Folding Forum

So would it be more productive for me to use Linux in a virtual machine? Should I just do that? Also, the version of Linux that I have is Kubuntu. Is Kubuntu compatible with Folding@home or only Ubuntu? I heard they were virtually identical aside from the interface.

Yes.
Yes.

I was in the process of changing over all my Q6600s (I have 14) to VM/Linux v6.34 folding when p11040 came out, and the windows machines got quite a few of them. That erased the 20% advantage Linux had, so I quit the switchover, leaving 4 Q6600s on Windows. When p1104x and similar p110xx WUs are not longer being assigned, I will complete the transition.

As 7im often points out, the pendulum. swings back and forth on which OS is the fastest. As to which is the most productive, it swings even faster due to WU variability and assignment patterns by OS. That is the beauty of having a Linux VM on every machine. When the pattern changes (the pendulum swings), you can -oneunit the currently folding OS and change to the other in a few minutes. I've switched a few machines to v7 and find that they tend to get a lot of p80xx WUs that are also very high producers that I've yet to see on a Windows or Linux 6.34 machine (could be an assignment flag setting), introducing yet another variable, client version, to the quest to maximize ppd.

Another VMware advantage is apparent when running with GPUs. THe GPU can be easily overclocked in Windows, but you'd have to flash the bios on the video card to overclock it in Linux. Running SMP in the VM and the OC'd GPU(s) in Windows is often faster than running both clients in native Linux. The overclock on the gpu more than make up for the VMware overhead.

ALL free VMware products limit the number of cores available on the VM, so if you have more than 4 cores, you have to be sure to get a version that supports the number of cores you need up to 8. If you need more than 8 in a free VM product you have to go to Virtual Box.

FAH v6.34 will run on almost all versions of Linux, native or VM, and it certainly works in kubuntu. If running v7, you may have problems with the installer in some versions of Linux.

ChasR wrote:ALL free VMware products limit the number of cores available on the VM, so if you have more than 4 cores, you have to be sure to get a version that supports the number of cores you need up to 8. If you need more than 8 in a free VM product you have to go to Virtual Box.

Oh. I have 8 cores (4 physical cores, hyperthreaded) and the free version of Oracle VirtualBox. If it's the free version, would it still work to its full extent because I have only 4 physical cores or would it count them as 8 cores? And I assume that, if it doesn't support all of the cores, it won't be as productive?

If Linux is really that much faster at completing WUs, I could run FAH on Windows during the day while I'm using the computer but at night I could restart it and boot Kubuntu instead of Windows 7 and have it run overnight. Would that be a good idea? Is there any way for them both to work on the same WU (i.e. if Windows hit 28% and I restarted it running Kubuntu for the night, would Kubuntu resume where Windows left off at 28%? Or would they have to work on totally different WUs)?

I believe that VirtualBox will support all the 8 CPUs. There will be some overhead but as mentioned above, there is a net gain.

In theory, it could be possible if BOTH installations (Windows and Kubuntu) pointed to the SAME data files. However, I haven't tried it so don't know. If you do separate installation, then both machines would be processing different WUs.

EDIT -> This is a very bad idea since the FahCores can be different for each OS which may cause issues. Moreover, some projects are OS Specific which can complicate things further.

The free version of VirtualBox only supports 4 cores max.

Just a footnote, given identical fahcore, Windows and Linux perform the same. So a better question is which fahcore runs faster with the current position of the performance pendulum?

P.S. For those who would complain about points inconsistencies between Windows and Linux, we could fix that by artificially crippling whatever version is currently faster, thus slowing the search for cures!

Stonecold wrote:

ChasR wrote:ALL free VMware products limit the number of cores available on the VM, so if you have more than 4 cores, you have to be sure to get a version that supports the number of cores you need up to 8. If you need more than 8 in a free VM product you have to go to Virtual Box.

Oh. I have 8 cores (4 physical cores, hyperthreaded) and the free version of Oracle VirtualBox. If it's the free version, would it still work to its full extent because I have only 4 physical cores or would it count them as 8 cores? And I assume that, if it doesn't support all of the cores, it won't be as productive?

If Linux is really that much faster at completing WUs, I could run FAH on Windows during the day while I'm using the computer but at night I could restart it and boot Kubuntu instead of Windows 7 and have it run overnight. Would that be a good idea? Is there any way for them both to work on the same WU (i.e. if Windows hit 28% and I restarted it running Kubuntu for the night, would Kubuntu resume where Windows left off at 28%? Or would they have to work on totally different WUs)?

You can't run the same WU in two OSes on the same machine. You would have to work on different WUs, which would approximately double the timeout on each and reduce your overall ppd by 65%.

My understanding of the latest version of free virtual box is that it supports up to 32 cores. From the Virtual Box manual:

Guest multiprocessing (SMP). VirtualBox can present up to 32 virtual CPUs to each virtual machine, irrespective of how many CPU cores are physically present on your host.

ChasR wrote:Guest multiprocessing (SMP). VirtualBox can present up to 32 virtual CPUs to each virtual machine, irrespective of how many CPU cores are physically present on your host.

But don't plan to use that capability to process bigadv. You'll be able to get the assignment and prove to yourself that when you miss the deadlines, you'd be better off running standard SMP assignments which will still be get QR Bonuses.

Creating "extra" cores can never increase performance -- and, in fact, reduces performance.