PDA

View Full Version : Wu's Crash on GTX 560 Ti !!!



STE\/E
03-11-13, 06:54 AM
Downloaded & Installed Driver v314.07 WHQL yesterday for the GTX 560 Ti but the System kept crashing Upon entering Windows 7, so I went back to the v310.90 Driver. The System runs okay with that Driver but the Collatz Wu's keep getting Computation Errors after a few Minutes running time. They ran okay with the GTX 470 that was in the Box but I sold that Card yesterday & replaced it with the GTX 560 Ti ... Either the 560 Ti is bad or the Collatz Wu's don't like to run on it, I'm running the 560 Ti at it's stock speed & the fan on 70 which reads about 60c on load. It's running 2 Wu's at once, maybe that's the problem, I don't know ???

Al
03-11-13, 08:10 AM
Ste\/e, I use 301.42 on my 560ti and they work fine on everything. I've never tried 2 collatz at a time, so you might to start there.

STE\/E
03-11-13, 10:09 AM
Well I'm running 2 because it is a Dual Core Card ...

Al
03-11-13, 10:37 AM
Well I'm running 2 because it is a Dual Core Card ...

Sorry, makes sense now. Doesn't Z have one of those?

STE\/E
03-11-13, 10:46 AM
I think he has several of them ...

Duke of Buckingham
03-11-13, 10:53 AM
Nice :D

http://media.bestofmicro.com/3/0/322812/original/560-roundup-title.jpg

Al
03-11-13, 11:03 AM
I think he has several of them ...

Maybe once he gets out of bed he can shed a little light.

zombie67
03-11-13, 12:07 PM
Sorry, makes sense now. Doesn't Z have one of those?

Are we talking about a GTX 560 Ti 2Win? I have one, yes. It's not running collatz right now. But I will give it a shot and report back.

Slicker
03-11-13, 01:54 PM
Downloaded & Installed Driver v314.07 WHQL yesterday for the GTX 560 Ti but the System kept crashing Upon entering Windows 7, so I went back to the v310.90 Driver. The System runs okay with that Driver but the Collatz Wu's keep getting Computation Errors after a few Minutes running time. They ran okay with the GTX 470 that was in the Box but I sold that Card yesterday & replaced it with the GTX 560 Ti ... Either the 560 Ti is bad or the Collatz Wu's don't like to run on it, I'm running the 560 Ti at it's stock speed & the fan on 70 which reads about 60c on load. It's running 2 Wu's at once, maybe that's the problem, I don't know ???

The collatz CUDA and OpenCL apps were never designed to run multiple instances on a single GPU. The v2.xx ATI apps were built using a customized Brook+ DLL specifically so multiple apps could run at once. That same Btrook+ DLL was used at MW which enabled Collatz and MW apps to "play nice" even before the BOINC client officially supported ATI cards. In other words, there's no guarantee running more than one Collatz CUDA or OpenCL at a time that they will work properly.

The whole reason there was a push to run multiple apps was due to projects with crappy GPU implementations that underutilized the GPU. Since the Collatz apps can be configured such that a WU will use so much GPU that the system is unresponsive, there has not been a need to run more than one app at a time.

However, with the release of the app_config.xml, it may be easier to run multiple apps than to tweak the app parameters. So, the next version of the Collatz apps will support running multiple copies. In addition, while OpenCL performance still sucks in general, the apps will use the CPU to verify the results so that a quorum of 2 is no longer required. The new "solo" apps will be GPU only so I may be tempted to change the OpenCL credit to match that of the CAL/Brook+ apps since AMD has already dropped Brook+ support and will soon be dropping CAL support.

STE\/E
03-11-13, 03:16 PM
Are we talking about a GTX 560 Ti 2Win? I have one, yes. It's not running collatz right now. But I will give it a shot and report back.

Yes the Ti 2Win Z ...

zombie67
03-11-13, 07:18 PM
I was able to crunch the bigger collatz task with my 560 Ti 2win, no problem.

http://boinc.thesonntags.com/collatz/result.php?resultid=136534708

It is on a machine that has the 2win and a 590 (4 GPUs total). Win7 64 (all updates), BOINC 7.0.42, 314.07 driver (boinc shows it at 314.7 in the event log).

STE\/E
03-11-13, 07:43 PM
Okay Z, but did you run just 1 or 2 at a time on the 2Win, and if only 1 how did you do that to restrict the Dual Core to only run 1 Collatz ??? Thanks

zombie67
03-11-13, 07:59 PM
Okay Z, but did you run just 1 or 2 at a time on the 2Win, and if only 1 how did you do that to restrict the Dual Core to only run 1 Collatz ??? Thanks

I am not sure I understand the question. I ran 4 tasks at once on the machine, 1 per GPU. There is no option to run multiple tasks per GPU with collatz. Do you mean using an app_info.xml? No, I don't use one for collatz. You don't need it, as the load is already maxed out. And according to slicker's comment earlier in this thread, the CUDA app is not designed to run more than one per GPU at a time.

STE\/E
03-12-13, 01:52 AM
I am not sure I understand the question. I ran 4 tasks at once on the machine, 1 per GPU. There is no option to run multiple tasks per GPU with collatz. Do you mean using an app_info.xml? No, I don't use one for collatz. You don't need it, as the load is already maxed out. And according to slicker's comment earlier in this thread, the CUDA app is not designed to run more than one per GPU at a time.

What I meant as did you run just 1 Wu on the 2Win or 2 Wu's as in 1 Per Core ... the 2Win I have seems to run 2 GPUGrid Wu's okay 1 on each Core but will Crash running 1 Collatz on each Core & the Computer won't even run with the 314.07 Driver, keeps rebooting when entering Windows, had to revert to the 310.90 Driver ...

zombie67
03-12-13, 02:03 AM
What I meant as did you run just 1 Wu on the 2Win or 2 Wu's as in 1 Per Core ... the 2Win I have seems to run 2 GPUGrid Wu's okay 1 on each Core but will Crash running 1 Collatz on each Core & the Computer won't even run with the 314.07 Driver, keeps rebooting when entering Windows, had to revert to the 310.90 Driver ...

Okay, I am very confused now. Or rather, I think maybe you are? The 2win is a dual GPU card (two 560 Ti), just like the 590 (two 580).

So, I ran just one task on each of two GPUs in my 560 Ti 2win...both at the same time. One task per GPU.

Let me ask you a question: Are you using an app_info.xml? Or are you having these problems completely "native"?

STE\/E
03-12-13, 02:37 AM
I'm not running any app file, the 2Win just wants to run 1 Wu on each GPU on it's own ...

zombie67
03-12-13, 10:20 AM
I'm not running any app file, the 2Win just wants to run 1 Wu on each GPU on it's own ...

Okay, then we are on the same page. I have the same set up, and the tasks complete just fine.

STE\/E
03-13-13, 08:15 AM
I installed the v314.07 Drivers on my other Box with a 560 2Win this morning & it seemed to run alright, have some Collatz running on it now to see if they Crash or not. I decided to try the other Box again with the 560 2Win in it with the v314.07 Drivers & this time the System didn't Crash, could have been just a bad install the first time because I didn't change anything since the last time I tried the v314.07 Drivers ...

Slicker
03-14-13, 12:14 AM
What I meant as did you run just 1 Wu on the 2Win or 2 Wu's as in 1 Per Core ... the 2Win I have seems to run 2 GPUGrid Wu's okay 1 on each Core but will Crash running 1 Collatz on each Core & the Computer won't even run with the

314.07 Driver, keeps rebooting when entering Windows, had to revert to the 310.90 Driver ...

I have a version that is working with the 314.07 drivers. Just finished initial testing today. Next tests are to see whether it can run multple WUs at one time via an app_config.xml. You guys want to help with testing it?

zombie67
03-14-13, 01:16 AM
I have a version that is working with the 314.07 drivers. Just finished initial testing today. Next tests are to see whether it can run multple WUs at one time via an app_config.xml. You guys want to help with testing it?

Of course, I'm always in, when it comes to testing.

STE\/E
03-14-13, 01:24 AM
Yup I'll give them a try too ... Just let me know what to do ...

STE\/E
03-15-13, 09:12 AM
Don't know if I picked up a bunch of the Test Wu's or not but got over 600 Cuda23 Wu's dumped on 1 Box, will have to abort some, don't think I can run that many before their deadlines but not sure about that either ... The Cuda23's seem to be running about 25% Faster than the Cuda40's were on my 560Ti 2Win ...b-)

STE\/E
03-15-13, 02:57 PM
Was still having trouble running the Collatz GPU Wu's on the one Box with the 560 2Win in it. It would only run at about 13% Usage while the other Box with a 560 2Win in it ran at about 87%-89% Usage when running the Collatz GPU Wu's. So to see if it was the Card or the System I swapped Cards . Same thing happened to the Card that was running okay when I put it in the other Box that wasn't running okay, the usage dropped to around 13%.

The Box that was running good was running the cuda23 Wu's while the Box that was running bad was running cuda40 Wu's. So I aborted the cuda40 Wu's out of the bad running box & luckily got some cuda23 Wu's back. The Box seems to be running okay now, only at about 70% Usage but I set the System back to it's stock BIOS setting so don't know if that had something to do with it running okay now too ... I'm running out a couple of GPUGrid Wu's & when they get done I'll set the BIOS to about 4.2 or 4.3 GHZ & see if it keeps running okay or not when running the cuda23 Collatz Wu's ...

STE\/E
03-16-13, 05:13 AM
Both my 560ti 2Wins ran the Collatz Wu's okay over night 2 on each Card, seems if I keep them supplied with the cuda23 Wu's they both run fine and at the same speed if I run the same CPU Wu's on each Box. But if they run the cuda42 Wu's then 1 run's okay but the other one will either get Hung Wu's or Crash some Wu's so I just abort any other than the cuda23 Wu's for now anyway ...

DrPop
03-16-13, 04:15 PM
Is the difference in the two sets of CUDA WUs the amount of "new" CUDA code they contain? For example, one set of WUs should be sent to cards made up to date X, and then the second set of WUs should be sent to only newer cards than date X?
I'm pretty sure Slicker could set a flag or something in his program and make that happen if that's the case - he runs that project with the least hiccups I've ever seen of any projects so far.

Slicker
03-17-13, 06:36 PM
The latest 64-bit windows BETA version of the Collatz app can be found at the link below.

1. Make sure the collatz.config is located in the project folder along with the app_info.xml, executable, and cuda runtime DLL.

2. Make sure the collatz.config contains the line "solo=0" or else your WUs will be invalid. That allows the app output to run in compatibility mode with the 2.xx and 3.xx apps. Once the apps are tested and the new work generator and validator are installed, Collatz will be moving to a quorum of 1 which means instant credit and the line can be changed to solo=1 at that time.

3. The app_config.xml is optional. If you find your GPU running at less than 98% utilization, copy it to the Collatz project folder and restart BOINC so it will run 2 WUs at a time on a single GPU. It seems this is becoming the preferred method rather than allowing users to tweak the app parameters. That allows the project to set the parameters so that it will run on a lowly 8400 GS but still keep the high end GPUs fully utilized.

http://boinc.thesonntags.com/collatz_test/download/test/

If testing of this meets or exceeds the 2.03 app speed (may be a little more difficult to compare if running multiple WUs at a time as they WILL take longer but total WUs completed each day should increase), I will also put out the 32-bit windows app for testing.

I've got CPU apps working for Linux and Windows, OpenCL for AMD are close to finished, so once Linux and OS X apps are ready and tested Collatz will be switching to these new apps across the board. The ATI 2.09 apps will remain for those running HD 38xx thru HD 6xxx GPUs so long as AMD supports CAL since OpenCL performance is still quite lacking.

System Requirements:
Windows x64
CUDA 5.0 drivers (created and tested using 314.07)
BOINC 7.0.xx or later client (if using the app_config.xml)

STE\/E
03-17-13, 08:15 PM
Okay, I'll have to wait until the morning to give it a try ... had company all day & going to bed now ...

STE\/E
03-18-13, 08:01 AM
Well that didn't go over so good, the Computer Locked completely up when putting the Files in the Collatz Project Directory Folder to start with, shouldn't have done that though, BOINC was shut down at the time. Then after taking 15 Minutes to even get the Box to Boot again BOINC promptly crashed over 400 Hours of RNA & CPDN & Lattice Wu's. Luckily the WUProp Wu didn't crash too & maybe I'll at least get the Hourly Credit there for the Wu's if not the Project Credit.

Then once I got BOINC running right again I had to suffer the Indignity of it all lol by the Project refusing to give me any GPU Work, say's I don't have the right things selected & or the Project has no work, something like that anyway,funny I was getting Wu's before I put the files in the Directory. I'm running the 314.07 Drivers & the BOINC Manager say's I'm running Cuda 5.0, I also have NVIDIA GPU selected to use:

Use CPU
Enforced by version 6.10+ no
Use ATI GPU
Enforced by version 6.10+ no
Use NVIDIA GPU
Enforced by version 6.10+ yes
Is it OK for Collatz Conjecture and your team (if any) to email you? yes
Should Collatz Conjecture show your computers on its web site? no
Default computer location ---
Maximum CPU % for graphics
0 ... 100 20
Run only the selected applications collatz: yes
mini_collatz: no
If no work for selected applications is available, accept work from other applications? no

I'll hold off on the other Box for now until I can get work for the Idle GPU in the 1'st Box ...

STE\/E
03-20-13, 04:27 PM
Trying the v314.21 BETA Drivers on both 560ti 2Win Box's, so far the results are good, running cuda40 on 1 Box & cuda23 on the other and both are at least under 1 Hr running time. The one Box was under 1 Hr running but the other was taking close to 2 Hr's to run the Wu's. Hopefully things stay stable overnight, had the one Box shut down twice today already just running at stock speeds ...

Slicker
03-21-13, 04:38 PM
I updated the files in http://boinc.thesonntags.com/collatz_test/download/test/collatz_4.02_windows_x86_64__cuda50.zip (http://boinc.thesonntags.com/collatz_test/download/test/collatz_4.02_windows_x86_64__cuda50.zip) so that it now correctly request work. For whatever reason, when using an app_info.xml with BOINC 7.0.52, it won't request GPU work unless there is a CPU app in the app_info.xml as well -- even if the project preferences say not to use it. It also pretends to be v3.11 with plan class cuda42.

I also got the "solo" versions of the ATI OpenCL version giving correct results so I'll need to find someone with a HD7xxx GPU to test that and make sure it doesn't error out right away.

DrPop
03-21-13, 05:48 PM
...I also got the "solo" versions of the ATI OpenCL version giving correct results so I'll need to find someone with a HD7xxx GPU to test that and make sure it doesn't error out right away.

I think Zombie has one of those GPUs? Maybe Al does too, he might be up for a little testing for you as soon as the Challenge is over in case it crashes. ;)

Al
03-21-13, 06:25 PM
My 7970 has been running Collatz for a couple of days with no error. I'll unhide the computers if you'd like to take a look.

STE\/E
03-21-13, 06:45 PM
My 7970 has been running Collatz for a couple of days with no error. I'll unhide the computers if you'd like to take a look.

What ATI Driver you using Al, I just ran 40 of them with error's in a few seconds, but then anything I run at Collatz seems to crash so it's to be expected I guess ... ;)

Al
03-21-13, 06:53 PM
What ATI Driver you using Al, I just ran 40 of them with error's in a few seconds, but then anything I run at Collatz seems to crash so it's to be expected I guess ... ;)

Ccc 12.10 1.4.1741 I've made the computers viewable if that helps.

zombie67
03-21-13, 07:25 PM
I think Zombie has one of those GPUs? Maybe Al does too, he might be up for a little testing for you as soon as the Challenge is over in case it crashes. ;)

I'll try it out right now. collatz or mini-collatz? Or does it matter?

Al
03-21-13, 07:37 PM
I'll try it out right now. collatz or mini-collatz? Or does it matter?

Collatz wouldn't send the minis to my 7970. Not sure why. I've since changed the preferences to not bother requesting them. Might be different now.

zombie67
03-21-13, 07:44 PM
Collatz wouldn't send the minis to my 7970. Not sure why. I've since changed the preferences to not bother requesting them. Might be different now.

collatz tasks seem to have completed just fine with my 7970s, just waiting for wingmen to

http://boinc.thesonntags.com/collatz/results.php?hostid=123801&offset=0&show_names=0&state=0

Edit: And yeah, the project will not send mini-collatz to my 7970s.

DrPop
03-21-13, 09:19 PM
Thanks Al and Zombie for pitching in, glad to hear the WUs are finishing! SteVe, do you think this is a driver issue or some other conflict on your rig?:confused:

STE\/E
03-22-13, 12:37 AM
. SteVe, do you think this is a driver issue or some other conflict on your rig?:confused:

I don't have any idea DP, just uninstalled the 12.11 Drivers that were in the System & installed the 12.10 WHQL Drivers Al said he was using. Promptly erred a Dozen more of the Collatz 3.11 (opencl_amd) Wu's GPU Wu's.

@All, are you running a app file for Collatz AMD Wu's ??? Post it here or PM it to me if you are so I can try it ... Thanks

zombie67
03-22-13, 12:59 AM
I don't have any idea DP, just uninstalled the 12.11 Drivers that were in the System & installed the 12.10 WHQL Drivers Al said he was using. Promptly erred a Dozen more of the Collatz 3.11 (opencl_amd) Wu's GPU Wu's.

@All, are you running a app file for Collatz AMD Wu's ??? Post it here or PM it to me if you are so I can try it ... Thanks

I am not running any app_info.xml or app_config.xml for collatz. I don't know of any reason to do so.

Are you trying to use one of the manual power apps? If so, then you probably shouldn't be doing that. According to the power app page: "At the present time, the optimized apps listed here are also the exact same as the stock applications which are downloaded automatically. Only the Linux CUDA and ATI applications need to be downloaded here as they are installed manually at present."

STE\/E
03-22-13, 01:19 AM
I'm not running any app myself, just what the Project sends me. The 7970 is running at it's stock speed too. The Wu's error out in less than 2 seconds after starting ... Windows 7 Pro 12.10 WHQL Drivers with NET4 Support ...

zombie67
03-22-13, 01:44 AM
I'm not running any app myself, just what the Project sends me. The 7970 is running at it's stock speed too. The Wu's error out in less than 2 seconds after starting ... Windows 7 Pro 12.10 WHQL Drivers with NET4 Support ...

Hrm. Mine are running 12.10. And the machine has .NET 4.5.

DrPop
03-22-13, 01:48 AM
Hrm. Mine are running 12.10. And the machine has .NET 4.5.

Maybe it's the .NET code that needs updating then? I wonder if AMD uses some of that code in their latest drivers.

STE\/E
03-22-13, 01:56 AM
I'll try & download the NET4.5 later & try it, gonna take a nap now ... :p ... all this Collatz stuff wore me out ... lol

STE\/E
03-22-13, 06:04 AM
Installed Net 4.5, Rebooted, saw a Dramatic improvement of 1 Second before the Wu gave a Computation Error, before Net 4.5 I didn't see any Seconds before a Computation error so we're making progress anyway ... lol

I don't have a clue why it won't run them, the Box runs GPU Grid, Donate, Milkyway without any problems. I tried a different 7970 Box with different Brand 7970 in it than the other Box. I Installed NET 4.5 but didn't reboot, Box has a CPDN Wu on it @ 95% done & 280+ Hr's & didn't know where the last Checkpoint was so will wait for the CPDN Wu to finish in ? Hr's ... I did download 1 Collatz GPU Wu & it promptly erred out like the other Box so a Reboot probably isn't going to do anything to improve it.

I could be missing an Windows Update, I install the OS & do the SP1 Update but then Disable Update's & never do another one so that's a possibility but to figure out which one it would be would be awful hard I would think ...

Al
03-22-13, 07:15 AM
No app file here either. I had the same problem a while back. I cleaned all the driver remnants with DriverSweeper and cleaned the registry with cc cleaner. Then reinstalled the drivers. Not sure if I did anything else, but I remember doing some research on the problem at the Collatz message boards. I now a lot of 7970 owners were having similar issues. At that time no one had a fix that worked for everyone.