Need to get my priority/affinity worked out with WinAFC!

[woody86]

Ok, so I'm folding 24/7 on my sig rig. I have 4 GPU clients, and two notfreds VM running on VM Player. I am also running WinAFC 64 to tweak my priorities/affinities. OS is currently Win7, although I've had the exact same issue on Vista, so I'm pretty sure it's not the OS.

The first couple days I had it running after I got my second 9800GX2, I hit 26k, and 27k points. However, that was sometime ago. The past two days, however, my points have really seemed to take a dive - hitting 14k and 15k respectively. After seeing that, I shutdown one of my VM's and assigned the remaining VM to CPU0 and CPU1, and all the GPU clients to CPU's 2 and 3. Now after leaving it overnight, HFM is showing me ~25k points. Great! That seems better, but I can't for the life of me figure out why when running two VM's at a low priority they are still drowning out my GPU clients, even if I set them to 'Real Time'.

So what seems to be your 'sweet spot' for getting the priorities setup? What do you have them set on, and do you notice any drop in your GPU clients? I also wonder if I need to change the option in the clients for their cpu "core lock"? I've always left it as is, but maybe I'm supposed to change this? Anyone else have any other ideas?? I'd be **VERY** happy to hear them!! I'm missing out on 2.5k ppd by having my VM down, so I'd like to get everything working again. Please feel free to share any ideas or advice you may have!

Woody


My current affinityinput file for WinAFC looks something like this:

Code: Select all

*FahCore* = [priority = High]
*vm.exe* =ALL [assign=2, priority=BelowNormal]

(Please note, this isn't the exact contents of the file. I just want to give you a good idea about how I set them up. I know there are some errors in the above that wouldn't work in WinAFC )

[ChasR]

You shouldn't need third party software to manage FAH and VM priority and affinity. In FAH config set the GPU thread priority to low and unlock affinity on the GPUs. In the virtual machines directory edit the .vmx file to include ungrabbed.priority= "idle". That should solve priority conflicts and windows task scheduler will be able to move the gpu processes to fill idle cpu cycles on all processors. You can set affinity of the vms by adding the following to your .vmx file (assuming a quad, I don't see a sig rig):

processor0.use= "TRUE"
processor1.use= "TRUE"
processor2.use= "FALSE"
processor3.use= "FALSE"

swap the true and false for the second VM.

Are you sure all of the decline was priority related and not just a run of lower producing GPU WUs (511 pointers)?

[woody86]

Ah yes, sorry I copied/pasted this from another forum I posted it on. My rig is running a Phenom II 940 Black Edition OC'ed to 3.4GHz, and two 9800GX2's with their shaders maxed out. All four cores on the GX2's are folding, and I have two VM's assigned to 2 cores each of my CPU.

Well regardless, isn't it all the same no matter how you set the priority? Or am I mistaken? I didn't think there would be any difference in the method I used to change priorities, but in *how* I set the priorities.

I'm more than likely positive this isn't a WU/point issue. I have checked on my progress using HFM, and at the most it shows I'm pushing around ~19k PPD. With one VM off, and the GPU clients taking up those two cores, it jumped up to ~25k running the same WU's. The past couple days seem to be even worse as I'm only getting around 15k PPD, and I know this is extremely low for my setup.

[woody86]

Well, if this is any indication, things are looking better again. I disabled the CPU affinity lock in all the clients, changed them all to Real Time, and dropped my VM's to Low. Currently running two VM's and four GPU clients, HFM is showing ~24k ppd (including two 511 WU's). Hopefully that should take care of things. I'll keep an eye on my actual points produced for the next couple days to see how it does, but again if anyone has any tips/ideas please share them

[ChasR]

You could experience problems running processes in real time and it is completely unnecessary.