GPU 2 Folding issue (Priority Issue?)

Norcalsteve said:

I have an issue with my #2 GPU (GTX 285), on a 353 Reward WU, i normally get 8500-9500 PPD, but now i am getting 4000 PPD on it. Still get 8k-9k on my 1st GPU.

Recently, i finished a -smp 7 VM Bigadv WU, with a -oneunit flag, to change it back to -smp 8, because, thanks to ChasR, i had my "Priority's" straight. Now, I have been gone all day... and the GPU only has folded 4 WU, and my other is up 9. Is there an environment variable i can input for my second GPU to get a priority boost in it?

My set up now consists of:
VM Ware with -smp 8 (Both VM priority.grab/ungrab "Idle" commands input to VM client)
GPU1 is Low priority ("slightly elevated)
GPU2 is Low priority ("slightly elevated)
and i have the NV_...... value of 15 Environment Variable input to Windows

Any thoughts? I know its not the WU's, cause history shows i have folded them before, and i should be getting 33sec/frame, now its up to ~2min/frame for same project.

Click to expand...

Jolly-Swagman said:

In config setup for GPU,s did you set Advanced methods yes and CPU Affinity lock disabled yes

Then set Affinity for GPU,s to use all cores in Task manager

Click to expand...

ChasR said:

I've had this happen on a dual gpu system. It seems the later drivers try to save power by dropping the clocks on the 2nd card. To fix this I forced 3d performance clocks with RivaTuner.
Start RT/Power User/Expand RivaTUner\NVIDIA\Overclocking/right click on EnablePerfLevelForcing/click on the light bulb and select OK. RT will close. Start it again. Main/Driver settings/customize/System settings/ select performance 3d in the Force Constant Performance Level for each gpu. Reboot.
Another easier solution is in the nVidia control Panel/Manage 3D settings/Global settings/ for some cards there is a selection to change from adaptive perfromance to performance. If you don't have these nVidia settings, use RT.

Click to expand...

Norcalsteve said:

Wow, the Affinity's only use 1,2,3,4 i had to "Select all" for 0,1,2,3,4,5,6,7... also, i have to do it every time i open the GPU clients... is there a setting to make the affinity's default to All cores/threads?

Click to expand...

Norcalsteve said:

Wow, the Affinity's only use 1,2,3,4 i had to "Select all" for 0,1,2,3,4,5,6,7... also, i have to do it every time i open the GPU clients... is there a setting to make the affinity's default to All cores/threads?

Click to expand...

ChasR said:

Tell me how you are currently configured.

Click to expand...

[settings]
username=dylskee
team=32
passkey=
asknet=no
machineid=3
bigpackets=big
extra_parms=
local=460

[http]
active=no
host=localhost
port=8080
usereg=no
proxy_name=
proxy_passwd=

[core]
priority=96
cpuusage=100
disableassembly=no
nocpulock=1
checkpoint=15

[power]
battery=no

[settings]
username=dylskee
team=32
passkey=
asknet=no
machineid=2
bigpackets=big
extra_parms=-gpu 1
local=414

[http]
active=no
host=localhost
port=8080
usereg=no
proxy_name=
proxy_passwd=

[core]
priority=96
cpuusage=100
disableassembly=no
nocpulock=1
checkpoint=15

[power]
battery=no

dylskee said:

I've got a monitor for each GPU, I've extended the desktop and tried everything that Norcalsteve has done and got the same results. Something kind of strange though, since I made that post HFM is reporting between 8200 and 8700 PPD for each card now. Not sure what I did but I have never seen both cards producing that much. Usually 1 GPU will hit about 7,000 and the other only 4,000. I'll keep my eye on them and see when the PPD drops.

Click to expand...

torin3 said:

If I remember right, you recently moved your GPU's priority from 0 to 96 (idle t low).

Is it possible that the change in PPD took place around the same time?

If you really want to test, you could switch back for a bit, but I wouldn't suggest that. But running the GPUs on idle priority before sound like might be responsible for your prior performance.

Click to expand...

ChasR said:

Are you running in SLI?

Add -gpu 0 to the first config in extra_params.

Do you have two monitors attached, one to each card, or a dummy plug on the 2nd card? If not, you have to run in SLI and won't make as many points.

You have priority correct on the gpus.

I presume you have the SMP client set at idle priority?

Click to expand...

Norcalsteve said:

Well, i have SLI disabled too, but no dummy plug, but Windows 7 still lets me extend my monitor to a second desktop on the card and the -gpu 1 (card #2) client sees that i have it up too. Even though its disabled and "unlinked" via the NV panel, should i just physically unlink them too? anyone got a link for a "build it your self dummy plug from old monitor cables?" lol, i will order one now.

Also, i guess the "NV_FAH_CPU_AFFINITY value 15" did not work in Environment Variables. i will use a "Affinity program" i been reading about online and have that prog auto set it for me.

Click to expand...

Norcalsteve said:

Also, i guess the "NV_FAH_CPU_AFFINITY value 15" did not work in Environment Variables. i will use a "Affinity program" i been reading about online and have that prog auto set it for me.

Click to expand...

Norcalsteve said:

Found a MAJOR issue with that Variable... a value of "15" for a core i7, is cores 1,2,3,4... That Value is used for dual cores up to hyper threading. so it turns off affinity for cores 0, 5,6 and 7. Any body know the affinity value for a core i7 for all 8 cores?

Click to expand...

Norcalsteve said:

AHHH HAAAA!!!!
OK FOUND IT!!!

For all you i7 GPU folders:
Environment Variable:
NV_FAH_CPU_AFFINITY Value 255

The 255 value on a core i7, locks all cores in for use!

WHOOT!

Value 15 is for you quad core guys

Click to expand...