CUDA miner

1101113151669

Comments

  • dominusdominus Member Posts: 36
    How do You like my 2x 750ti hashrate? B)
    ./ethminer -F http://ethereumpool.co/[email protected]. -U -t 2 --cuda-turbo
    http://postimg.org/image/q8n3uu2g3/
    The pool says Real Current Hashrate: 20066826 hash/s -> 20.07 MHASH/s
    It is the frontier version compiled with cuda 7.5.
    There is no CPU usage at all.
    Tnx genoil ;)

  • RichieRichie United KingdomMember Posts: 10
    Phantom said:

    @Richie : use flag -t n (n = how many GPU you have)

    Sample: ethminer -F http://xxxxxx -U -t 2

    I have 2 GPUs i.e. DUAL-SLI NVIDIA GT755M on the laptop.
  • SuchFakeAccountSuchFakeAccount Member Posts: 14
    edited August 2015
    I have problem compiling the PTX branch on linux. My PC compleetly freezes at this point. Like the mouse even wont move :D
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @SuchFakeAccount do I see correctly in that picture that you have 4GB RAM? Last time I built on Linux, needed 8...
  • SuchFakeAccountSuchFakeAccount Member Posts: 14
    edited August 2015
    Genoil said:

    @SuchFakeAccount do I see correctly in that picture that you have 4GB RAM? Last time I built on Linux, needed 8...

    Hah :blush: :smiley: Yes I have 4GB RAM.

    Thing is that your Cudaminer branch compiles without problems...
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @SuchFakeAccount I don't know what they did to make it so demanding. Cudaminer is based on 0.9.23, cudaminer-frontier and ptx are 0.9.40.

    I might just copy the needed functionality from .40 and plug that into cudaminer.23. Makes miners happier I guess.
  • SuchFakeAccountSuchFakeAccount Member Posts: 14
    edited August 2015
    @Genoli

    My PC is trying to build the frontier branch. Look at it sweat... using 2.8GB o Swap :smiley:



    Edit: after 10 minutes it finished, but with a error because I forgot to remove -werror :smile:

    Edit2: tried the waiting method with PTX. 25minutes later it finished successfully :smile:
    Post edited by SuchFakeAccount on
  • PhantomPhantom Member Posts: 46
    @SuchFakeAccount: I believe that when you build ccminer on Linux, you need the powerful processor. Before I used AMD Sempron 145, it took me about 30 minutes. Then I switched to AMD FX6300, building process was super fast.

    @Genoil : I am done overclock my GTX 750 ti. All my gtx 750 ti rigs are stable ~ 58 Mh/s. However, I got problem to overclock my GTX 960. GTX 960 has 4 performance level (0 to 3), but I can't jump to level 3 to apply my tune. I am still researching.

  • antonio8antonio8 Member Posts: 44
    Have not posted in a while but decided to fire up one rig that dual boots Windows 7/10 with 750ti's.

    This was the first time going onto Windows 10 and I don't have the problem where it locks my system up, cpu 100%, no matter what miner I used. Right now just very minimum cpu usage and I can still do other things with it.

    Unfortunately though the 750ti's only mine at about 1.8-2.0 Mh/s so it is still not worth it to mine on the rig but at least I know now that it is a Windows 7 issue that I am having.
  • PhantomPhantom Member Posts: 46
    @antonio8 : did u try on Ubuntu? Before, I mined on Windows, but I got a lot of issues with ethminer. So, I switched to Ubuntu, both QUARK and ETH mining were improved a lot: higher hashrate and rig running stable 24/7.

    The only problem I have now is that overclock GTX 960.
  • antonio8antonio8 Member Posts: 44
    Phantom said:

    @antonio8 : did u try on Ubuntu? Before, I mined on Windows, but I got a lot of issues with ethminer. So, I switched to Ubuntu, both QUARK and ETH mining were improved a lot: higher hashrate and rig running stable 24/7.

    The only problem I have now is that overclock GTX 960.

    Unfortunately I have never ran any version of Linux/Ubuntu.
  • PhantomPhantom Member Posts: 46
    @antonio8 : you should try it. Dual boot Linux/Windows is not a bad idea. If you need any helps, I will try.
  • dominusdominus Member Posts: 36
    It is frustrating... I get more from an 750ti then 960 :)
    2x 750ti producing 21.35-21.5Mh(pool accepted): http://s13.postimg.org/4wkzf7bzb/snapshot2.png
    The cards are not modded in any way. It is only gpu clock and memory oc, it is working for more than 11 hours now...
    One of those is a reference one without the additional pcie power.
    B)
  • PhantomPhantom Member Posts: 46
    @dominus: that's why those cards draw a lot of power from the wall and they were very hot. I guess that you put those cards into your ATX pc case, not a opened case. You should lower gpu and memory clock to keep temp lower or use a fan. The rig will run 24/7, so keep it cool. You'll save a lot later on.
  • SuchFakeAccountSuchFakeAccount Member Posts: 14
    Phantom said:

    @SuchFakeAccount: I believe that when you build ccminer on Linux, you need the powerful processor. Before I used AMD Sempron 145, it took me about 30 minutes. Then I switched to AMD FX6300, building process was super fast.

    The old cudaminer builds very fast on my PC too (Not more than 3min). Problem is with the newest version that it needs more than 4GB of RAM. When the RAM fills up then computer will start using SWAP (HDD or SSD, in my case it is HDD) which is slow AF :smile: and the system basically freezes up because of the HDD slowness. But the build will eventually finish if the computer has enough SWAP.

  • dominusdominus Member Posts: 36
    Phantom said:

    @dominus: that's why those cards draw a lot of power from the wall and they were very hot. I guess that you put those cards into your ATX pc case, not a opened case. You should lower gpu and memory clock to keep temp lower or use a fan. The rig will run 24/7, so keep it cool. You'll save a lot later on.

    The cards are not hot at all. 63° and 68° C(ambient temp.=28°C) is very low for maxwell cards. The power draw is also not an issue. The reference card cannot draw more than 75W and the other not much more. Both are sitting outside of the case.
  • jakobitojakobito Member Posts: 9
    Successfully compiled PTX branch on lubuntu 15.04 with CUDA 7.5.
    Compiler found a problem with an unused variable "int targetBlocktime = 15;" found in ./ethminer/MinerAux.h
    Since I'm really newbie in programming, I just tried commenting it and it worked!
    3 x GTX750Ti > around 28-30Mhs without any OC.
    Many thanks @Genoil
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @jakobito what's the difference in ahshrate between -G and -U flags?
    (the unused int should indeed be gone)
  • PhantomPhantom Member Posts: 46
    I know that any cards could run at that temp, but it is not good for long term. My rig with 6 x GTX 750 ti which sit very tight in a custom case, the temp is never go above 60°C. I have been in this game long enough to understand the issue. Back to AMD Radeon HD 7950, R9 280X or R9 290/290X, overclock max with 100% fan. Then, the fans were broken, cards couldn't boot up or showed the blue screen...I sold them all before too late. Then, I build my rigs with gtx 750 ti (only get 2 fans), always run at stock clock in Summer time and lightly overclock in Winter. However, I had to RMA 2 of 40 Gigabyte GTX 750 ti. Trust me, it takes a lot of time to do RMA or you will never get ROI.
  • jakobitojakobito Member Posts: 9
    Im at work atm and I dont remember the exact values. This evening I will post the benchmark results.
    I performed several benchmarks using:
    For OpenCL: --cl-global-work 16384
    For CUDA: --cuda-grid-size 8192 --cuda-block-size 128 --cuda-turbo
    and found higher hashrates using CUDA. I think Im not getting >10Mh per card because I havent OC my cards yet.
  • dominusdominus Member Posts: 36
    edited September 2015
    @Phantom
    I know the risks. And I know also the radeons runs much hotter(could get close to 100°C) ;) But I will always consider safe running a maxwell below 70° with fan below 60%. All Maxwell's I have are temperature limited at 80°C(default throttling). It must work. If it fails it will be rma-ed ;)
    I forgot: The power limit and the voltages are not increased. The cards are not throttling so the power limit is not reached and the cards are not handling any more power as were designed for. It is valid for eth mining only offcourse.
  • PhantomPhantom Member Posts: 46
    The most issues I got is broken fan. Anyway, keep it cold :)
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    Important notice to Compute 3.0 (and lower) miners (GTX 6x0, 760, 770, Grid K520)

    This includes Amazon EC2 instances. Please use the cudaminer-uint64 branch for now. Turns out those cards didn't like the sp_ mods, leading to severe drops in hashrate. Thanks @Mardofuk for pointing this out.

    I'll make sure an upcoming release takes this into account.
  • PhantomPhantom Member Posts: 46
    @Genoil : I try to build on Cuda-opencl-ptx branch, but hashrate goes up and down all the time, not stable as before.

    5 x gtx 750 ti: hashrate jumps from 43-52 Mh/s
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @Phantom is that on CUDA or OpenCL? Try halving --cuda-grid-size/-cl-global-work and see if it is more stable.

    Anyway the opencl-ptx branch is not really meant for production. It's just still a test.

    To sum up current useful branches:

    cudaminer: best for modern GPUs
    cudaminer-frontier: work in progress to merge cudaminer ito the main repo
    cudaminer-uint64: best for compute 3.0 and lower, also works fine on newer gpus

    The rest is not really useful atm.

    I had a closer look to sp_'s keccak optimizations today. turns out that on Compute > 3.0 it also isn't faster than the original. I reverted parts of it, now keccak is almost as fast again as in the cudaminer-uint64 branch.

    Ultimately the speed of keccak doesn't really make a difference, as the bottleneck of the code is in the dagger stage.
  • cidmocidmo Member Posts: 446 ✭✭✭
    edited September 2015
    these ptx are gems
    290x bouncing 26-29Mh

    setx GPU_MAX_ALLOC_PERCENT 100
    setx GPU_USE_SYNC_OBJECTS 1
    setx GPU_MAX_HEAP_SIZE 100
    ethminer-ptx-4 -F http://eth2.suprnova.cc:3000/workername.1/29 -G --cl-global-work 16384

    gpu-1010
    mem-1300
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    @cidmo actually they aren't supposed to be any faster or slower on AMD. Inline PTX assembly is only applied to newer Nvidia cards. But if it works for you, great :).
  • cidmocidmo Member Posts: 446 ✭✭✭
    edited September 2015
    Genoil said:

    @cidmo actually they aren't supposed to be any faster or slower on AMD. Inline PTX assembly is only applied to newer Nvidia cards. But if it works for you, great :).

    haha yea i figured id give em a try and so far its the highest hash ive seen
    more so than cudaminer in cl with the gpu-batch-size string which was 25-26
    tweaking still 8192 is pretty much the same as 16384 atm but goes 27-28.5 lil more stable i suppose
    gonna play some more and install linux later and compile my own to see what i can do

    thx again every extra hash counts
  • GenoilGenoil 0xeb9310b185455f863f526dab3d245809f6854b4dMember Posts: 769 ✭✭✭
    when the hashrate starts bouncing, it's just a measurement error, not actual fluctuation in speed.
  • cidmocidmo Member Posts: 446 ✭✭✭
    edited September 2015
    gotcha so 16384 is better since it hits close to 30
    1 question tho, so im using the same ptx-4 for a 750ti, with the strings u posted -U --cuda-grid-size 8192 --cuda-block-size 128 --cuda-turbo
    any reason the 750ti is using over 1gig of ram and the 290x only uses 82MB?
    each on different rigs both win7

    32768=26-32Mh
    shows up on pool too
Sign In or Register to comment.