Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
Once you have a kernel with BFS, you will see better gains. When you configure the kernel, make sure you set the following:
General Setup ->
Select BFS
Zen-Tune Profile = Server
Processor Type and Features ->
Processor Family = Core 2/newer Xeon
Preemption Model = No Forced Preemption (Server)
Timer Frequency = 100 Hz.
I also do a bunch of stuff such as removing all drivers except for what I need, removing AMD support (since it's an Intel box), etc., etc. Just be careful and you'll be fine.
Two GTX 460s will net you approaching 30K PPD all by themselves provided with good OCs. It would be difficult to extract that much more from an i7 with 8 threads, four of which are real cores, to match or exceed the production of two Fermis to be advantageous in removing them, even with the most efficient OS possible.I suppose I'll just need to try to get GPUs working again, then, as the benefit to -bigadv still leaves me 10k ppd short of where i was with windows and 2x 460s
Got this compiled, and I also downloaded and compiled source from ck-sources. "Eselect" sees all three sources so I'm good to pick and choose.
If you want, I could try both out and let you know if I see any differences on the SR-2.
When I generated the "genkernel" it also asked me to add an "initrd" (initialize root disk) entry into GRUB.CONF under /boot/grub/grub.conf after the initial "kernel" entry. When I do a manual compiling of the kernel it doesn't list this as required in grub.conf, nor the "real_root" entry.
Is this ok, or does the kernel with manual config have a built-in module for this? I just want to make sure I don't have to use the liveCD again because I'm a n00b.
Thanks much!
Two GTX 460s will net you approaching 30K PPD all by themselves provided with good OCs. It would be difficult to extract that much more from an i7 with 8 threads, four of which are real cores, to match or exceed the production of two Fermis to be advantageous in removing them, even with the most efficient OS possible.
Now that would be very interesting. I might try this on one system that could absolutely use the benefits of Linux and it has one GPU client running.I realize... I'm trying to get the 460s running under wine with the gpu3 client. I'd say I'm about half way there... then I need to figure out overclocking under linux for those..... should be fun =)
Are you running zen-kernel and did you set the "Zen Profile" to server? It makes a big difference compared to the other options.getting tpfs of 29:41 on an i7 920 at 3.9ghz with c-states enabled on gentobit. compared to my old windows times of 36:15 with -smp 7, that doesn't seem like as much of a gain as I should be seeing. Any suggestions?
Then no, I had tried telling you in IRC that we had to make one change to his config after you downloaded it but you kept ping timing out. Switching the zen-profile to server increased his performance considerably, saving upwards of 3m per frame.Tobit, the kernel config I'm using is the one you did for musky, with a couple of drivers added in for my hardware... so yes.
Probably dumb on my part to use genkernel and core2 CFLags, but that's how one learns, and this is more exciting than my job right now which has me learning a 20 year O/S again. That's right, Novell
I realize... I'm trying to get the 460s running under wine with the gpu3 client. I'd say I'm about half way there... then I need to figure out overclocking under linux for those..... should be fun =)
Section "Device"
Identifier "Videocard0"
Driver "nvidia"
VendorName "Videocard vendor"
BoardName "nVidia Corporation Unknown device 0391"
Option "Coolbits" "1"
EndSection
IINM, the core and shader clocks are linked on these cards and they cannot be adjusted independently. My GTS 450 is also linked. I believe it's that way with all the 400-series cards.the tool available when I did it last couldn't separate the shader from core clock...
I will look around for you though...
IINM, the core and shader clocks are linked on these cards and they cannot be adjusted independently. I believe it's that way with all the 400-series cards. My GTS 450 is also linked.
I'm not seeing it... this should be in nvidia-settings? coolbits are enabled, 260.19.36 drivers
require a full reboot, or just an x restart?
Update - SR2#3 - Ubuntu 10 Musky Mix DFS kernal. Running at 4.3Ghz.
Changed checkpoint from 3 to 12 mins -saves 5 seconds a frame - which I can't recall being the case on windows.Nothing to sneeze at.
So that takes me to consistent 9mins 54sec (Windows average p6901 - 11mins 12sec)
Numa disabled. Cstates enabled. Captured p6901 from Musky.
which takes me to 197,700 ppd = 20.3% more than windows.
Interesting test: c-states has no effect on the x5680 running with turbo off. On the x5660 with turbo on, c-states has a large effect and should be left on. You cut 2 hours off your BigAdv processing time too, compared to Windows. Nice.
Musky is a legend.
If you need any help with that, just shout. I know it very well
H.
Hey all,
A quick update to Musky's post here on how to install CK source/BFS(cheduler):
http://hardforum.com/showpost.php?p=1036935472&postcount=259
If you get a "command not found" error when running "sudo add-apt-repository ppa:chogydan/ppa && sudo apt-get update"
You need to do:
sudo apt-get install python-software-properties
FYI - and thanks Musky for your fine post. Otherwise worked perfectly fine for me on SkankBox48.
You shouldn't need to install python-software-properties with Desktop. You do with Server. I purposely left it out, but it won't hurt anything to run the command in and Ubuntu release.
7:43[/COLOR] frame times consistently with stock (generic) kernel and NUMA on, nearly 10 minutes with the CK patches added and NUMA on (????) and much worse with NUMA off, which is the exact opposite of Windows Server 2k8 R2 x64 which shows great speed upgrades when NUMA support is OFF but still 3 minutes off Ubuntu.
that is the inconsistency that musky saw... cept his would start low and finish slow... and have spikes and other stuff...Think I need to do some more testing on the Dual Hex - Ubuntu Server 10.10 was showing a potential improvment, even though on a slower drive ....
Using OpenSuSE 11.2, CFS scheduler I'm seeing something a little odd though .. the last WU it completed was a 6901 (R9,C12,G3) it chewed through 100 frames at a consistent 22:01 - 22:04. The next WU was another 6901 (R14,C16,G3), first 7 frames were at 22:50, then dropped to 21:36 - 21:39 for the last 27 frames.
H.
but he is no longer seeing this with BFS.that is the inconsistency that musky saw... cept his would start low and finish slow... and have spikes and other stuff...
but he is no longer seeing this with BFS.
correct...that and ext3
but he is no longer seeing this with BFS.
correct...that and ext3
BFS fixed the CPU not loading 100%/inconsistent frame times issue. Switching from ext4 (Ubuntu's default) to ext3 fixed the issue where the client would wait 45 minutes between finishing a unit and starting to upload the results. This held true for an Intel single hex, an Intel dual quad, and a few Intel dual hex machines. I feel confident in saying that you should use the BFS for any Linux install on an Intel machine. The jury is still out for higher end AMD machines. You should use the ext3 file system regardless of platform.
BFS fixed the CPU not loading 100%/inconsistent frame times issue. Switching from ext4 (Ubuntu's default) to ext3 fixed the issue where the client would wait 45 minutes between finishing a unit and starting to upload the results. This held true for an Intel single hex, an Intel dual quad, and a few Intel dual hex machines. I feel confident in saying that you should use the BFS for any Linux install on an Intel machine. The jury is still out for higher end AMD machines. You should use the ext3 file system regardless of platform.
Hello from the Land of OZ.
what a mighty impressive thread this one is along with some amazing PC's congratulations!
I have recently started folding with ubuntu 10.10 also and wanted to see if any of our friends from across the globe were having trouble receiving -bigadv units under linux? Is there a different way to force -bigadv under this OS?
I personally am referring to my 2600K and have received all A3 smp units so far despite the appropriate -smp 8 -bigadv settings.
Curious also to know what kind of ppd any of you are receiving on the smp units - currently im about 26K ppd
Click to view full size!
Recently i have stepped up to run some comps at OCAU head quarters in hopes to get some more output. Things will get better over time i hope
I would like to add its great to see [H] going so strongly. Your output and team support in the forums is something to admire. I pray that you crush the evil forces that are EVGA (whom i secretly name DVDA - not sure if that computes over there)