George Smart - M1GEO

///////////////////////////////////////////
Folding AT Home with CPU and Nvidia GPU on Ubuntu

Posted: 08 Apr 2020 07:35 PM PDT
https://www.george-smart.co.uk/2020/...gpu-on-ubuntu/



With the outbreak of the novel coronavirus (COVID-19) in late 2019 and
spreading through Europe early 2020, a lot of tech and IT media wrote to
encouraging people to take up Folding at Home (FAH, F@H). I decided to join
in, which for the most part was pretty easy, just a case of installing the
software more on this later. But on my larger PCs, I wanted to make use of
my Nvidia GeForce GTX 1070. Using a GPU adds another workhorse to the
project, and is often faster/more suited to some of the computation tasks
(called Work Units or WU).



Installing the Software




The Folding at Home software comes in 3 parts: the client (FAHClient) which
does the computation work; the controller (FAHControl) which provides a GUI
for the client; and a viewer (FAHViewer) which renders pictures and
illustrations of the molecules being computed, albeit at the expense of
some compute performance.




The FAH Downloads Page has the files you need to get started. On Linux,
youll need the following files:



fahclient_7.5.1_amd64.debfahcontrol_7.5.1_amd64.de b (or my version with
fixed dependencies:
fahcontrol_7.5.1_amd64_m1geo_deps.deb)fahviewer_7. 5.1_amd64.deb




These should be installed with the following command:



$ sudo dpkg -i package_deb_file
$ sudo apt --fix-missing install




The first line installs the package and the second will obtain any
dependencies for the package. You may see errors after installing the first
package, but after the second, you should have any dependencies met. You
may need to tinker around and help the package manager fix any missing
issues.




Youll notice that I have supplied a modified version of FAHControl. This
has one of the dependencies removed, which means it is possible to install
on newer Ubuntu versions.




Once you have the programs installed, the client will start. You can stop
and start it with:



$ sudo service FAHClient stop # stop the client
$ sudo service FAHClient start # start the client




Youre ready to go. Use the control program to make changes to settings,
etc. Youre ready to start folding on your CPU!




You may be interested to benchmark your machine with FAHBench. The process
takes a little over a minute to run a 1 minute test, and it reports if
everything is working correctly. Fror







Setting up the Nvidia GPU




The more tricky part, for me at least, was getting the environment set up
for the Nvidia display.




The first thing to do is to open the Additional Drivers dialogue box from
the Software and Updates menu. You can jump directly to this by running the
following in a terminal:



$ software-properties-gtk --open-tab=4




Typically, you should select the newest driver available. At the time of
writing, nvidia-driver-435.








After pressing Apply Changes, the system will download the latest drivers
and install them with the required dependencies. Well add some extra
packages of our own too:



nvidia-opencl-devocl-icd-opencl-dev




Install the above libraries using the following:



$ sudo apt install nvidia-opencl-dev ocl-icd-opencl-dev




At this point, you should be ready to go



Advanced GPU Tinkering




You may also wish to tinker with the GPU a little bit, choosing to either
underclock or overclock the GPU to get either better reliability or
improved performance. One thing to note with overclocking is that while the
occasional glitch or artefact is acceptable during gameplay, it will cause
the FAH WU to fail.




Initially with my configuration, I fount the factory overclock on my Asus
ROG Strix GTX1070 O8G to be slightly unstable in FAHBench. I found that the
GPU would fail at at between 7% and 9% of the way through the test.



GPU Benchmark StartsGPU Benchmark Fails




Further investigation using the Unigine Superposition as a GPU stress test
allowed me to tinker with the settings to find a reasonable GPU Core and
Memory overclock parameters, as well as appropriate power limits. The Asus
ROG Strix GTX1070 has good cooling and so I was able to get away with
increasing the power limit and adjusting the clocks for long-term
stability. Bear in mind that stability is way more important than getting
that last couple of percent from the card.




Overclocking and tinkering was done using a mix of Green With Envy and the
Nvidia-Settings tools on Xubuntu 19.10 which come with the driver.



Other Considerations




If you dont fancy pushing your expensive graphics card hard, you may be
able to find a custom GPU card previously used for mining
crypto-currencies. These cards feature the same GPU chipsets, but are
crippled such that they cannot easily (although it is possible) generate
video. The Nvidia P106 offers features similar to the GTX 1060, while the
Nvidia P104 offers features similar to the GTX 1070.




Since GPUs are no longer viable to generate crypto-currency, theyre often
sold second hand on eBay and similar sites for relatively cheap prices.
While a modest Nvidia GTX 1060 with 3GB of VRAM costs around £150 on eBay
at the time of writing, an Nvidia P106-90 (GTX 1060 with 3GB VRAM) costs as
little as £26 each or £20 in bulk! At a once off, thats nearly a sixth the
price.








Watch this space for an update on how these perform