for the whole machine at the wall.

    Ok sounds good on the donation I'll send it to Genoil , once I get some coins mined.

    If onl these cards were not so expensive I really like the SMI stuff and cuda in genarl vs AMD. I also like how the cards still run cool even though they are working wuite hard.
  • Marvell9Marvell9 Member Posts: 593 ✭✭✭
    Genoil said:

    Cubirez said:

    @Marvell9 it should loook like this :
    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS};--std=c++11;--disable-warnings;--ptxas-options=-v;-use_fast_math;-lineinfo)


    this is now rolled back. sorry for the inconvenience.
    @Genoil no problem at all I got it to work with @Cubirez help
  • blueboxbluebox Member Posts: 181 ✭✭
    @Marvell9 Playing around with mem o/c it appears that for a 10% increase in clock rate from base (3GHz), power consumption goes up 20%, more when you approach the max (4GHz). Those are rough observations, so take them for what their worth, but I start seeing nonce errors at 3.8GHz.
  • Marvell9Marvell9 Member Posts: 593 ✭✭✭
    @bluebox yeah but all i did was set the maximum mem and clock states i.e perforamce mode

    I'm not overclocking over the stock 3505 mem and around 1455 clock for all the cards.

    how do i raise the memmory speed with SMI ? seems i cant go over the 3505

    I did find that looking at monitoring my strix cards are pulling close to 170 watts each so i used
    nvidia-smi -pl 150 to cap them all at 150 and the hash rate has stayed the same I just dont get that.

    power is capping now at 155 on all the cards and hash rate has not decreased much if any, I'm thinking I can use a lower power state of maybe 140 even Ill watch things for 24 hours first though. Still no blocks mined tho :(




  • eddy4eddy4 Member Posts: 30
    Hello!

    I've read as much as possible but after one week I have no idea how to increase the hash rate for my two 750Tis any further.

    Hardware:

    Nvidia 750Ti x2
    AMD Sempron 145 Single Core
    8 GB RAM
    32 GB SSD

    First try:

    Xubuntu 15.10
    CUDA 7.5
    Nvidia ?

    gcc 5.2
    cmake 3.4

    git clone https://github.com/Genoil/cpp-ethereum

    cmake -DBUNDLE=cudaminer -DCOMPUTE=50

    make

    ./ethminer --farm-recheck 200 -U -F http://127.0.0.1:8080/rig1 --cuda-devices 0 1

    = 5.1 MH/s x2


    Second try:


    Xubuntu 14.04.4
    CUDA 6.5
    Nvidia 352

    gcc 4.8.4
    cmake 3.2.2 = highest ppa version

    git clone https://github.com/Genoil/cpp-ethereum

    cmake -DBUNDLE=cudaminer -DCOMPUTE=50

    make

    ./ethminer --farm-recheck 200 -U -F http://127.0.0.1:8080/rig1 --cuda-devices 0 1

    = 5.1 MH/s x2


    Any ideas how I can get about 9 MH/s for each 750Ti?
  • ekqtqvwekqtqvw Member Posts: 13
    Hello,
    what's best ethminer and setting for Geforce GTX960 (windows 7 unlimited )?
  • blueboxbluebox Member Posts: 181 ✭✭
    @Marvell9 First set the P0 state as you've done with nvidia-smi -ac, then run something like msi afterburner (follow their directions for setting it up first) and you can set the memclock offset, e.g. +200 to equal 3705MHz.

    Are you sure the -pl flag is actually changing anything on the card? Check the wattage at the wall, hard to believe you can just drop the cap and have it operate the same hash rate without effect... All I know is with the old cuda6 version, -pl only worked on Kepler gpgpu compute and quadro cards, not consumer/gamer models.
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    edited March 2016
    @eddy4 this is the best you're going to get. 750Ti's are heavily affected by growing DAG size. Actually using Linux is by far the best option, on Windows it is much worse. 5.1MH/s is not bad at all for this card.

    @isamamin my cuda miner, any 1.0.x release will do. Default settings are fine. Launch with -U flasg instead of -G.

    btw still fighting with Compute 3.0 support. Got to 12MH/s on Grid K520, but still some nasty bug loitering around causing no valid shares...
    Post edited by Genoil on
  • eddy4eddy4 Member Posts: 30
    edited March 2016
    @Genoil First of all thank you very much for your awesome work!

    Someone over there gets at the moment 8.3 MH/s.

    https://www.reddit.com/r/EtherMining/comments/49fs6o/whats_your_max_hash_rate_with_a_nvidia_750ti/d0ugt03

    The single core AMD Sempron 145 CPU isn't a problem, right?

    Can I change something before cmake or by changing the cmake command?
  • eddy4eddy4 Member Posts: 30
    Marvell9 said:


    cmake -DBUNDLE=miner -DETHASHCU=1
    ./ethminer -U -F http://(yourpool) --gpu-devices 0 1 2

    If I'm not wrong because I've installed Ubuntu 14.04 with ethminer just hours ago, you can't use "-U" and "cmake -DBUNDLE=miner". Use "-G" with "cmake -DBUNDLE=miner" or "-U --cuda-devices 0 1 2" (for 3 GPUs) with "cmake -DBUNDLE=cudaminer".
  • Marvell9Marvell9 Member Posts: 593 ✭✭✭
    bluebox said:

    @Marvell9 First set the P0 state as you've done with nvidia-smi -ac, then run something like msi afterburner (follow their directions for setting it up first) and you can set the memclock offset, e.g. +200 to equal 3705MHz.

    Are you sure the -pl flag is actually changing anything on the card? Check the wattage at the wall, hard to believe you can just drop the cap and have it operate the same hash rate without effect... All I know is with the old cuda6 version, -pl only worked on Kepler gpgpu compute and quadro cards, not consumer/gamer models.

    You make too much sennse man stop it lol yeah at the wall i dont see any huge difference but
    I use a metered PDU which rates in 240 watt increments so a small drop of 10-20 watt per six gpus might not register unless it drops the wattage down by over 240.


    the thing is though the temps have dropped on the monitor , they used to be 76-78 now i see temps from
    69-72 so it must be doing something.

    MSI afterburner has a linux version ? I had no idea
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @eddy4 perhaps they get 8-9MH/s on the benchmark. That's using a 1GB DAG. Current DAG is ~1.3GB. It really does affect the hashrate severely, esp on GTX750ti.

    I'm parking the compute 3.0 support for now. Can't debug properly on AWS Ubuntu, plus I hate paying by the hour. I'll see if I can get my hands on a Compute 3.0 card next week to figure out what's going wrong.
  • eddy4eddy4 Member Posts: 30
    edited March 2016
    @Genoil Are the 2 GB RAM of the 750Ti a problem or something different? If the RAM is the problem, would you suggest instead the AMD Radeon 7950 with 3 GB RAM to keep mining till PoS?
  • blueboxbluebox Member Posts: 181 ✭✭
    @Marvell9 Sorry, should have realized you were on linux. Afterburner is windows only, one of the few reasons I've (reluctantly) stuck with the shittiest OS on the planet... I don't know a way to boost these game card clocks over P0 state on linux, but I'll look.

    Trust me, if I had the time I'd have converted to Mint, or better yet CentOS 7, which I believe can get all the dependencies now to compile ethminer. RHEL/CentOS is my day job (HPC), so I'm much more comfortable with linux but the gaming world is windows, as are these overclocking tools for gaming cards. :'(
  • sikuntsikunt Member Posts: 13


    this is my benchmark for 3 titans, watercooled.
    does this mean thats the hashrate for each at the end or total?
Sign In or Register to comment.