Dumped WU When Switching GPU

[jseyfert3]

Hey all,

Relatively new here. Used to run some FAH with BOINC alongside other projects in the past, but got out of it, mostly due to excessive heat in the summertime and never restarting in the cooler months.

Anyway, my current rig is a pieced together old server box with a Xeon E3-1225 CPU and an GTX 760 Ti GPU running Linux Mint 19.3 (primary) with Windows 10 as dual-boot. When folding with GPU, any graphics, including just scrolling webpages, becomes painfully slow, which matches my memory with BOINC back in the day.

Anyway, I was halfway through a GPU WU, and I wanted to see if I could reboot and use my onboard GPU to run a monitor (instead of dual monitors off my 760 Ti), so I could use the computer normally for non-graphics intense applications but let FAH run the GPU while I used the computer. I booted on the integrated graphics, tried running FAH, it gave an error that no supported GPUs were detected. Ah well I thought, rebooted and selected the Nvidia card again. Booted up FAHControl, and rather than resuming at ~50% now that the supported GPU was back it dumped the WU.

Code: Select all

14:38:37:WU02:FS01:Sending unit results: id:02 state:SEND error:DUMPED project:14533 run:0 clone:7779 gen:6 core:0x22 unit:0x0000001080fccb025e72f26fee06ce8c

Anyhow, just wanted to check if this was expected or if it could be a potential oversight, as not many people probably have done what I just did.

Going forward, I'll just let it run when I'm away from the computer and pause when I'm on the computer and stop trying to mess with getting monitors running while the GPU folds unless I'm on a machine with two dedicated GPUs in the future.

[Jesse_V]

Hi and welcome to the forum.

This is expected and fairly normal. The software tries to refer to each GPU with a specific OpenCL or CUDA identifier and when no hardware matches that ID, the processing will abort and the software will dump the workunit. For example, it knew that an index of 0 referred to the one GPU on your system, the 760TI. When you told the motherboard to switch to the internal GPU, the F@h software could no longer see the 760, so an index of 0 didn't refer to anything anymore. When you booted back into the 760, the software dumped the workunit. At least, that's how I think that it went.

[jonault]

It should be possible for you to run the desktop off the integrated GPU & reserve the 760Ti for folding; I have a system where I'm doing that. I don't remember exactly what I did to set that up correctly with F@H. I would get the video set up the way you want it, then uninstall & reinstall F@H; it should detect the new GPU setup correctly. If it does create a folding slot for the integrated GPU, you can just delete it.

Changing the hardware configuration while a work unit is in progress is a risky proposition, I've lost work units before under similar circumstances.

[jseyfert3]

It should be possible for you to run the desktop off the integrated GPU & reserve the 760Ti for folding; I have a system where I'm doing that. I don't remember exactly what I did to set that up correctly with F@H. I would get the video set up the way you want it, then uninstall & reinstall F@H; it should detect the new GPU setup correctly. If it does create a folding slot for the integrated GPU, you can just delete it.

That's what I figured I could do, but I didn't try reinstalling FAH. So typically I have a dual monitor setup on the 760 Ti. My first go at trying to use integrated GPU I rebooted, found an option in BIOS to output on the PCI GPU while also using the integrated GPU. Booted up that way.

My two monitors were connected to the 760Ti with DVI cables. I connected one of two using a VGA cable to the integrated GPU. Once I booted with the dual output BIOS setting I could tell Linux to use either the DVI to that one monitor or the VGA. So the integrated card was outputting. Problem was even if I shut down both DVI outputs in Linux, when folding using the 760 the VGA output had the same stutter and lag as the DVI outputs from the 760. So it was like the 760 was still running the X window server and then feeding the graphics to the integrated GPU, rather than the integrated GPU running the X window server.

I'm not a Linux expert by any means, so I wasn't sure how to try and fix this within Linux. So my next try was rebooting and selecting only the integrated GPU, thinking it would display only on that but FAH could still use the 760 for computing. Not so much. On that boot is when it gave me the error for no compatible GPU found, so I rebooted, changed BIOS back to what it was before I started messing around (outputting only on the 760), and when FAH started that's when it dumped the WU.

My guess is I need to go back to the BIOS option of outputting to both PCI and integrated GPUs, then once I boot I'll have to use some Linux command line to stop the X window server on the 760 and start it on the integrated GPU. But I'm not sure how I'd do that. Google could probably help but I'm a bit lazy.

In any case I'll wait till a GPU WU is completed before trying to mess with the GPU options again.

[jonault]

On my system, I only have one monitor connected to the MB graphics & nothing connected to the GPU cards; I've never tried the kind of dual monitor setup you describe (two monitors connected to two different GPUs). I'm running Windows, not Linux, so I can't give any advice on modifying the XWindows config.