Guide to running multiple GPU work units

[mstenholm]

Well, came home from work to 2 rigs sitting idle with no work. It's back to a single instance for me.

I gave up as well. To much GPU time wasted.

 

[[Ion]]

Well, came home from work to 2 rigs sitting idle with no work. It's back to a single instance for me. I should have my Son's spare HD6850 in a week, so i might try it on that card.

Well, given your challenges, I've decided to just leave my cards as-is. Stability and reliability is more important to me--I just don't have the time to babysit them.

 

[t_ski]

Driver crahses:
If you are having driver crahses, the following registry modification might prevent that:
(Source: Bun-Bun from XS)

Windows Registry Editor Version 5.00

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Watchdog]
"DisableBugCheck"="1"

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Watchdog\Display]
"EaRecovery"="0"

I can't seem to find this on my system? I have no "watchdog" key

 

[mstenholm]

I gave up as well. To much GPU time wasted.

And I gave it a second change. GPU x 3 and GFAM on a i7-920.
When I ask not to receive new work it reports right away.
When I ask to receive new work the client doesn't report (set to report every 0.0 day) and hence no new work is downloaded.
When I ask to do only 40% CPU (just 3 CPU aiding the 3 GPU WUs) it still download pure CPU work. wtf? I can't download enough GPU WUs to last a night so I have to lose all WU and run 1 WU during nights.

If I leave the rig unused it does a GPU WU in 3:00 in an average. With only 1 GPU at the time it did it in 2:20 so a nice boost.

Next step must be to try the GPU only version.

Update - installed app_info (HCC and HCC version) on my 680 rig with a two GPU WUs setting. More to follow. Updated: It ran out after two hours. Back to 1 unit again.
Update 2: My 7770 just did auto send and update when it ran out of GPU work. Let's see if it works all night. Updated: It played nice all night I think.

 

[KieX]

And I gave it a second change. GPU x 3 and GFAM on a i7-920.
When I ask not to receive new work it reports right away.
When I ask to receive new work the client doesn't report (set to report every 0.0 day) and hence no new work is downloaded.
When I ask to do only 40% CPU (just 3 CPU aiding the 3 GPU WUs) it still download pure CPU work. wtf? I can't download enough GPU WUs to last a night so I have to lose all WU and run 1 WU during nights.

If I leave the rig unused it does a GPU WU in 3:00 in an average. With only 1 GPU at the time it did it in 2:20 so a nice boost.

Next step must be to try the GPU only version.

Update - installed app_info (HCC and HCC version) on my 680 rig with a two GPU WUs setting. More to follow.
Update 2: My 7770 just did auto send and update when it ran out of GPU work. Let's see if it works all night.

What did you do differently to get it to work? Could be useful to know in case someone else has same problem.

 

[manofthem]

So I made my app_info.xml and followed the steps. This is what I'm seeing.

I'm getting en error in the Even Log, saying "[error] app version returned from anonymous platform project; ingnored"

So i'm obviously not getting any work on my gpus, so who can tell me what i'm doing wrong?

Please keep in mind i'm on Win8 x64... thanks

 

[t_ski]

That's the same error I was getting until I copied the text from the CODE quote in the beginning of the thread. Something was happening from what I was copying from ThE_MaD_ShOt. I think that there could be an extra space or something throwing if off just one character.

PM me your email address and I can email you my app file.

 

[t_ski]

OK, so I came home from work and I had more stuck WU's (one of which was for over 7 hours ), so I paused everything and closed the client. I disabled Crossfire and relaunched the client. It ran for about 5 hours so far without any more stuck WUs.

Keeping my fingers crossed that it's fixed...

 

[F150_Raptor]

So I made my app_info.xml and followed the steps. This is what I'm seeing.
http://img.techpowerup.org/121102/wcg1.png

I'm getting en error in the Even Log, saying "[error] app version returned from anonymous platform project; ingnored"

I have that same error in mine but it's not effecting any thing.

So i'm obviously not getting any work on my gpus, so who can tell me what i'm doing wrong?

Please keep in mind i'm on Win8 x64... thanks

Click on tools and computing preferences, then make sure you have the box checked for use gpu while computer is in use. Those wu's you have are for gpu's.

 

[manofthem]

I have that same error in mine but it's not effecting any thing.


Click on tools and computing preferences, then make sure you have the box checked for use gpu while computer is in use. Those wu's you have are for gpu's.

OMG, i can't believe I forgot to re-check that! (forgot to after the reinstall) Well, went ahead and checked it, applied, and now it's working. Seems to be taking a long time, however. 4 WUs per card, may have try 3 because it's saying it's going to take 40 minutes each, which is nuts. It was doing 1 every 1:30, now 40min?

edit: that's why, gpu usage is like 0.... gotta try to restart it.

 

[F150_Raptor]

Give it some time and see what it's at when it finishes. That's just a rough estimate for the time, it's not the actual amount of time it'll take.

 

[manofthem]

Give it some time and see what it's at when it finishes. That's just a rough estimate for the time, it's not the actual amount of time it'll take.

I think it's because the driver crashed. It crashes and then gpu usage plummets. So I may try a different driver, go back to 12.8 or 12.9, as the 12.11 beta may be an issue.

driver keeps crashing and usage drops. obviously that's my problem, but now as to why.....

 

[t_ski]

You have two cards in Crossfire, try disabling it.

 

[manofthem]

I'm getting a lot of driver crashes, almost immediately. I disabled cfx, nothing. Tried 4 WU's, no good. 3 WU's, no good. I'll try 2 but I fear that it'll be the same.

I just set it back to 1 each, and it's running normally, each finishing quickly: the last 3 under 1:20. I dropped the clocks back to stock and it seems that they are finishing faster!


I thought maybe it was MSI AB being open, but I tried with it closed and no difference.

edit: seems my cpu is not being used like it used to, so that could account for the speed increase I may have to reinstall the client.

edit2: tried 2 and it had a problem. Driver crashed immediately and then stabalized. Card1 then did nothing, but Card 2 continued to work. Card 1 usage is down at 0, while 2 is sitting up around 98 most of the time.

edit3: driver crashed and it all dropped. I'm going back to normal now, at least it works

 

[mstenholm]

What did you do differently to get it to work? Could be useful to know in case someone else has same problem.

I don't know beside using the HCC and HCC app_info file this time.

 

[Steve-007UK]

Going back to windows 7 so will be offline for an hour or 2. Driver always crashes on windows 8 and i don't like windows 8 much anyway lol.

 

[F150_Raptor]

I'm getting a lot of driver crashes, almost immediately. I disabled cfx, nothing. Tried 4 WU's, no good. 3 WU's, no good. I'll try 2 but I fear that it'll be the same.

I just set it back to 1 each, and it's running normally, each finishing quickly: the last 3 under 1:20. I dropped the clocks back to stock and it seems that they are finishing faster!


I thought maybe it was MSI AB being open, but I tried with it closed and no difference.

edit: seems my cpu is not being used like it used to, so that could account for the speed increase I may have to reinstall the client.

edit2: tried 2 and it had a problem. Driver crashed immediately and then stabalized. Card1 then did nothing, but Card 2 continued to work. Card 1 usage is down at 0, while 2 is sitting up around 98 most of the time.

edit3: driver crashed and it all dropped. I'm going back to normal now, at least it works

Going back to windows 7 so will be offline for an hour or 2. Driver always crashes on windows 8 and i don't like windows 8 much anyway lol.

@ manofthem, it could be windows 8 giving you all the problems.

 

[manofthem]

@ manofthem, it could be windows 8 giving you all the problems.

I believe you are right, I started thinking the same thing. I've been looking around and just read this post from the other day:

If you want to keep running multiple WUs on your GPUs, you might want to stick with 7. Supposedly the newest BOINC (not the WCG site one) that is required for Windows 8 has issues with running multiple WUs. My 7950 crashed like a mother trucker until I removed the app_info.xml.

I really don't feel like reinstalling 7, so I'll let it be as is for now. Hopefully someone will find and post a fix to get multi WU's working well on 8. I really had wanted to get 4 WU's on each card going....

 

[F150_Raptor]

You could always install windows 7 and wcg on another hard drive and only run it when your at work or sleeping. Just swap them when you want to use the rig for some thing else.

 

[manofthem]

You could always install windows 7 and wcg on another hard drive and only run it when your at work or sleeping. Just swap them when you want to use the rig for some thing else.

Yeah maybe I'll give that a go today. Only available drive is an old 64gb ssd that's sitting around; it's not a great drive but it should handle it. I'll report back with what happens.

 

[t_ski]

So if I understand this right, there is one CPU thread devoted to each GPU WU, correct? So if I get a 7770 card for the other rig I was trying to get up and running, it may not be able to do multiple WUs on that GPU since it's a single core Pentium 4. Am I right?

If so, I may have to put it in the kids' rig and take their GTX280 and put it on the P4. That might work out better that way.

 

[Norton]

So if I understand this right, there is one CPU thread devoted to each GPU WU, correct? So if I get a 7770 card for the other rig I was trying to get up and running, it may not be able to do multiple WUs on that GPU since it's a single core Pentium 4. Am I right?

If so, I may have to put it in the kids' rig and take their GTX280 and put it on the P4. That might work out better that way.

Correct- 1 CPU core per GPU so multiple wu's won't work with a single core....

 

[manofthem]

Well, after fiddling with it for a while, I finally got it working. man this stuff can be finicky!

So far it's looking to good. Each card is running 4 WU's and finishing each at around 2:30
Let's hope this proves stable; I'll be monitoring it for the next few hours and then it'll be running all night.

Thanks for your help guys!

 

[manofthem]

I know you guys are getting sick of me , but I appreciate the help
Everything ran well for a while, but then no new work was coming in. SAW the following errors in the Event log, which i'm assuming accounts for it:

Any ideas why it's not working?

 

[ALMOSTunseen]

I know you guys are getting sick of me , but I appreciate the help
Everything ran well for a while, but then no new work was coming in. SAW the following errors in the Event log, which i'm assuming accounts for it:
http://img.techpowerup.org/121103/wcg4.png

Any ideas why it's not working?

Is it 64bit or 32 bit? Looks like it's running with 32 bit.