GPUGrid and SLI

[Plan.B]

If you are running, or have attempted to run, GPUGrid on an SLI setup, please share your experiences, tips, advice, and 2 cents. I've been looking at a x2 gtx280 setup for better gaming fps over a single gtx280, and for the fun and challenge of my first sli rig, watercooled, but have been reluctant to pull the trigger, which would place me clearly outside "the budget" and incur the wrath of the warden. If it will also successfully run GPUGrid at roughly double the output of a single gtx280, then I probably won't be able to stop myself. But seriously, I'm looking for some wisdom on this topic. Do GPUGrid and sli play nice together, and what kind of output over a single card can be expected? Thanks

[OldChap]

I thought we're waiting for grid to move to cuda 2.3 for this to work "seamlessly" however this might interest you

http://www.gpugrid.net/forum_thread.php?id=1270

[Kondik]

Works


Here

190.38 drivers SLI on eVGA SLI Patch 14 Installed , two GTX 285's , WIndows 7 RTM 64 bit
When I want to play I exit the client and play SLI on, then turn off the game , turn on the B client on and it works.

http://img29.imageshack.us/img29/8819/bionic.th.png (http://img29.imageshack.us/i/bionic.png/)

[Gamekiller]

Sli + gpugrid + gtx 295 + 190.38 = working fine. Good luck.

[Plan.B]

Thank you gents

[INFRNL]

190.38/SLI working for me as well. no errors that I have seen.

Looks like wus take longer to complete.

MUST have sli enabled for grid. In oc's linke a guy said if you don't grid will only detect 1 card....If I remember correctly; I couldnt get the grid to see any cards at all untill I enabled SLI. Really sucks cause I have 3 monitors and only want SLI If I choose to play a game

EDIT: I was just looking at my stats: 190.38+SLI= Junk.
Unless the system still needs to be updated; with 2 gpu's in about 20hrs time I have only completed 3 wu's for a rough total of 15kish points. That is less than or about half of normal production :mad::mad::mad:.

If this is the case; I am going back to 186.18 drivers. I will double check later to see, but this is unacceptable.

[PoppaGeek]

Found this on GPUGrid forums and thought might help guys with multi-gpu systems. (http://www.gpugrid.net/forum_thread.php?id=1272&nowrap=true#12061)


Decided to make some relocation of GPU's on the MoBo (ASUS WS Evolution, having 4 PCI-E slots). After movement of GTX 260 to another slot (closer to other GTX), strated BOINC again & suddenly found, that all GTX 260 is in the system & recognized by BOINC. Very happy. That means, that PCI-E IRQ (or else) distribution managed by MoBo is quite important for systems with multiple GPU's.

[INFRNL]

ok; I have a slight update to my problems I think I have accidentally been posting in a few different threads :ROTF: :shrug:...Just trying to keep you guys on your toes

I am now currently running 6.10.0 on my main rig (dual gpu's), 190.62 drivers, and precision 1.8.0.

At first it was acting weird; I could see that the clocks were running at half speed or less (probably 2d speeds), boinc kept some of the previous wu's (cuda) and was running them. I suspended them and finally got the cuda23 wu's. started running 2 wu's, seemed to be ok. Something happened and said driver failed and had to restart. So I lost those wu's. only had 1 wu left ready to go. for some reason boinc only downloaded 3 cuda23 wu's and would not get anymore. So I thought maybe I need to abort the reg cuda tasks. now it says reached max of 10 wu's per day. so I am left with only 1 wu to run; so tomorrow will be another crappy day for this rig.

So, now to figure out the half speed deal on precision? shut everything down rebooted. now it seems to be running fine, but is that cause I am only running 1 wu now? or will it actually be ok. I guess I will find out when the server allows me more wrk.

They really need to get rid of the max wu per day deal. If I can get this running; I think I will run this setup on my triple gpu rig; hopefully I can figure out 3way sli and gpu grid. I saw someone having problems with 3-way sli on another forum in one of PG's links. :shrug:

I also need to figure out how I can run 3 monitors with the newer drivers since sli HAS to be enabled. It will be nice just to have 1 monitor hooked up to ea rig, but need 3 for main rig....possibly with 3-way sli or sli and 1 xtra gpu?

I will keep you guys posted on my findings

I do know there is a ppd hit on multiple gpu's on a rig though, but its probably not worth it to build more rigs to run just 1 gpu per rig. Although it would help out on WCG :ROTF: :yepp: :D

In my situation I would need to add 3 more rigs for that config, and I do not think my wife would be happy. maybe If I got a low consuming rig it would not add much more, but she is already :mad::mad::mad: about the utility bill :rofl:.

Lastly for this post; sorry to drag on. I wonder if there would be some power savings with running rigs in sli compared to individually running the cards on the same rig?:shrug: Theoretically sli should consume less power over 2 separate gpus, correct?

[INFRNL]

Todays update: I got 1 wu to finish in normal time on 6.10/190.62/sli/1 gpu. Looks like it errored out on its next wu, and it finally started the second gpu. For some reason when it runs 2 gpu's it throttles both cards way way down. I went to disable the second monitor to see if that did anything and errored out th 2 wu's it was working on.

1 gpu= works great
2gpu's. 15+ hrs to complete 1 wu

If I could figure out how to force the 3d setting; that may resolve my issue.

[INFRNL]

I give up. Can only get 1 card to run full speed, even when forcing 3d. I have lost a good 50k in the last 2 days due to trying to update. Back to 186.xx drivers for me with trouble free production. of course gpugrid will screw me when I switch over and not allow any new work till tomorrow sometime. oh well, gotta get back on track

[PoppaGeek]

I thought I would see how F@H would run on the 216 shader 260. The one that runs 190.38 and GPUgrid without error. Nope, can not get F@H to run on it. How/why did Nvidia GPUs get so damn picky. the 192 shader 260 runs 182.50 drivers and F@H without error but no other driver. The 216 GPUgrid only and 190.xx fine.

Fine with me one on each project but geeze no flexibility at all.

[INFRNL]

I do not know whats going on either PG. I guess as long as your producing, its a start.

as I mentioned, went back to 186.xx driver but still on boinc 6.10.0; said I maxed out on wu's for the day, bastards.

With all the messing around I have been doing has screwed up my system too. Good time for a fresh install. might go to win7, not sure yet. I am afraid to try 190.xx drivers on my win7 rig; do not want to lose a bunch more work. by tomorrow; I will have lost approx 75k in work. Should have went with that saying...if its not broke, don't touch it! Oh well, live and learn I guess; who would have know I would have so many issues. I should hopefully be back on track for tomorrow. This weekend I think I will do a fresh install and maybe check over everything on my system.
I would like to try and oc it a touch more too, but its at a comfortable level; maybe one day when I am really bored, or possibly on a rig with only 1 gpu.

[PoppaGeek]

I really wonder how much it is worth it to fight with a card that only wants to do one thing. I know what you are trying to do and I admire tenacity but sometimes I think ya gotta realize ya can't have what ya want. Of course sometimes stepping away and getting it off your mind then coming back later and starting from scratch will get ya there.

We had a modem bank we needed to bring up for off site people to dialin. We had a name brand rack of modems and I needed to hook it into a server running terminal server. I wanted to use a Unix although it was supported by NT 4. There were drivers for both Linux and FreeBSD. I spent 2 weeks compiling and recompiling kernels, reading forums and FAQs and setting IRQs moving stuff around, wiping it all and starting again. Ended up with NT 4. Nothing I could do could get it to work even though it should have under Linux or Freebsd. Worked fine under NT 4. Had to reboot NT 4 ever other day due to memory leaks but it worked. I had a rep for tenacity but it comes to a point where it is costing you and you are getting behind in other things.

Anyway tenacity had always worked for me and I was sure it would again, I just had to keep trying. I am not fighting this 192 card anymore. If it gets where I no longer can run a project on 182.50 drivers I'll move the card to linux or maybe FreeBSD with Linux Libs. Or sell it.

Whatever ya decide good luck!

[PoppaGeek]

If you have 2 GTX 260s in one machine is it required they be in SLI now? I have all AMD/ATI chipset machines, they do not do SLI. Will GPUGrid use them if NOT in sli?

[OldChap]

Although I never got my 260's to complete any more than about 20% of the Wu's they would run fine setup as per Fah on x38 and xp sp3 6.6.37 and 190.37/8 and Grid recognised both

[INFRNL]

They should run without SLI, unless I missed an update somewhere. shouldn't be any different than my 275's or any other set of cards. I run 2 on main rig non-sli, and 3 on another: non sli and gpu grid runs both just fine minus my other small problems. I actually have problems running them in SLI; however I installed the new 6.10.3 boinc last night, not sure what they did different, and not sure if they fixed the downclocking to 2d clocks? If they fixed that issue then I may try SLI again. will have to browse their forum again to check on progress about sli issues

[PoppaGeek]

I thought the 190.62 drivers required SLI.

[INFRNL]

thats right, I forgot. is that driver related or boinc related though? you could always back to 186.18 drivers, but cannot recall if you had issues with that driver

[jcool]

Dual GTX 280 on 190.62 doesn't really work here. It runs 2 WUs on both cards for 1-2 minutes, then pauses. Then starts again. WTF...

[OldChap]

jc are you using an sli capable MB? sli on?

[jcool]

Yeah, an x58. SLI on.

[OldChap]

What are you seeing in display properties ...attached? 2 screens?

What does riva see? (if you use it)

I ask because I've had a situation where prior to sweeper and a re-install of drivers I had 4 screens showing in each even though I could not extend to 2 of them and folding was flakey

[jcool]

Since when do you need to do that extend screen BS with GPUgrid? I thought that was only for FAH?
Anyway I'm off the machines now, gonna check the display properties tomorrow.

[jcool]

Ok here's the situation:

http://database.he-computer.de/Bilder/temp/gpugridwtf.jpg

SLI is on, both GPUs recognized, and running - sometimes. "Verdrängt" means pushed aside by other WUs - it runs GPUGrid for a few minutes, then stops, then starts back again. Seems to me like some BOINC setting issue?!

Monitors displayed is only one active, the 2nd inactive, no 3rd or 4th.

[OldChap]

Looking at yours and comparing mine the only difference I see is that I have Phys-X off.

[Plan.B]

Here's my GPUGrid SLI setup so far... The rest of the system is ready to go

http://i74.photobucket.com/albums/i256/truckstopcbrambo/CIMG1531.jpg

[INFRNL]

Anyone running 190.xx and triple sli yet? just wondering if anyone has it working, I might try to see if I can get it to work, but I do not think the mobo I have 3 cards in came with a 3-way bridge and the spacing is different from the bridge I have for another mobo. will have to see what i can do. might try it in the next couple of days and post my findings

[INFRNL]

Todays update. I really need help from you multigpu guys. I feel like I am in this by myself :ROTF:

I am thinking of going back again. I thought everything was going well this time, but looks like it may not be.

It seems as though sometimes it might work and other times it doesnt. My daily ppd for my dual gpu rig has dropped in half again. Everytime I look it seems that the wu's are completing in around 7 hrs. Sometimes when I check a few have been running longer than normal though.

when I look at my stats on grid, I am still getting bonus's, very few errors or redundant wu's. At the same time I thought the rig is completing more than what shows up on the grid site, almost like some are getting lost :shrug: thats or I am imagining things :ROTF:.

Bottom line is I do not think they have all the bugs worked out, so I think when the last few wu's are done, I will revert back to 186.xx drivers. I am losing 20k ppd right now on avg, maybe not quite that much, but enough to not justify runnin grid & sli.



EDIT: I wonder if it is because I recently started having an issue where boinc would stop some wu's in the middile of things and start other wu's. could be as that would hurt my production. Yesterday looks a little better, but not sure. I have a few wu's left and set for NO new tasks. Maybe when these are done, I can get more and see if it cooperates correctly. Then if I still have issues I will revert drivers back and not run sli.
Does anyone else have any thoughts, ideas, experience with the same thing? I do not even know how many of you are running multi-gpu rigs with sli :shrug:

[PoppaGeek]

I had a couple of WUs missing too, they did show up though. It looks like most of the time the WUs show up on the web site as soon as they are reported. So when it was showing 2 less than I expected I was getting upset. Things were finally quiet and now missing WUs! I did updates on the project, everything reported. Started to record WU numbers and compare. A couple of hours later they showed up. Some lag in their system maybe. But I would wait a day before I called a MIA on WUs and see if they show up. I also checked the forums to see if there problems and someone was reporting missing WUs and had the numbers/names of them. So it happens.

Since I have all AMD MBs I cannot do SLI and there fore cannot offer any help my friend.

[shadowwind]

Todays update. I really need help from you multigpu guys. I feel like I am in this by myself :ROTF:

I am thinking of going back again. I thought everything was going well this time, but looks like it may not be.

It seems as though sometimes it might work and other times it doesnt. My daily ppd for my dual gpu rig has dropped in half again. Everytime I look it seems that the wu's are completing in around 7 hrs. Sometimes when I check a few have been running longer than normal though.

when I look at my stats on grid, I am still getting bonus's, very few errors or redundant wu's. At the same time I thought the rig is completing more than what shows up on the grid site, almost like some are getting lost :shrug: thats or I am imagining things :ROTF:.

Bottom line is I do not think they have all the bugs worked out, so I think when the last few wu's are done, I will revert back to 186.xx drivers. I am losing 20k ppd right now on avg, maybe not quite that much, but enough to not justify runnin grid & sli.



EDIT: I wonder if it is because I recently started having an issue where boinc would stop some wu's in the middile of things and start other wu's. could be as that would hurt my production. Yesterday looks a little better, but not sure. I have a few wu's left and set for NO new tasks. Maybe when these are done, I can get more and see if it cooperates correctly. Then if I still have issues I will revert drivers back and not run sli.
Does anyone else have any thoughts, ideas, experience with the same thing? I do not even know how many of you are running multi-gpu rigs with sli :shrug:

im not sure if you are running the bionic 6.6.36 or not if you are their is a run time bug that is taking longer to complete the work units its assigning work for the main card or slot 0 and when slot 1 gets it then it starts off at a crawl. i was reading it on the gpugrid the thread is run time. so wu are slower because of the bug its not your cards fault.:up:

[INFRNL]

Thanks PG

Shadow: I am currently running Boinc 6.10.4 getting ready to try 6.10.5. I am also on 190.68 or whatever the lastest driver is.

I went to check the system a little bit ago. I had a wu@ 37% or something like that. boinc put it on hold and started another on, bastards. They all have the same download and due times, I dont get it. I will wait for them to finish, upgrade boinc and see what happens. worst case I set to no new tasks, finish some wu's tomorrow and revert back to 186.xx drivers.

[shadowwind]

Thanks PG

Shadow: I am currently running Boinc 6.10.4 getting ready to try 6.10.5. I am also on 190.68 or whatever the lastest driver is.

I went to check the system a little bit ago. I had a wu@ 37% or something like that. boinc put it on hold and started another on, bastards. They all have the same download and due times, I dont get it. I will wait for them to finish, upgrade boinc and see what happens. worst case I set to no new tasks, finish some wu's tomorrow and revert back to 186.xx drivers.

sounds good. i had to put my new build on hold i bought the wrong psu i need an 8 pin for my cpu then when i was talking on phone i decided to try and send back the 2 gtx 280s for a gtx295 all i got to say is the egg rocks :up:

[INFRNL]

Newegg customer service has always been great to me. The only gripe I have now is their pricing on hot/popular items. I also wish they still used fedex supersaver shipping. I have to order on sunday, possibly monday to get package by friday.

Good luck with the 295 and new build. keep us updated

[shadowwind]

i will keep you updated when you see me getting ppd im hoping to get into the top 10 and get 5 of us in their but the way my work units are bouncing up and down who knows.:shrug:i cant count on getting the same rac per day adding 3 more cards 2 gtx 285s and 1 gtx 295 to what i all ready got going so its a wait and see what i get for points deal.:D

[INFRNL]

I know what you mean, My one rig is up and down and down and down and up :rofl: :shrug:

I am now on latest drivers and boinc 6.10.6. still having issues. I have been running 2 wu's for 4 hrs now and barely to 19%. I shutdown precision and restarted, restarted boinc; now boinc has paused 1 wu and started another.

Crap just took another look, it pause the other wu it started and started yet another wu; WTH :mad: :shakes: :down:

So it looks like I will finish out these wu's and revert back. I need to finish most of them otherwise I will have to wait 24 hrs, although I wouldnt be losing much; maybe put them on F@H for a day.:shrug:

Either way I am going to revert back tomorrow. I have 2 cards producing 1 cards worth of production :rofl: :ROTF: :shakes:

[shadowwind]

the 6.10.6 isnt that the new beta driver for bionic?rigt now im installing fedora 11 to try it out on the gpugrid if i can get it working i seen your post for a linux distro go ubuntu its easy to use and poppa or my self can walk you threw it. just waiting to see what the old man is going to say.:D

[PoppaGeek]

You two wanna quit talking so loud, ya woke me up. What are waiting for Shadow?:wth:



:rofl:

[shadowwind]

im waiting for you he asked you the question in the evga thread.:D
wow i woke you up from your nap.:rofl: i need to download some drivers not sure im liking fedora 11 i may need to get use to it kind of feels like the old solaris hmm might be good after all, a least i could navigate that os.:D

[PoppaGeek]

I used Fedora alot long time ago. Very good and stable. I was admin on many very big boxes running Solaris. Very sophisticated OS. AIX is impressive as well.

[shadowwind]

so far its a pain to find things in it but i did find bionic im going to take my time and look around before i try and tackle nvidia newer drivers right now im on limited display drivers
i didnt install any off the dvd.:D so i shouldn't have a lot of crap to un-install later on i hope.

[PoppaGeek]

I played games running Fedora for a good while and I do not remember the Nvidia drivers being hard to install. Seems to me it was pretty easy. One person in GPUGrid forums said it was easier to install the Nvidia drivers downloaded from Nvidia than using the package manager.

[shadowwind]

hey pops had it working on the grid for 2 hours then error out i got something wrong in the kernal or didnt add some thing so back to start and check it over 3 times before i push enter this time.:D at least i know it works.

[Plan.B]

so I got the gtx280 SLI setup running. it's in bleed/test mode, and working as it should. after I flash my vertex to the latest firmware will be the fresh os install, then drivers, then wcg & gpugrid crunching (and gaming of course). I'm looking forward to cranking out some wu's

http://i74.photobucket.com/albums/i256/truckstopcbrambo/CIMG1540.jpg

[shadowwind]

nice job Plan.B are you going to permanently mount your fill tube?
but it still nice work:up:

[Plan.B]

mount fill tube? No. once the all the air is bled from the coolant I will turn the red knob (ball valve) to close the system and remove the fill tube.

[shadowwind]

ahhh i didn't see a ball valve . but now i got a better picture of what its going to look like when done. cant wait to see it in the finish stage pictures.:up:

[Plan.B]

build finished and working. now onto gpugrid and wcg crunching

http://i74.photobucket.com/albums/i256/truckstopcbrambo/CIMG1541.jpg

[shadowwind]

nice work. :up:now if you got some time you can come do my rig:D i got to many projects going and i need to get back to the brake presses to finish of my heat project for venting heat. water cooling is the way to go but if you got 5 or more rigs that can get very expensive.:)

[Plan.B]

nvidia driver 190.62 working for anybody in sli? Thanks

[Plan.B]

190.62 did not work for me - bionc client didn't recognize my gpu's. using bionc 6.2.28 with nvidia 190.38 now. seems to be working, at least it recognizes my hardware, but message log keeps telling me "no work sent" from server. I'm going to let it go for awhile and see what happens.

Edit: now it's reporting no nvidia gpu present, with sli enabled and sli disabled - in vista64. At least sli & games are working properly. Will try again tomorrow.

[Plan.B]

gtx280 sli + vista64 + nvidia 190.62 + bionc 6.10.5 = computation errors

[Plan.B]

wcg/gpugrid/sli setup = i7 920 + gtx280 sli + vista64 + nvidia 190.62 + bionc 6.10.5

cpu settings at 80% of processors and 80% cpu time. I now have 2 tasks running, each one running on 0.14 cpu and 1.0 gpu; while 8 wcg tasks are running. estimated time for the 2 gpugrid tasks is around 2.5 hours. seems to have fixed the computation errors, should have figured this one out sooner. any advice on these settings? Thanks

Edit: I can tell from watching the cpu/gpu temps the tasks are cycling on and off quite a bit. wcg seems to be active only about 1/2 the time. is this a resource share setting issue? right now wcg and gpugrid are set at 50% each. advice on this setting? Thanks

Edit: estimated time to completion for gpu tasks has risen to 3+ hours. I know these estimates should not be given too much credit, I will wait and see what happens.

Edit: computation errors

[shadowwind]

can you try out 1 vid card out and disconnect the other card to see if it completes a work unit on its own , just curious if you can complete a work unit on 1 card.after try the other card out so we can say its not the vidcards if both work, i know its a pain ars to pull cards with them set up in water cooling but we should rule out the cards first.:up:

[Plan.B]

great idea. (hehe duh on me) I will try that and report back. Thanks.

[06F150fx4]

MY GTX 295 works fine on gpu 1 but gpu 2 errors out. GPU 2 worked fine on folding and works fine in games.

[shadowwind]

MY GTX 295 works fine on gpu 1 but gpu 2 errors out. GPU 2 worked fine on folding and works fine in games.

thats the down side of the 295s on the grid only a few people
i know of have that same problem as you have its a bug like
the 260s have but the good news they are aware of it and trying to fix the problem, time will tell,:up:

[OldChap]

Have you tried a later boinc

[Plan.B]

It looks like 6.10.13 is working.

[OldChap]

Keep us posted on this ...good news though so far :up:

[Plan.B]

computation errors. the video driver "stops responding" (as per the vista error msg) for a split second, creates the computation error, stops the gpugrid wu. been trying different combos of vid drivers and bionc client. now I must try shadowwind's suggestion of physically 1 card at a time

[shadowwind]

im hoping that works for you PB, that galaxy PSU you got what is it rated at?
i had a similar experience with a psu going bad just moving my mouse would cause it to error out, it may sound weird to you but when i changed it out things started working.so im thinking 1 card should work.:up:i got my fingers crossed things will work out.:)

[Plan.B]

its an enermax galaxy 1000W. the cards run fine in 3+ hour overclocked sli gaming sessions. the error usually happens in the first 15 minutes of a wu. I thought it was the bionc screensaver kicking in, disabled all screensavers and power management stuff, not the problem. It's not temps, low volts to the gpu's, or overclocked settings (mobo or vid), tried those. I'll keep digging...

Edit: I keep wondering if there is some kind of throttling going on with the vid cards. I've been researching the cards and looking through rivatuner, can't find anything

[Plan.B]

ok. bionc 6.6.41 + nvidia 191.07 + vista64 + 2x gtx280 + sli enabled = over 1 hour of gpugrid crunching with zero errors (while also crunching wcg). since all my previous problems were popping up in the first 15 -20 minutes, I think I got it. wooters

Errors. F*ck it.