<<  >> (p.297)
    Author Topic: [ mining os ] nvoc  (Read 418569 times)
    LuKePicci
    Jr. Member
    *
    Offline Offline

    Activity: 128
    Merit: 1


    View Profile
    February 05, 2018, 10:24:50 PM
     #5921

    Hi again, I have some news to share with you.

    First of all, ethminer is finally adding an amazing "exit on cuda error" feature, I bet somebody here will love this.

    Second, my extensive ethash miners testing revealed something strange with newer versions of all major miners, so I asket ethminer devs to give me some kind of changelog for the new stable release, and they did it.
    Code:
    ethminer 0.13.0
     chfast released this 13 days ago

    Move stale detection to the source, eliminate solution reporting inefficiencies.
    Minor cosmetic console output changes.
    CUDA: Much faster job switch time. Higher stable hash rate.
    CUDA: Refactored host software. Efficiency and simplification.
    OCL: Added support for experimental kernel. Performs better on some cards.
    Dropped support for CUDA 8.0
    Blank API server fields correctly populated.
    Compiler and linker build options optimized.

    So I have to ask here, does the update script also update CUDA to version 9? If no how can I safely do that?

    Last thing, I agreed with the ethminer devs it would be nice to have OC settings reported near the miner crash report, they said it wouldn't be so easy to do so I would ask here if it would be possible to include the output of "nvOC gpuinfo" near the error log snapshot.


    Update script wont install cuda 9 because so many miners still don't have cuda 9 support and it will break them. You can use Nvidia Documentation on how to install Cuda 9.

    I installed cuda 9 on a test rig and many miners start giving all kinds of errors, but I'm mining with ethminer 0.13.0 on my rigs with cuda 8 with no problems.

    I have no idea how come some miners work and some doesn't with different cuda versions.
    It would be nice if someone explain it to us

    Sounds reasonable. Actually I started getting frequent cuda errors as soon as I updated to 0.13.0, I removed any OC and the error frequency drastically reduced but I was still getting errors too often (1,2 times per hour) with both latest claymore and latest ethminer. Meanwhile I noticed the reported OC'd hashrates from the ethminer log increased a little from an average of 179.5 MH/s to something around 182 but apart from the crashes my effective average hashrate reported by ethermine (the pool) with both ethminer and claymore (latest stable version of both) dropped from ~175 to 145 after a full-day test. The situation recovered as soon as I switched back to 0.12.0. I'm running 6x1070 and the currently installed nvidia driver version is 387.34

    Have you noticed any improvements to the effective share rate after updating your miner? Which gpus/driver are you using?

    After some minutes spent in the dev channel of ethminer I saw thare should exist a "kern.log" file from which we should be able to understand, also in case of OC-related crashes, which gpu actually crashed. I'm looking for it because every time I got a crash there were -I think always- CUDA errors reported in the console for all gpu's. Is it normal?
Page 296
Viewing Page: 297