suitable Linux distros for BOINC?

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5842
Credit: 109403504651
RAC: 35516541

cecht wrote:... the rx570

cecht wrote:
... the rx570 (device 0) reverting to default clock settings ... 

I have a vague recollection of seeing something like that ... somewhere ... but I didn't pay much attention because I don't change GPU settings.   I've been following a lot of stuff at Phoronix, particularly amdgpu development, so it may have been on that site.  I'll have a bit of a browse and see if I can find it again.

 

Cheers,
Gary.

cecht
cecht
Joined: 7 Mar 18
Posts: 1421
Credit: 2445552956
RAC: 1497520

Alas, I like working with

Alas, I like working with Ubuntu, I really do, but could not (yet) figure out how to run the RX570 along with the RX460 and keep the temperature of the RX570 below 80 C. By itself, the 570 ran just great at default settings. Compared to Windows 7 on that same PC, I saw a 5% task productivity increase with a 3% tasks/KWh efficiency increase under Ubuntu. But, unlike Win7, when I included the RX460 in crunching tasks, the RX570 GPU temp increased from 75 C to over 80C, which didn't happen under Windows. The RX460 ran hotter too. Using AMD's ROCm-smi program, I could not find a way to stably underclock the 570 or power limit either card. I did use it to keep temps down with increased GPU fan speeds, but I didn't think that was a good long-term strategy. The result was I was running more watts at the wall and running the cards harder than I cared to.

Gary Roberts wrote:
With regard to keeping your temperatures in check, there are two factors to consider.  As you've already found out, frequency is one, but core voltage may be an even bigger one.

I tried undervolting the rx570 with rocm-smi, but it spat an error and said no-go because it couldn't access the needed file. Maybe that's where the ohgodatool comes in, but I didn't go there.. (Big Dragon).

With AMD's WattMan in Windows, where I can lower voltages, I haven't because the rx570 seems to automatically run at lower voltages. The speeds & voltages in the default state table has the top three speeds all paired with 1150mV (or something like that). While running E@H at the top GPU speed, however, the average voltage is ~960mM, as monitored on GPU-Z. Before I realized this, when I tried to incrementally lower the voltage on the top speeds (1150 -> 1100), I saw a big increase in GPU temp and fan speed. So a manual adjustment of voltage nixes the automatic undervolting and forces it to the set voltage. I suppose I could have set the top speed states to ~950mV, which is what some cryptominers say is best for the rx570, but with it averaging itself around 960mV, I didn't seed the need to intervene. For the same reasoning, I didn't poke around too much with altering state tables in Linux.

Gary Roberts wrote:
...You then don't have to worry about being careful to preserve the separate boinc ownership and permissions regime.

The only issue I initially had with Boinc Manager was getting access to the app_config file, but got around the permissions issue with:
$ sudo leafpad /var/lib/boinc-client/projects/einstein.phys.uwm.edu/app_config.xml
, where leafpad is the default Lubuntu GUI text editor. I think other flavors of Ubuntu use gedit as the default. I haven't yet ventured into doing anything more than that with BOINC, but when I do I'll keep in mind the tip to install BOINC in my home directory.

Until then, sadly, I'm back to running the cards on Windows 7 and using AMD's WattMan utility to gain power efficiencies (and top up my flagging RAC). Both cards are now running at ~74 C, with lower fan speeds, and pulling 40 fewer watts at the wall.  I will stay on the lookout for a Linux GPU utility that can do what AMD WattMan does on Windows. I wonder whether using AMPGPU-PRO drivers or installing AMD's full ROCm system would make any difference? My old mother board might not take advantage of what those upgrades could offer.

Regarding the GPU clock reset:

Gary Roberts wrote:
I have a vague recollection of seeing something like that ... somewhere ...

I saw this on a cryptominer's blog: "Note that the (rocm-smi) clock setting is only temporary for the next opencl program to run.  Just run the rocm-smi command each time."  https://nerdralph.blogspot.com/2017/03/amdgpu-pro-1660-on-ubuntu-kernel-4105.html The FGRPB1G app is a an openCL program, right? So perhaps it's something in how E@H tasks tick over that triggers the rocm-smi clock speed reset, (Modified fan speeds and other parameters are not reset.) I tried to pin down when the GPU clock reverted to default during E@H runs, but ran out of patience; it wasn't as simple as when a single new task started running (while running 2x tasks).

Thanks for the other tips too Gary. I haven't given up on Linux, not by far, just regrouping. Cheers.

Gary Roberts wrote:
I really know nothing about Ubuntu and how things are organised there but I'm sure it would be fairly simple to make the transition - particularly for a flourishing Linux Pro like yourself :-).

  Now I just feel bad. :-(  (LOL)

Ideas are not fixed, nor should they be; we live in model-dependent reality.

kingcarcas
kingcarcas
Joined: 24 Feb 15
Posts: 2
Credit: 198731294
RAC: 0

Anybody know how to hide the

Anybody know how to hide the Boinc taskbar icon like you can in Windows? It's a small annoyance, but anywho, TC if you're trying to run a 400 series card and a 500 on the same machine hats off to you, but that seems like just asking for a headache.

cecht
cecht
Joined: 7 Mar 18
Posts: 1421
Credit: 2445552956
RAC: 1497520

Thanks to a post from Keith

Thanks to a post from Keith Myers over on the All Things Vega VII discussion https://einsteinathome.org/goto/comment/169630, I'm back in the Linux/Ubuntu game.  The utility he mentioned, amdgpu-utils on GitHub, is working very well to stably power cap both my cards so I can now efficiently run 2x tasks.

Ideas are not fixed, nor should they be; we live in model-dependent reality.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.