Topic: ethminer-0.9.41-genoil-1.1 - page 54. (Read 397360 times)

Genoil, I tried the new release, checkout the 110 branch.
Now I get a new error: [ 61%] Building NVCC (Device) object libethash-cuda/CMakeFiles/ethash-cuda.dir/ethash-cuda_generated_ethash_cuda_miner_kernel.cu.o
nvcc fatal   : Unsupported gpu architecture 'compute_61'

Please tell me how to fix it. Also, can you look at my previous post on this page, the problems I mentioned are still not fixed in the new release. The compile still hangs at the " "nvcc fatal : redefinition of argument 'std'" error, and the CL/cl.hpp file appears to be there in the 110 branch, but after I run cmake in the build folder it's not there, although i am not sure it's necessary, since now I get a completely new error. Please advise.

edit: I saw your post on the other forum, I understand now why you keep those error generating arguments. Ubuntu uses gcc-5.3, and it's normal to have that as the target as most people use it. I can't stand it for some reason.

awesome
thanks
I will test it tonight when i get home

Pushed some changes to the 110 branch.
- binaries built with CUDA 8RC for Compute 2.0 and up
- CUDA kernel housekeeping
- some makefile fixes

Nothing special really, updating not required if all is fine.

I'm going to see if a simpler stratum client implementation will improve things.

I'll need to do more testing - I know from checking Suprnova consistently that Ethminer tends to underreport, but I can't tell by how much. Claymore's is no doubt more accurate.

sounds like library missing, did you install all the libraries mentioned on the README.md ?

from the Genoil's cpp-ethereum readme:

there is a way to mine with only the second gpu or third or any other minus the first? -t does not work, it run only the first gpu, if you set 0 or 1 or whatever

ok found it --cuda-devices N

At this point, it's only a start, as the first increases I got were small ones - but still significant - from rewriting Ethash in a different way. I was going to wait until I had more speeds from the optimizations I planned on, as this isn't all that great, but I do have records from a couple days ago of the very first bump.

This was done at the clocks shown, exact driver version shown, and needless to say, within minutes of each other to ensure the DAG was the same. The stock miner used was Genoil's, from his 110 GitHub branch.

Stock (NSFW): https://ottrbutt.com/miner/eth-290x-stock-05272016.png
WIP Rewrite (NSFW): https://ottrbutt.com/miner/ethwolf-290x-05272016.png

In my tests, it seemed to do ~6.500% better consistently. Taking the shown records and using the difference between the peak of each, it's 5.274% better. As I said, I've plenty more to do on it, but it's a start.

So yesterday I managed to compile genoil's miner for the first time on manjaro linux (arch based), after a very long time of struggles. I did it on a test virtualbox installation. Today I tried to install it on a real computer, but had no luck, following the same instructions I used before. I tried again on virtualbox and again no luck, the same error appeared. I am totally clueless now to what's going on or what exactly I did to make it work (I deleted the old virtualbox drive).
This is the error I get http://pastebin.com/cKx10qbb
I use these command to get it installed:

sudo pacman -Syyu
sudo pacman -S cmake boost yaourt screen crypto++ leveldb miniupnpc libcpuid opencl-headers cuda gcc49 libmicrohttpd
yaourt libjson-rpc-cpp

export CC=gcc-4.9
export CXX=g++-4.9
git clone https://github.com/Genoil/cpp-ethereum
cd cpp-ethereum && git checkout 110 && mkdir build && cd build && cmake -DBUNDLE=cudaminer ..
cd libethash-cl/ && mkdir CL && cd CL && wget https://www.khronos.org/registry/cl/api/1.1/cl.hpp
cd ~/cpp-ethereum/build/libethash-cuda/CMakeFiles/ethash-cuda.dir
nano ethash-cuda_generated_ethash_cuda_miner_kernel.cu.o.cmake ###line 79, remove "-std;-c++;"
cd ~/cpp-ethereum/build && make

What am I doing wrong here? Btw genoil can you please add cl.hpp to /cpp-ethereum/libethash-cl/CL/ in the 110 branch? It doesn't compile without that file.
Also can you check ethash-cuda_generated_ethash_cuda_miner_kernel.cu.o.cmake in /cpp-ethereum/build/libethash-cuda/CMakeFiles/ethash-cuda.dir and fix the "nvcc fatal : redefinition of argument 'std'" error ? I just removed -std;-c++; from line 79 but maybe that's not the perfect way to fix it.

Gigabyte AM1M-S2H
works like a charm
costs nothing
power usage is almost nonexistent
 

Actually, this kernel is a MASSIVE pain in the ass to make faster. It doesn't react to most of what you do, and the AMD OCL compiler is having none of this shit - hates the algo so much that it tends to erroneously compile your shit FAR more than usual. So, small tweaks aren't gonna fly with this one.

To get my speed increases, I first had to rewrite Ethash entirely in a different way - which got me the very first substantial increase. After that, I just said fuck it - the compiler is too much of a pain in my ass, so I modified Genoil's Ethminer implementation to load binaries. After that, I dropped to GCN assembly to continue working, where I can control things a little better.

a strange thing is happening when i mine in stratum mode on cointron and etheremine

basically my gpu show 0% usage but core and mem are at full speed, and it's not working, i need to kill the miner and start it again

what it make even stranger is that this thing only happen with one rig, the one with the low end cpu(g1840), that it is used at 40%

something wrong with the new version, 1.1 and weak cpu i think

edit

now happened with the other rig too, 100% that it has to do with the stratum mode thing, when i use dwarf and their stratum it does not happen ever

 Where did you find an AM1 motherboard with 3 PCI-E slots?

Is it possible to send out the miner hashing rates to a pool? I use the ethpool.org. I want to monitor my miners.

Two is pretty good. I do it in... I think like four-ish seconds, but I have 4 GPUs in it.

If your rigs run stable, I don't see a real reason to upgrade from 1.0.x to 1.0.y, unless you're looking for a specific feature. 1.1 isn't final yet, hence the different branch that I have it available in now. It will have the biggest change in it compared to the others: as in the streaming DAG files. It will quite likely als be the last release in this 0.9.41 series. I'm going to try something different maybe.

whats the oldest available branch you would suggest someone use before its imperative to upgrade?

In my rig, the apt-get install cuda, didn't work, i get "unknown package", so i removed it, i already had cuda installed for the ccminer from tpruvot and sp

also i had to do
cmake -DBUNDLE=cudaminer ..  instead of just miner, or it will not compile with cuda support
prettymuch everything else is straightforward, go for it.

git pull will update you, then try a checkout. Or, if you just want a new dir, nuke the old one and:

git clone -b 110 https://github.com/Genoil/cpp-ethereum

For the 1.1 prerelease you'll need to checkout the 110 branch.

I am thinking about trying this miner with some Nvidia cards and Linux.  I found the following notes on the releases page..  Is there anything else that we need to get this working?  Any other software, or known issues that I should be aware of?  Any tips on getting Nvidia cards up and working with this miner would be awesome.  Thanks.

From the Release Page:
wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_7.5-18_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1404_7.5-18_amd64.deb
sudo apt-get -y install software-properties-common
sudo add-apt-repository -y ppa:ethereum/ethereum
sudo apt-get update
sudo apt-get install git cmake libcryptopp-dev libleveldb-dev libjsoncpp-dev libjson-rpc-cpp-dev libboost-all-dev libgmp-dev libreadline-dev libcurl4-gnutls-dev ocl-icd-libopencl1 opencl-headers mesa-common-dev libmicrohttpd-dev build-essential cuda -y
git clone https://github.com/Genoil/cpp-ethereum/
cd cpp-ethereum/
mkdir build
cd build
cmake -DBUNDLE=miner ..
make -j8