Folding@Home New Client Available

[Nick Name]

Via Dr. Greg Bowman on Twitter:

"In response to popular demand, we have created an update to the Folding@home software that allows you to prioritize COVID-19 projects. We encourage you to upgrade as the new software includes important bug fixes and security updates. Downloads are available here. Please also join me in thanking the Center for the Science and Engineering of Living Systems (CSELS) at Washington University in St. Louis for funding the development of this software update. "

I'll get this installed on at least one machine this weekend, if it goes smoothly I'll upgrade the other as well.

 

[Nick Name]

I tried installing this on the Win10 machine last night, and had some problems that made me think it had an showstopping bug. I still think there's a bug, but it's a problem trying to run it from the command line rather than with the new client. The short version is, if you're not running from the command line with the --chdir parameter it's probably fine. I haven't had enough time yet to see if it improves work fetching or not, but if I don't have any problems over the next day or two I'll install it on #1.

 

[BeauZaux]

Waiting for present folding tasks to complete. Will then update. Will it install (or update) on top of the old app or does it need to be removed? Speaking Linux.

 

[Sandman192]

Waiting for present folding tasks to complete. Will then update. Will it install (or update) on top of the old app or does it need to be removed? Speaking Linux.

Any work F@H has will delete all work and start from new (never finish old work). It happened to me. So waiting for present folding is wise. Speaking Windows but I would believe Linux is no different.

 

[Nick Name]

Any work F@H has will delete all work and start from new (never finish old work). It happened to me. So waiting for present folding is wise. Speaking Windows but I would believe Linux is no different.

Thanks for the tip. I installed over the top but I did set it to Finish and let it complete before installation. I haven't gotten it working on Linux yet, the plan was to work on that this weekend but I don't think I will have time.

 

[BeauZaux]

Tango Uniform for new folding monitor on my Linux. Client and tasks are running, but I can't monitor even from web.
Flashes "local connect/update inactive"

 

[Nick Name]

Tango Uniform for new folding monitor on my Linux. Client and tasks are running, but I can't monitor even from web.
Flashes "local connect/update inactive"

I've seen a lot of reports about that issue on the F@H forums. Can you run the Advanced Controller?

 

[BeauZaux]

I meant to say controller. I'll check the forums. Hope I can revert back if there's no fix.

 

[BeauZaux]

Installed beta 7.6.10. All is well...so far.

 

[Nick Name]

Installed beta 7.6.10. All is well...so far.

Sounds good, hope it stays that way. I've upgraded both machines here, no problems so far. I've been trying to get my Radeon VII working but no success so far. I'm not sure if I should blame Windows or the F@H clieint. When I get time I'll give it another shot.

 

[BeauZaux]

Two systems running fine with beta version of F@H in Linux. Does the VII work with other GPU projects or is it just the new F@H version making problems?

 

[Nick Name]

Two systems running fine with beta version of F@H in Linux. Does the VII work with other GPU projects or is it just the new F@H version making problems?

It works fine otherwise, I got it primarily for MilkyWay last year and it's great for that. To flesh this out, it's in the system I call #1 which also has two 2080 Tis. I run a separate BOINC client for each GPU, plus another for the CPU. That gives me maximum control over exactly what's running. F@H tries to use every processor by default, which I didn't really want to do, I'd much rather save CPU resources for CPU-only projects. I also want to keep the system working 100% either on a BOINC project or F@H. F@H work was all but impossible to get for awhile, it's still not as consistent as BOINC work. I've only been running it on a single 2080 Ti, and I have a program exclusion set for when the card is folding. It works well for one card, but if I were folding on all three, most likely only one or perhaps two GPUs would actually be folding at a time due to the inconsistent work, and the program exclusion would stop all BOINC work.

When I first started folding it looked like there was a separate app for AMD cards, but I don't think that's the case anymore if it was even true then, so I'd need a better way - meaning more complex scripts - to keep the VII working 100%, if I could get it working at all. The problem I have now is that I can't get the F@H client to properly assign work to it. Troubleshooting has been difficult because I'm not sure if the client isn't getting work because it's not available, or because there's a setup problem on my side, but as best I can tell either the F@H client crashes or the work it gets crashes, and the OpenCL error seems to indicate it's trying to assign the wrong type of work to the card.

Here's the F@H log showing the cards detected and their order.
GPUs: 3
GPU 0: Bus:5 Slot:0 Func:0 AMD:5 Vega 20 [Radeon VII]
GPU 1: Bus:1 Slot:0 Func:0 NVIDIA:8 TU102 [GeForce RTX 2080 Ti Rev. A]
M 13448
GPU 2: Bus:2 Slot:0 Func:0 NVIDIA:8 TU102 [GeForce RTX 2080 Ti Rev. A]
M 13448
CUDA Device 0: Platform:0 Device:0 Bus:1 Slot:0 Compute:7.5 Driver:11.0
CUDA Device 1: Platform:0 Device:1 Bus:2 Slot:0 Compute:7.5 Driver:11.0
OpenCL Device 0: Platform:0 Device:0 Bus:1 Slot:0 Compute:1.2 Driver:445.75
OpenCL Device 1: Platform:0 Device:1 Bus:2 Slot:0 Compute:1.2 Driver:445.75
OpenCL Device 2: Platform:1 Device:0 Bus:5 Slot:0 Compute:1.2 Driver:2841.19

You can see the VII is card GPU 0, but OpenCL device 2. No idea why that is but I think it's the main reason why I'm having problems.

Here's the config for my 2080 Ti that's working.
<!-- Folding Slots -->
<slot id='0' type='GPU'>
<cuda-index v='1'/>
<gpu-index v='1'/>
<opencl-index v='1'/>

Based on that, and how the client identifies the VII, which it does, the config for that should look like this.
<slot id='1' type='GPU'>
<gpu-index v='0'/>
<opencl-index v='2'/>

But this doesn't work, it gives an error. I also tried setting it up with the defaults and removing the Nvidia slots, but that didn't work either. I haven't had time to try with the latest client.

 

[Nick Name]

I spent a few hours tonight working on this again and I've concluded the F@H client just can't handle mixed GPU assignment, at least in my case. The VII fails even running with the default configuration, all GPUs are detected but VII work fails. I'm not going to spend anymore time on it.

I also tried getting the Rx 590 running under Linux, and that didn't work either. It works under BOINC, but for some reason the F@H client doesn't see it, even when running as root. I'm stuck with Windows there at least for the time being. I thought Nvidia + Linux was a giant pain but AMD is much worse.

 

[doneske]

Are you loading the AMD proprietary driver from their website? I would think that the VII would be recognized by the open source AMDGPU kernel module. If the card is recognized, I assumed that you would only need to install the ROCm module for the compute capability. Have tried any of it myself so not sure how robust the open source stuff is versus the proprietary stuff. I may have to spend a few bucks, get a vega card, and try it out.

 

[Nick Name]

Are you loading the AMD proprietary driver from their website? I would think that the VII would be recognized by the open source AMDGPU kernel module. If the card is recognized, I assumed that you would only need to install the ROCm module for the compute capability. Have tried any of it myself so not sure how robust the open source stuff is versus the proprietary stuff. I may have to spend a few bucks, get a vega card, and try it out.

The VII is in my Windows machine, I haven't tried it in Linux. It's water cooled and I'd have to do a lot of work to move it to another system. I still need Windows for my daily driver, I've thought about booting into Linux on the weekends but as much trouble as I've had the thought of trying to run Nvidia and AMD in the same Linux box gives me a headache.

There are a couple problems currently with AMD drivers for Linux, at least from what I know.

1. The open source drivers don't include the OpenCL libraries required for computing so you have to get that from the proprietary driver.
2. AMD's proprietary drivers are only supported on certain distros, you can get them working on others but it takes some effort. I've grown accustomed to Pop!OS which is an Ubuntu derivative, while the proprietary driver works on Ubuntu LTS 18 it doesn't work on Pop.

I didn't have any luck with RocM, the OpenCL was never detected. I found this on the MilkyWay forum and was able to get the proprietary driver working. It works fine under BOINC, just F@H has the problem.

amdgpu-install --opencl=legacy,pal --headless --no-dkms

Most of my problems likely are a result of me just being terrible at Linux.

 

[doneske]

Reading through the Milky Way thread, I have a hard time understanding why OpenCL works on some projects and not others unless it's due to compute level requirements like 1.2, 2.0, 3.0 etc. I would think OpenCL is OpenCL. Once you have it working it should just work. Like I said, I'm going to have to get a card and try it out. Problem I have is, I can't just get a card. I would have to get a motherboard to support the PCI-e requirements and all that entails like new processor, memory etc. In other words, build a new system.

 

[doneske]

have you tried adding the fahclient userid to the video group? I think you have to be a member of the group to access the GPU. I would be interested if the boinc userid is in the video group. Additionally, when you installed ROCm, did just install ROCm or did opencl-rocm-dev get installed too? There are a few pieces to ROCm depending on what you want to do..

 

[Nick Name]

have you tried adding the fahclient userid to the video group? I think you have to be a member of the group to access the GPU. I would be interested if the boinc userid is in the video group. Additionally, when you installed ROCm, did just install ROCm or did opencl-rocm-dev get installed too? There are a few pieces to ROCm depending on what you want to do..

I did add fahclient to the video group. It's very confusing because there is fahclient and FAHClient, the latter being what you need to actually run the program. That flummoxed me for a bit. As best I can recall, when I installed BOINC it added itself to the video group without me needing to do anything, although I've seen reports of folks having to do that.

I can't recall those details re. ROCm, I just remember it wasn't working so I uninstalled it. It's possible and even likely that part didn't get installed.

I found this while I was trying to get it running:

"...you need to tweak the startup script for the client (/etc/init.d/FAHClient), changing the username from "fahclient" to "root"..."

And another post with a theory about what the problem is and how to fix it here.

By the time I found that I was already a couple hours into troubleshooting and I may have missed something, but it was time to go to bed.

I also installed from a repo someone created and it seemed to work fine, but it's possible something didn't go right. So admittedly there are some potential self-induced bugs in the process that may have broken something.

The strangest part is the client sees an OpenCL device, which is the GPU, but still reports zero GPUs.

 

[Nick Name]

I haven't observed any serious problems with the new client but neither has it improved anything for me. Work requests still seem to hang after some period of time and doesn't get work until the client is restarted. I've seen my Nvidia machine not get any work for 24+ hours. I suppose it's possible there really isn't any work from any project, but that seems unlikely to me when the AMD card is still getting work. In short things seem much the same as they've been.

 

[Nick Name]

I debated starting a new thread but decided to post here. I FINALLY got the client working on Linux, or at least the GPU recognized and working. It took me 3-4 hours, there's something weird going on with my wireless connection on that machine which might be a failing adapter. Anyway I installed the OS from scratch, I already had the USB stick ready and that process only takes about 10 minutes.

This worked for me on Pop!OS.

1. Download the AMD driver from the AMD website, extract and run ./*directory where you extracted the driver*/amdgpu-install --opencl=legacy,pal --headless --no-dkms
2. Install the clinfo utility to make sure the necessary OpenCL components were installed: sudo apt install clinfo, then clinfo. If no devices are reported some troubleshooting will be needed but my GPU was picked up right away.
3. Download and install the BETA F@H client, 7.6.13 at the time of posting. It looks like there's only a 64-bit client available here. I don't have any suggestion if your system is 32-bit. You should install the fahcontrol if you need advanced control of the client, e.g. to set up remote access, and the fahviewer for simple monitoring. This also fixes the outdated Python dependency problem. NOTE: I first installed from the repo mentioned here but the GPU wasn't detected. I then installed the Beta client - just over the top - and it was. The installer showed the installation as a downgrade, not sure why but it worked. I suggest installing the Beta client to start with.
4. Run FAHClient and start FAHControl to see what's happening. There's a nice shortcut in the Pop!OS menu to the control app, but you'll need to start the client from the terminal.
5. If you're running BOINC as well and want to set up an exclusive GPU app, the F@H GPU name is FahCore_22.

I like to run it with the --chdir directive so I can easily access the config files etc. without needing to root through the system folders. The only thing I'm a bit concerned about is controlling the fan speed. I can easily do that in Windows, for some reason it's still a problem in Linux. Temps aren't a problem right now but I'll have to watch more closely than before.