problems folding on gtx295's

[Fitseries3]

hey guys...

i managed to get gpu2 running on 3 outta 4 of my cores on my 2 gtx295's.

the 4th will start but then gets one of the following errors...

"UNSTABLE_MACHINE"
"EUE limit exceeded"
"connection closed"

or it will keep looping saying that "results where sent" but nothing has even been done.

the other 3 gpu's are fine and the 4th is configured exactly as the other 3 are.

machine IDs are set to 10, 11, 12, 13 for the cards 0 1 2 3

i have the correct -gpu 0, -gpu 1, -gpu 2, -gpu 3 flags amended and the targets point to 4 separate gpu folders.

i have tried different drivers. none of the cards are OCed. fan is at 60% and temps dont go above 70c max.

im running vista 64bit

any help?


EDIT: i can only get 2 gpu's to fold now. same problem with the other though.

 

[WhiteLotus]

The problem on the same card, or just the second core on both?

 

[Fitseries3]

depends really.

sometimes cores 0 and 1 will run but not the others

sometimes cores 0 and 2 will run but not the others

kinda random really.

 

[Solaris17]

I can shed some light on this. Due to vistas single driver issue this is impossible to accomplish however you can get it to do it though their are still alot of bugs. Vijay and the stanford team is trying to figure out how to do this more stabily. but with all the smp and various other high performance clients. they are on the beta page for a reason. first you need to disable SLI this cannot be enabled when running more than 2-3 GPU's.

you then need to connect a KVM switch or multiple monitors. and set it to extend the desktop. this shoulde allow the GPU's to fold without problems.

 

[wolf]

I can shed some light on this. Due to vistas single driver issue this is impossible to accomplish however you can get it to do it though their are still alot of bugs. Vijay and the stanford team is trying to figure out how to do this more stabily. but with all the smp and various other high performance clients. they are on the beta page for a reason. first you need to disable SLI this cannot be enabled when running more than 2-3 GPU's.

you then need to connect a KVM switch or multiple monitors. and set it to extend the desktop. this shoulde allow the GPU's to fold without problems.

good to know 295 owners unite!

 

[Fitseries3]

i have 3 dummy dongles and one monitor. all the displays are marked active.

SLI was disabled a long time ago. the bridge isnt on the cards either.

it will run the client but just fails to work on anything.

 

[Solaris17]

i have 3 dummy dongles and one monitor. all the displays are marked active.

SLI was disabled a long time ago. the bridge isnt on the cards either.

it will run the client but just fails to work on anything.

as far as i understand you need dongles on all the connections. because each one can disply to a monitor. regardless of weather or not they are shown as active. when i disable SLI on my rig i have the option to pick between 8 diff desktops. however F@H will not run if it does not detect it. you need a dummy dongle on all monitor outs of the vid cards.

 

[Fitseries3]

did i not just say i do have a dongle on each?

maybe you need a pic for reassurance?

 

[Solaris17]

did i not just say i do have a dongle on each?

you said you have 3


you need 4

let me reitirate.

i have 3 dummy dongles and one monitor..

 

[Fitseries3]

why 4? i have 3 and a lcd. thats 4

there is no where else to plug any more in

 

[Solaris17]

why 4? i have 3 and a lcd. thats 4

there is no where else to plug any more in

do you have a vga input on your monitor?

EDIT: dinner

 

[WhiteLotus]

Reading some other forums which post similar questions that you are, the only answer i can give is that the FAHcore is not fully optimized for the secondary GPUcores

what kind of drivers are you running the cards on?

Although i will continue to search!

 

[Fitseries3]

im running the suggested cuda drivers 181.20

and i forgot i have 2 lcd's and 2 dongles...

 

[WhiteLotus]

Fit, are you running a -forcegpu nvidia_g80 flag on your clients? if not then try it out

 

[Fitseries3]

no... why would i do that?

 

[WhiteLotus]

no... why would i do that?

no idea why, but a guy at Hardforum says that you should.

10e [H]ard|Gawd, 2.6 Years

Hey Nitro.

I responded to your other thread titled "GTX 295"

You may try adding the -forcegpu nvidia_g80 flag in your command line for each client with the -gpu 0 for client 1 and -gpu 1 for client 2, also ensuring the machine ID is unique (as mentioned).

If that doesn't work, you may have a bad second GPU. I have two GTX 295 cards folding in one box successfully with a 1000w PSU...

 

[Solaris17]

lol o fit.

 

[Fitseries3]

got 0 1 and 2 running but 3 says no way....

each one has "-gpu # -forcegpu nvidia_g80"

 

[Fitseries3]

any other ideas?

 

[DaedalusHelios]

I tried but gave up. GTX 295 lacks the proper support IMO. It was the main reason I obtained a GTX 295 in the first place, for folding.

I got rid of it and I am now back to a GTX 280. I figure it will be easier with a single GPU.

 

[Solaris17]

the gtx will do it its relatively easy with dual GPU cards its just running more than 2. fit try running the force gpu tag without the core specification.

 

[WhiteLotus]

try deleting your work folder and your queue file in each client folder...

 

[Fitseries3]

I tried but gave up. GTX 295 lacks the proper support IMO. It was the main reason I obtained a GTX 295 in the first place, for folding.


I got rid of it and I am now back to a GTX 280. I figure it will be easier with a single GPU.

it was actually really easy to get it folding.

the gtx will do it its relatively easy with dual GPU cards its just running more than 2. fit try running the force gpu tag without the core specification.

wont work.

try deleting your work folder and your queue file in each client folder...

did that 100 times.

 

[WhiteLotus]

well im out of ideas. I have no more ideas.

I still think that the client just has a problem running on the 295. Maybe they will update the core soon and it'll fix it.

 

[Fitseries3]

i even tried deleting the whole gpu3 folder and copying the gpu0 folder and renaming it gpu3 and deleting the work folder as well as the log file, and que.

doesnt work.