Windows vs Linux

[Nert]

Just curious ... was this hardware stable under Windows, or are these new systems ?

[Aurum]

Something went awry with the Linux installation. I'm trying to reinstall Linux and make a fresh start but I cannot and only get the BSOD. Don't know what to do.

[Aurum]

Just curious ... was this hardware stable under Windows, or are these new systems ?

This is the first of 3 Linux Mint 18.3 rigs I've converted. The 2nd and 3rd run F@H fine. I just want to start over with a fresh Linux install. I've tried DVD and bootable USB drive but both freeze up with a BSOD.

[bruce]

My first guess would be heat.

Are the fans all turning? Is the thermal paste properly applied and the heat-sinks doing their job? Does the case need better ventilation?

[Aurum]

I can't log into that rig to tell you, but, it's pretty hard to overheat 1080 Tis when there's only 3 on a MB. I do know the CPU folds fine sans overheating. It's an open rig with fans on the GPUs.
I'm going to try removing the SSD and installing a HDD and installing Linux.

[bruce]

Do all of the GPUs exhaust their heat out the back or do they dump in into the case? Is one GPU hotter than it's neighbors?

[Aurum]

An "open rig" has no case.
All 3 GPUs Failed.

[_r2w_ben]

Getting back on topic, this OpenMM issue gives a bit of insight.

When running on Windows, the OpenCL platform already does a yield() after each time step. We assume that if you're on Windows, it's most likely F@H and you want to sacrifice some performance to keep the UI more responsive. On Linux, we assume you're more likely running on a server and you want the best simulation performance possible.

Some of the performance deficit on Windows is by design to prevent people from quitting folding because Windows appears to freeze or scrolling is janky.

[Kuno]

Then the real problem begins. My folding rigs are headless and I use TightVNC to run them. But I cannot get TightVNC to run even though it looks like it installed. Looking for a guide to help me with that.
I also like to have Piriform Speccy, CPUID CPU-Z, GPU-Z and Windows Task Manager Performance so I can see the load on all CPU threads and balance them right. I've yet to find anything comparable that runs under Linux. If anyone knows good utilities to maintain headless PCs please post them. TIA
BOINC installed and ran fine using this guide: https://boinc.berkeley.edu/wiki/Install ... _on_Ubuntu. The only difference is when connecting to a new project one has to use the username when registering for that project as opposed to using your registered email address on Windows version.
I tried installing Storj but haven't gotten it to work. Storj is a breeze under Win7.
This was a new build and I had a HDD and an M.2 SSD plugged into to the new MSI X99A Raider MB. Oddly the BIOS did not see them. I went ahead and tried installing Linux Mint 18.3 and it installed on the M.2 SSD but doesn't see the HDD. Still much to learn about using Linux.
I can disconnect the monitor, mouse & keyboard and move it from my desk to the garage rack and it boots and runs fine.

It's possible you might need an HDMI or DVI dummy plug. I know with my linux system I had to get an HDMI dummy plug or else no RDP software would work at all. You can pick them up on Amazon for like 5 bucks or so.

I also agree with you on the sheer amount of outdated information that is out there for LInux. It's extremely frustrating trying to do all these things and the guides are sometimes years old, or they expect you have to have a PHD level of knowledge of the command line in Linux already. Honestly if it wasn't for Chris21010's tutorial I'd still be running windows for folding and I'd be losing points!

[Aurum]

Really??? Dummy loads. I haven't used one of those in at least 3 years. I still have some I made back in the day that I'll find and try.
But, if that's the case when I tried to set up RDP it was sitting on my desk with a monitor plugged in. I'm going to try TeamViewer tomorrow and see if I have better luck.
It feels like Linux GPUs fold at least 10% faster.

[v00d00]

Do a minimal install. You dont need X to run FAH and unless you are taking a particular interest in learning how to use Linux, don't bother. Your interest is in how to fold on Linux. Do what is required to achieve that goal.

Running X doesn't teach you Linux, it teaches you to do the same things using a GUI without ever learning how to do the same thing in a text terminal.

Its just that learning Linux requires a commitment, specifically you dump windows for a year or two and use Linux as your primary system. If you are just looking to maximise points and have no real interest in learning anything, then go for a headless build and control it from elsewhere over the lan.

Besides FAHControl requires X and if you are following the minimal approach you wont have X installed.

Has anyone written a guide (script?) so that a person can start X on a minimal install system but avoids automatically starting it? I'd be happy with a system that opens only a terminal screen but in which can start X manually when I want it. My objective would be to permit drivers to be updated without the need to re-blacklist nouveau to install nvidia drivers, I suspect that this isn't easy since the drivers get linked into the kernel. NVidia's Windows installer incorporates all those steps into a single command ... assuming you're running it's GUI and then ends by restarting it's GUI -- which in Linux would be unnecessary if it could be done manually.

I depend on a single copy of FAHControl in some GUI (easiest in one copy of Windows) to control all of the clients within the reach of my LAN. This prevents the errors caused by my inevitable sloppy typing when I change something.

From systemd i belive you change the target from graphical to text, but dont quote me on that as I don't use it. On Init systems, you alter /etc/inittab

Code: Select all

# Default runlevel. (Do not set to 0 or 6)
id:4:initdefault:
or
# Default runlevel. (Do not set to 0 or 6)
id:5:initdefault:

to

Code: Select all

# Default runlevel. (Do not set to 0 or 6)
id:3:initdefault:

X can be started (assuming you installed everything that is required to run it) by issuing the command startx at the console after logging in. This is the way I do it on my main machine. When i first boot the machine their is virtually nothing running besides 2x fahclient instances. When I run X, X takes up an additional amount of cpu time (8-20% on avg), but in my setup it doesnt take it from the cores assigned to fahclient processes, even though the knock on effect/slowdown i suspect would be negligible. I have my setup a specific way to avoid loss of cpu time to clients/cores due to X using it (isolcpus=x and client segregation via taskset -c xx <process>).

Generally to install nvidias driver and not the precompiled binary you find in most repositories, you have to manually blacklist nouveau, which is done in different ways on different distros. You also need kernel source and build tools. But once you have all of that, you generally just need to log in to a term running at runlevel 3 or text, run the installer, either let nvidias config generator build you an xorg.conf or do it yourself (/etc/X11/xorg.conf, find driver under Videocard section and change it to nvidia). For example.

Alter BusID to your own settings. You can find it by running "lspci | grep NVIDIA | grep VGA" and making note of the first two numbers, eg.

Code: Select all

FAH-GPU1$  lspci | grep NVIDIA | grep VGA

01:00.0 VGA compatible controller: NVIDIA Corporation Device 1c82 (rev a1)
05:00.0 VGA compatible controller: NVIDIA Corporation Device 13c2 (rev a1)

So the BusID for you cards are PCI:1:0:0 and PCI:5:0:0

Code: Select all

Section "Device"
        Identifier      "Videocard0"
        Driver          "nvidia"
        VendorName      "Videocard vendor"
        BoardName       "nVidia Corporation Unknown device 01df"
        BusID           "PCI:1:0:0"
        Option          "NoLogo"        "true"
        Option          "RenderAccel"   "true"
        Option          "backingstore"  "false"
        Option          "CoolBits"      "28"
EndSection

For two cards.

Code: Select all

Section "Device"
        Identifier      "Videocard1"
        Driver          "nvidia"
        VendorName      "Videocard vendor"
        BoardName       "nVidia Corporation Unknown device 01df"
        BusID           "PCI:5:0:0"
        Option          "NoLogo"        "true"
        Option          "RenderAccel"   "true"
        Option          "backingstore"  "false"
        Option          "CoolBits"      "28"
EndSection

Section "Device"
        Identifier      "Videocard0"
        Driver          "nvidia"
        VendorName      "Videocard vendor"
        BoardName       "nVidia Corporation Unknown device 01df"
        BusID           "PCI:1:0:0"
        Option          "NoLogo"        "true"
        Option          "RenderAccel"   "true"
        Option          "backingstore"  "false"
        Option          "CoolBits"      "28"
EndSection

Sorry for the rambling post. I haven't been able to check the forum for a bit. Also not sure if this is what you are looking for bruce.

[foldy]

@_r2w_ben: If the performance drop on Windows is by design they could design it as an option. Default: FAH little slower, Windows no lag. Option: Yield=false makes FAH faster but Windows more laggy. User can decide. And even more advanced: If Windows idle, no user interaction, then dont yield. If mouse moved, then yield.

[toTOW]

Getting back on topic, this OpenMM issue gives a bit of insight.

When running on Windows, the OpenCL platform already does a yield() after each time step. We assume that if you're on Windows, it's most likely F@H and you want to sacrifice some performance to keep the UI more responsive. On Linux, we assume you're more likely running on a server and you want the best simulation performance possible.

Some of the performance deficit on Windows is by design to prevent people from quitting folding because Windows appears to freeze or scrolling is janky.

Can we get an environment variable added to OpenMM/Fahcore to get rid of this thing on Windows dedicated systems ?

[bruce]

A n enhancement request for an option in FAH's software would seem appropriate, though I'm not sure anything would happen to it.
(It seems that Development needs to perform a yield() more frequently.)

[toTOW]

We had such options back in the CUDA/Brooks core days ...

People could choose between performance (and UI lag) or more user friendly experience (which is a good default behaviour for most donors). But I like to choose myself ...