Bigbeta in a VM

That's probably it. Especially if, when you typed that, you said "zed" in your head instead of "zee"....:)

Well, you wanted BIOS screens, so here they come. I took the opportunity to take shots of all of the relevant pages since I feel pretty happy with how I have the overclock settings dialed in. I could probably drop vcore a bit more, but I already had to boost VTT to keep things stable (Linux in a VM requires 0.05v more VTT to stay stable compared to Windows), so I don't want to rock the boat any further at this point. Oh and the inflated IOH and screwy signal timings are part of a half way successful attempt to clear up static in the sound under high CPU loads.

By the way, these came at a high cost. I lost a 6904 at about 2/3 completion, so I hope they are helpful to somebody :(
Apparently even going through the allegedly proper 'Ctrl C then ACPI shutdown' procedure isn't enough to guarantee a WU will stop and happily resume later. Sorry about the bad quality. That's a little one handed camera phone action going on there.



imag0168x.jpg


imag0169h.jpg


imag0170w.jpg


imag0171i.jpg


imag0172r.jpg


imag0173t.jpg


imag0174r.jpg


imag0175x.jpg


imag0176u.jpg


imag0177vk.jpg


imag0178h.jpg


imag0179ts.jpg


imag0182c.jpg


imag0183e.jpg


imag0184tm.jpg


imag0185j.jpg
 
Last edited:
Lost a 6904? :( (many Bothans died to bring us this information)

I have gone through mine and cant find any differences that I imagine would alter things... sheet. Thanks for doing so, it means we can rule out a lot of things.

Let me know if you are getting 100% CPU usage on windows, or ~95%

I have tried c states both ways, no difference like it does on native Liknux.
 
I see 95% as well. Discouraging news. Maybe test drive a fresh windows and vb install? Maybe something got mucked up along the way.
 
This info from the folding image site might be of help:

Shutting down or restarting

To shutdown the virtual machine, run this command:

shutdown -h now

To reboot the virtual machine, run this command:

reboot
 
I see 95% as well. Discouraging news. Maybe test drive a fresh windows and vb install? Maybe something got mucked up along the way.

Hmmm, interesting - keep us posted if your frame times are still respectable with 95% usage.

Very odd that on your rig you can beat native Windows times and I cannot.

Do I need to worry about Virtualbox Guest additions? Installing them on a Ubuntu guest was easy, but on a CLI I am in noobland...

original.jpg
 
I'm helpless in linux period. I did not install guest tools.

The 95% CPU has been consistent for two days. TJ is your experience similar?
 
ok, just installed Ubuntu as guest in VB and am getting same results as linuxforge - but with benefit of GUI for us dummies.

EDIT - most important, GUI makes it easy to backup FAH folder while running before control C shutdown. Check in work folder to make sure this did not happen during checkpoint writing - if all files are a few minutes old, you are good. I have not lost a WU since I have been doing this.

Ubuntu 11.04 with BFS kernel 2.6.35-28-generic-ck only loads the CPUs to 50-60% and is a disaster. From past experience I know that 11.04 with 2.6.39-0-generic is basically as good BUT this is not available in repository right now. Googled v3.0-rc7 and it works well - had to install from .deb download rather than a repository, but got it in the end. Testing more now.
 
Last edited:
Jebo you are a legend.

tom-cruise-rock-of-ages-first-pic.jpg

(picture of Jebo relaxing)

SUCCESS. Finally I have a workble solution, after months of faffing around. :D:D:D

Ubuntu 11.04 on latest kernel v3.0-rc7 as guest on Win7 host using Virtualbox. Numa disabled, krakken does nothing but I have it there anyway so I can sneakernet to real linux.

Numa enabled has a penalty - 13 mins 18sec to 14 mins 02 sec. (Krakken cant help)

I have a 2685 running now, that gets about 3% less PPD in a Linux Guest VM than native in windows. But overnight I can boot to native linux, sneakernet in the folder from the VM and crank it up for 45% boost in PPD.

6901 test WU got about 11% less PPD in the VM than on windows, but this is made up for by the chance to run 6903/6904 and real Linux overnight.

Or leave SR2#2 and #3 running linux most of the time, and boot to windows when I need the render power, and continue on in a VM...

All I need to do now is get help writing a script to automate the tedious backup and sneaknet. ;)

  1. Backup FAH folder while running.
  2. 2 minutes later backup a second copy, in case the first was running during a checkpoint
  3. ctrl-c FAH
  4. copy FAH folder and two backups to my media center shared folder on windows.

reboot machine manually from win7 to Linux or vice versa

  1. run script to delete FAH folder, and copy over the latest folder from the media centre.
  2. mod the executables to be executable - permissions are ok as they were copied to a nice dumb windows drive.

then manually run FAH - hovering over it ready to ctrl-c if it resumes from corrupt checkpoints... then watch the PPD roll in. :cool:

Actually a variation of the first script to automate and backup the FAH folder every 6 hrs would be awesome too - those 6904s run for so long, a power outage could be bad news.
 
You did install BFS correct?

It's interesting you're still seeing a 10% penalty. I am running my first regular bigadv right now, a 2686, and I'm seeing 13:20 TPF for right at 100,000 PPD. I don't actually know what native windows did on those units after the points nerf, but I don't think it was 110,000.

Anybody with an SR2 near 3.6ghz have TPF for a 2686?
 
Backing up the fah folder every six hours could fairly easily be scheduled in cron (assuming crond is running). You could even back it up to a different folder every time it ran during 24 hours. So you would have four cron tab entries each rsyncing the files to its own folder.
 
10% penalty on captured 6901, 3% penalty on live 2685.

Also on the 2685 i am running the vm minimized and disabled the preview window in VB. Might as well.

BFS kernel 2.6.35-28-generic-ck installed as per Mr Musky guide gave me 50% CPU usage, so I don't think it was working right.

What ubuntu are you using? 11.04?
 
I'm using the linuxforge image. Exactly the way it comes at stock except for the BFS installation. Then I just manually launch ./fah6 -smp -bigbeta
 
Oh, if only I had a dual proc box to test with....anyone have a spare Z8NA laying around? :)
 
Sorry for showing up late to the party to respond to you jebo(It's a left coast thing) but I am getting 100% usage from the CPU. Well it was before I took the snapshot. :D

vmstuff.jpg
 
Gotcha. I guess that could be the difference between 12 threads and 24 threads.

I think when MIBW gets BFS installed properly he will see the same gains I have been seeing, plus the possibility of bigbeta units.
 
Do I need to worry about Virtualbox Guest additions? Installing them on a Ubuntu guest was easy, but on a CLI I am in noobland...

It could be the newer version of Virtualbox. With each new version, new vbox-additions are made available and the image is built off the older VB version. I plan to update the image in the next few days with kernel 3 and the latest vbox-additions. Although I'm not sure that vbox-additions is going to affect performance at all.
 
Cool. Thanks for the input.

I think we've found BFS works great for Intel systems, but not for AMD. At least, that was the case in native linux.

Thanks, that is good to know.

I am pretty helpless once I'm in linux (hence the whole reason I'm trying to run FAH in a VM in the first place) so I wasn't able to install the Kraken. To able to fully take advantage of the Kraken, wouldn't VB need to be NUMA compatible?

I can't remember if VB will do NUMA or not. I thought there was an option for it but I'll check when I get home. Since you are getting good performance with your VB setup, it may be best to leave as is.
 
How can I tell from the startup messages that I have BFS installed correctly?
 
Skimmed through this thread, and after seeing the results, I've decided to try a linux-from-scratch build (that I was working on anyway) with fah in a vm, and see how it does. The os install is still a work in progress, so I'm not sure how well it will take to a non-stock kernel, but hopefully I'll have some results to share by the end of the weekend.
 
The benefit should be 100% load versus 1 idle core. I say should because there is a Windows OS is the way of what the Linux VM is doing. Who knows....I am glad you guys are seeing decent results anyway...
 
With the 2686 I am currently working on I enabled BFS after a few frames which gave me a bit of a before and after thing.
Before BFS enabled: average tpf 19:42​
After BFS enabled: average tpf: 19:30​

I'm sold on BFS. :)
 
My SR2 picked up its first 6903 and I'm averaging 30:09 TPF for 160,000 PPD
 
My SR2 picked up its first 6903 and I'm averaging 30:09 TPF for 160,000 PPD
Kendrak reports his SR2 at 3.5ghz gets TPF in the neighborhood of 27:30 for about 184,000 PPD, so it looks like the penalty for running Windows is around 15%, give or take.

That's a sacrifice I'm willing to make :)
 
Got Linux from Scratch all set up with BFS on my i7 920 in a vm over windows 7 64 bit. Haven't pulled a bigadv yet, but I'll post some tpfs once I do. Curious to see how this compares to the gains with Ubuntu etc.
 
This is very strange. The VirtualBox Manager froze up but the VM is still folding. I'm not going to mess with anything until I'm between WUs.
 
Scary.

You can close the vb manager independently of the vb window. I do it every time.
 
-smp 8 in vbox has shaved ~20 seconds off my tpf for a 2684. This box does have 2gpus in it, so I'll run the next unit at -smp 7 to see if that still helps performance.
 
Another reason that BFS is a good thing: Compared to the last 6903 I ran within a VM my tpf dropped by 49 seconds once I engaged BFS. My ppd went from 90k to 95. :eek:
 
That's awesome TJ. Free 5% boost is nothing to complain about. I think the increase is even better for 2p rigs.
 
I've got it running on ht2pc, the 2685 it got is running 22:22tpf versus the native 19:00 I was getting. This is with BFS and I'm pulling about 90% cpu utilization.

After it finishes a few frames I'll check out the bios and double check all my virtualization is enabled.
 
Found the problem - it is turbo!

Turbo is not engaging when folding in a VM.

Folding windows native:
original.jpg


Folding with Linux guest on Win7 Host in Virtualbox
original.jpg


Now with C States enabled or disabled in the BIOS, we get the same result, but I remembered a C1E work-around developed by sfield: You can disable it on both processors by running two instances of CPU Tweaker with affinity set to eg: cpu0, cpu12.

Bang. Turbo on both. :D

EDIT - took my TPF on a 6901 bench from 13:10 to 12:05. which is within a few seconds of windows. So now I truly have an always better than Windows option. You beauty!
 
Last edited:
Great news MIBW. I knew that TJ and I weren't the only ones smart enough to figure this stuff out (God help us if that's true :p)

Now my question is, why did you have to go through those steps and I didn't? My first guess says we are using different BIOS versions.

I don't recall which version I'm using off the top of my head, but Aida64 says I'm using Bios Version 080016 with a date of 10/28/10, (edit) which Google is telling me is version A50.

Which version are you running?
 
Now with C States enabled or disabled in the BIOS, we get the same result, but I remembered a C1E work-around developed by sfield: You can disable it on both processors by running two instances of CPU Tweaker with affinity set to eg: cpu0, cpu12.

Bang. Turbo on both. :D

EDIT - took my TPF on a 6901 bench from 13:10 to 12:05. which is within a few seconds of windows. So now I truly have an always better than Windows option. You beauty!

Great news MIBW, could you elaborate a bit on how to setup the c-state hack? I'm also having trouble with mine.
 
Back
Top