New to FTC mining: Issues with miners (in general) - NVidia (specifically)

[doomtower]

Hey all,

I’m new to Feathercoin mining and I had a few issues surrounding miners in general. It’s not necessarily a specific issue with a specific miner, and since I have to start in this forum category I thought I might as well hit the ground running.

I originally posted this to reddit, but I figured I might get more technical help here, so I’m elaborating on it.

I’m not new to mining or cryptocurrencies; got started in 2013 mining BTC on old junker rigs that I had laying around. Given that a) the machines were old and b) that was when the difficulty started to spike, I didn’t have much success and eventually I read the writing on the wall that it’d be an uphill battle and quit.

I decided to look into altcoins (and Feathercoin in particular) as right now I’m running a decently-new CAD workstation; 6-core Xeon, DDR4, Quadro GPU. Might as well put it through the paces for the significant portions of the day when I’m not around.

The issue I’m having is that miners nowadays are notoriously hard to get running correctly, at least in my perspective. When you take a bunch of different miners, a bunch of different coins, and varying dependencies and put them together it doesn’t yield for easy setup, so I’ve been trying for the better part of an entire day without much success. I’m running Ubuntu 16.04 and a fairly new install of it - but it seems that all the miners I’ve looked at that handle Neoscrypt-based currencies well are either for Windows, they don’t like Nvidia GPUs much, or they rely on at-best awkward Bash script installs that haven’t worked so far.

As I said in the post above, I’m not terminally-inept and I use it for a great degree of manipulation and software (if something says it needs a dependency I can usually figure out how to get that installed); it’s that there’s so much variability in the miner arena that nothing seems to work right for my setup. It’s not that I’m unable to figure it out, but needless to say the idea of installing (or trying to install) a new miner would elicit groans at this point.

Eventually, the cgminer-neoscript seemed to get up and running alright, but I have some concerns about it. I’m working off the Blocks Factory pool for FTC and making (according to the miner) about 42 Kh/s average.

• The repo I cloned from indicates the version is 3.7.8 but the CGMiner instance running says 3.7.1. It does indicate both Neoscrypt and Feathercoin operation at the top, so I’m not sure what to think.
• It seems to detect a new block on average about once a minute or more, and it always indicates detection by stratum from pool 0.
• The network difficulty is around 950K+. This could be right; I’m used to BTC difficulties so I really don’t have a reference point here.
• Been running this instance for a solid 15-20 minutes now and The Blocks Factory site interface still doesn’t report any hashrate.

About an hour later, The Blocks Factory interface still does not register my contributions even though the miner appears to be churning away. It also doesn’t help that my machine is extremely quiet, so I’m not hearing the old obvious signs of impending neighborhood brownout even though intensity is at 20.

Now comes the second part, and the part that I’ll (begrudgingly) fall back on if I can’t get this setup working: new software. I’m not looking for bleeding-edge hashing kernels; if I can get FTC mining properly set up I’m more than happy to take the slow-and-steady approach and be in it for the long-term. If I’m to install new software, I’d need it to meet three criteria:

• Be compatible with Ubuntu 16.04. At the very least, the software has to have a Linux-capable variety or be cross-platform if developed for another platform. I’m not going to extend the goose chase to installing a different operating system ($/FTC isn’t that high yet!).
• Be compatible with Nvidia GPUs, or at least the subset of functionality involving OpenCL. I’ve tried setting up cudaminer; the CUDA runtime is so nefarious that it’s damn near impossible to install from any angle I’ve tried so far.
• Be relatively painless to install. I’ve dealt with enough frustration trying to get a miner to work and don’t want to fight the Hydra again if I don’t have to.

Now, for the closing statements. First off, if you’ve read through all of this, thank you.

First off, do any of you have any ideas what may be the holdup on CGMiner-Neoscrypt? Am I worried about nothing and the Blocks Factory just happens to be slow with updates? I’m new enough to this avenue of cryptocurrencies that I don’t have a compass orientation to it all yet.

And if I have to fall back on the second option, what software would you recommend that meets the above criteria (compatible with my system and easy to install)? Back in the BTC days the nicer pools would often release their own GUI-based Java miners that worked without a hitch or any fuss at the slight expense of hash rate. I haven’t found anything like that yet, and what I wouldn’t do to find that option.

Once again, thanks for bearing with me.

[wrapper]

Hi and welcome to Feathercoin.

Ghostlander has written NSGminer specifically for FTC // Neoscrypt for some of the reasons you describe. Certainly, doesn’t Blocks factory have their own miner / support?

Although miners use the FTC code to mine, they are not now part of the core development, like when there was a CPU miner in the wallet.

Although I think there is a chance of cross compatibility of GPU types mined by the same miner, AMD and nVidia use different Graphics programming, Cuda and OpenCL, so that might not be optimal. In particular nVidia has chosen to be none standard and proprietary.

[bluebox]

@doomtower said in New to FTC mining; issues with miners (in general):

If I’m to install new software, I’d need it to meet three criteria:

• Be compatible with Ubuntu 16.04. At the very least, the software has to have a Linux-capable variety or be cross-platform if developed for another platform. I’m not going to extend the goose chase to installing a different operating system ($/FTC isn’t that high yet!).
• Be compatible with Nvidia GPUs, or at least the subset of functionality involving OpenCL. I’ve tried setting up cudaminer; the CUDA runtime is so nefarious that it’s damn near impossible to install from any angle I’ve tried so far.
• Be relatively painless to install. I’ve dealt with enough frustration trying to get a miner to work and don’t want to fight the Hydra again if I don’t have to.
  […]
  And if I have to fall back on the second option, what software would you recommend that meets the above criteria (compatible with my system and easy to install)? Back in the BTC days the nicer pools would often release their own GUI-based Java miners that worked without a hitch or any fuss at the slight expense of hash rate. I haven’t found anything like that yet, and what I wouldn’t do to find that option.

Once again, thanks for bearing with me.

In light of what @wrapper said, I still find CUDA-based miners simple to compile and use compared to AMD/opencl, so long as the code is kept tidy, and even in some cases is easily fixed (missing include statements, etc). There are pitfalls and trip-ups galore in compiling and using each, most of which can only be avoided by experience.

Specifically for neoscrypt and nvidia gpu’s, I run djm34’s fork of tpruvot’s ccminer-sp. Very stable, fastest hashrate you can get, and still coded with CUDA6.5 and 7.5 to cover a broad range of gpu’s. You simply do not want to run opencl-based miners on nvidia gpu’s if you can help it.

(Why CentOS? Stability, pure and simple; it’s my natural preference since I work in a data center. ELRepo repository has kmod-nvidia, so driver maintenance is dead simple; no more init 3 installs unless you absolutely need the latest-greatest.)

On Ubuntu you should be able to easily install the CUDA 6.5/7.5 toolkit (or both versions if your card(s) are suported, switching between them with the /usr/local/cuda symlink and an ldconfig). There are local runfiles or .deb installers available from nvidia, though Ubuntu 16.x isn’t supported yet the 15.x install should still work. Compiling ccminer just requires apt-installing a few deps (you may already have most already) and configure/make.

Regarding your hash not showing on the pool, sometimes it’s a mismatch between the pool config and the miner. I’d just try another couple of pools like coinotron or give-me-coins and see if they work (always have for me).

Since you’re running linux, you should know nothing is exactly “painless”, it’s just a different variety of pain compared to Windows. :) And, at last check, FTC mining on nvidia is >50% more profitable than the next best coin (ethereum).

All that said, I really really need to post a how-to for nvidia mining… “time… not enough…” :/

[doomtower]

@wrapper said in New to FTC mining; issues with miners (in general):

Hi and welcome to Feathercoin.

Ghostlander has written NSGminer specifically for FTC // Neoscrypt for some of the reasons you describe. Certainly, doesn’t Blocks factory have their own miner / support?

Although miners use the FTC code to mine, they are not now part of the core development, like when there was a CPU miner in the wallet.

Although I think there is a chance of cross compatibility of GPU types mined by the same miner, AMD and nVidia use different Graphics programming, Cuda and OpenCL, so that might not be optimal. In particular nVidia has chosen to be none standard and proprietary.

I’ll look into NSGminer; as to Blocks factory having their own miner - if they do, I certainly haven’t found it.

And I’m aware of Nvidia being arch-enemy #1 of open-source (we all know they received the Linus Torvalds middle finger), but I’m running myself on the constraint that I have to use the hardware I’ve got. If it makes me a healthy-enough ROI, I got an additional x16 bay on this workstation motherboard that I can use for an AMD card. But, first things first.

As to CUDA vs. OpenCL, I’d certainly be using native CUDA runtimes and things built for it if I could find a way to install the runtimes properly. Nvidia binary drivers don’t like X and every time I kill X (service lightdm stop the screen just goes black and I have no way of configuring the driver.

@bluebox said in New to FTC mining; issues with miners (in general):

@doomtower said in New to FTC mining; issues with miners (in general):

If I’m to install new software, I’d need it to meet three criteria:

• Be compatible with Ubuntu 16.04. At the very least, the software has to have a Linux-capable variety or be cross-platform if developed for another platform. I’m not going to extend the goose chase to installing a different operating system ($/FTC isn’t that high yet!).
• Be compatible with Nvidia GPUs, or at least the subset of functionality involving OpenCL. I’ve tried setting up cudaminer; the CUDA runtime is so nefarious that it’s damn near impossible to install from any angle I’ve tried so far.
• Be relatively painless to install. I’ve dealt with enough frustration trying to get a miner to work and don’t want to fight the Hydra again if I don’t have to.
  […]
  And if I have to fall back on the second option, what software would you recommend that meets the above criteria (compatible with my system and easy to install)? Back in the BTC days the nicer pools would often release their own GUI-based Java miners that worked without a hitch or any fuss at the slight expense of hash rate. I haven’t found anything like that yet, and what I wouldn’t do to find that option.

Once again, thanks for bearing with me.

In light of what @wrapper said, I still find CUDA-based miners simple to compile and use compared to AMD/opencl, so long as the code is kept tidy, and even in some cases is easily fixed (missing include statements, etc). There are pitfalls and trip-ups galore in compiling and using each, most of which can only be avoided by experience.

Specifically for neoscrypt and nvidia gpu’s, I run djm34’s fork of tpruvot’s ccminer-sp. Very stable, fastest hashrate you can get, and still coded with CUDA6.5 and 7.5 to cover a broad range of gpu’s. You simply do not want to run opencl-based miners on nvidia gpu’s if you can help it.

(Why CentOS? Stability, pure and simple; it’s my natural preference since I work in a data center. ELRepo repository has kmod-nvidia, so driver maintenance is dead simple; no more init 3 installs unless you absolutely need the latest-greatest.)

On Ubuntu you should be able to easily install the CUDA 6.5/7.5 toolkit (or both versions if your card(s) are suported, switching between them with the /usr/local/cuda symlink and an ldconfig). There are local runfiles or .deb installers available from nvidia, though Ubuntu 16.x isn’t supported yet the 15.x install should still work. Compiling ccminer just requires apt-installing a few deps (you may already have most already) and configure/make.

Regarding your hash not showing on the pool, sometimes it’s a mismatch between the pool config and the miner. I’d just try another couple of pools like coinotron or give-me-coins and see if they work (always have for me).

Since you’re running linux, you should know nothing is exactly “painless”, it’s just a different variety of pain compared to Windows. :) And, at last check, FTC mining on nvidia is >50% more profitable than the next best coin (ethereum).

I’ve considered using CentOS or other Red Hat-children in the past as I have experience with Scientific Linux, but I have a generalized dislike of yum (haha). I’m too familiar with apt-get for my own good.

Tell you what, if this crosses into tomorrow (or later tonight) I’ll scrap the Ubuntu install and slap on CentOS or Fedora to give it a whirl. This disk is still new enough to the point where a 1GB flash stick is still a good backup strategy.

I’m thinking of giving ccminer-sp-neoscrypt a shot. Now, in the readme it specifies:

./configure has an option named --with-cuda that allows you to specify
where your CUDA 6.5 toolkit is installed (usually /usr/local/cuda,
but some distros may have a different default location)

How do I find where the CUDA toolkit is installed (and if it’s the proper version)? I’ve tried the usual, i.e. sudo find /usr/ -name "*cuda*", but that brings up every damn file and I’m not sure I know what to look for.

And, if I were to scrap the Ubuntu install and run Fedora or CentOS (I’m pulling both x86-64 ISOs right now, slight preference towards Fedora because I’ve used it before but it’s up in the air), how would I install the CUDA drivers/runtimes/toolkits/black magic properly using either one?

Is it as simple as yum install or would I need to work manually with RPMs?

All that said, I really really need to post a how-to for nvidia mining… “time… not enough…” :/

You’re telling me. I’m currently trying to write a Unix kernel as a learning exercise - my time and life force are measured in negative numbers.

[doomtower]

@bluebox said in New to FTC mining; issues with miners (in general):

Why CentOS? Stability, pure and simple; it’s my natural preference since I work in a data center. ELRepo repository has kmod-nvidia, so driver maintenance is dead simple; no more init 3 installs unless you absolutely need the latest-greatest.)

I took this advice, ripped out the Ubuntu installation and am in the process of installing CentOS 7 off the “everything-install” iso. Installed Ubuntu one day prior as a means to test a Linux dual-boot on this machine, so changing distributions only costs me about an hour of running dd on a flash stick and then the installation.

It also comes with the benefit of avoiding issues with prior dependencies and the general clutter that comes from unsuccessfully attempting installs of at least 4-5 other miners with varying dependencies and constraints.

Working with a Btrfs file system, KDE environment, full developer installation to boot to give the best chance of hitting the ground running and avoiding too many non-local dependencies.

So far, I’ve only run one command, namely yum update just to freshen the system. Nada software other than what is already present with the full installation.

So, given a completely clean, “zero-day” installation on a CentOS 7 machine with a Quadro K2200 GPU, what would you do to install the necessary CUDA runtimes and then ccminer-sp-neoscrypt? Consider this the start of your tutorial :smile:

[doomtower]

Well, I’ll post this here even though it seems everyone has abandoned this thread. I’ve attempted to install ccminer-sp-neoscrypt (unsuccessfully) and I figured I’d log what I did and the error messages to see if anyone could help. As I’ve said above, I’m running a relatively clean CentOS 7 image.

General stuff went successfully up until the point of cloning the repository and installing.

• Updated via yum update, grabbed latest ELRepo RPM and installed kmod-nvidia.
• Downloaded the CUDA 7.5 toolkit, typical yum install cuda.
• I needed OpenSSL’s development package, so I grabbed openssl-devel.

Then came the fun stuff. I cloned the repository down, and as per the install file:

You can use ./build.sh to configure and build with default options.

It is advised to run ./autogen.sh before ./configure (autoconf and automake
need to be installed on your system for autogen.sh to work)

./configure has an option named --with-cuda that allows you to specify
where your CUDA 6.5 toolkit is installed (usually /usr/local/cuda,
but some distros may have a different default location)

See README.txt

So, I figured because I had autoconf and automake, I’d work the manual install just in case the toolkit was in a different location than the default. ls -al showed it was indeed in /usr/local/cuda. I made the autogen.sh and configure.sh executable via chmod, and proceeded.

The autogen.sh output is less than ten lines:

configure.ac:4: installing './config.sub'
configure.ac:6: installing './missing'
Makefile.am: installing './depcomp'
configure.ac:103: error: possibly undefined macro: AC_MSG_ERROR
        If this token and others are legitimate, please use m4_pattern_allow.
        See the Autoconf documentation.

If I attempt to run configure.sh anyway, I get a bunch of verbose output, but the two error lines at the end are:

./configure: line 6217: syntax error near unexpected token `,'
./configure: line 6217: `LIBCURL_CHECK_CONFIG(, 7.15.2, ,'

I’ve RM’d the attempt and am waiting further ideas about how to get this working. This is what I was referring to above - it always seems to be something.

[Wellenreiter]

The thread hasn’t abandoned, but probably people don’t have experience with compiling sg-miner for nvidia.

I suggest! That you open an issue on github for the sg-miner repository, so the developers can help you with the issue.
I think, that the issue is in autogensh, as autogen creates the configure script, and as your autogen shows errors, the configure script is incomplete or better not adapted to your system.

I can´t dierctly help with your problem, but as an alternative I can explain my setup and what I did to compile a Nvidia capable miner.

My system:

• Opensuse 13.2
• Nvidia drivers installed with drivers downloaded from Nvidia
• cuda 7.5
• gcc and required developement tools/libraries installed

I first installed the drivers and made sure, that the graphics was working. I´m using KDE.

• installed cuda 7.5
• downloaded cudaminer from here: https://github.com/cbuchner1/CudaMiner
• run autogen.sh, configure and make

done

I get about 230khash with my Geoforce 750 Ti, what is reasonable

I hope this helps

[Wellenreiter]

I moved the thread to the mining section, so miners will notice better