Printable View
Just bought a used 4870 for my 2nd rig for the express purpose of crunching MW. It works fine on Collatz, but fails all MW wu immediately. With updated drivers it would make it to 80% complete, then get a computational error. With the original disk drivers it fails at 1 second. I'm a Nvidia guy, so any help from the ati folks would be helpful. BTW, it is not OCed. Ask if you need more info, I'm sure I didn't give you everything you need for a diagnosis.
Al
xp 32 - Intel Quad q6600
Are you running or have you run any of the optimized aps? If there is an old ap_info.xml file floating around, that will do it. I am running 11.3 driver version for my 4870. Not sure if that has anything to do with it. I remember having that same problem, but don"t remember the cause, because I was dealing with a few issues at that time. I do run the same Quad though. I'll keep it in mind and if I think of anything else, I'll let you know.
no app_info file. I did move it to another rig, with a better power supply thinking that might be the issue. No luck. All fail at between 50 & 75%.
What kind of error message does it show in the messages tab?
7/15/2011 2:27:10 PM | Milkyway@home | Aborting task ps_separation_13_3s_free_2_1051303_0: exceeded elapsed time limit 98.80 (2960409.21G/29964.59G)
It aborts, finishes and then says "Computational Error."
That's interesting. Never seen that one before. I'm sure that the project doesn't care how long it takes you to crunch, so that can't be it. Must be something where it finishes one part of the WU and then needs to switch to something else, but is hanging up. Where's Maxwell or Crunch3r when you need them.;)
Well since I'm not helping, my advice would be to detach. Make sure that you delete the MW directory and then reattach. Maybe that will excorsize the ghosts out of the machine.
IIRC there was a thread on MW discussing that several months ago. I'm thinking that the time limit was something of a protection thing because some wu were hanging up GPUs. I don't remember if there was a work around or whether they thought they had fixed it on the wu side.
I was just about to PM you on this one. Figured you were about the most knowledgable about it. If I get time I'll search on the MW site.Quote:
Originally Posted by Bryan
This thread suggests an app_info.xml file might fix the problem. It might be worth a try.
http://milkyway.cs.rpi.edu/milkyway/...?id=2468#49488
Have you tried looking at the stderr and stdout files in the 'slot' folder while it's crunching? I've found some useful info doing that before.Quote:
Originally Posted by Al McAdams
I think that is the answer CB! I was crunching MW back then and I always run an app_info on that project. I never did see the problem myself but now remember that a lot of others were having trouble.
Thanks for the suggestions. I read the MW thread, I tried the app_info file with and without the optimized app. No luck. Still aborts, finishes and gives a computational error. Works file on Collatz, PG Fails immediately too.
Al
Well Al of course the smart a$$ answer is to run Collatz. :))
What CAT driver are you running? I know that 10.10, 10.12, 11.3, and 11.5 work on my 5870s. Also OS and BOINC version?
Well, I don't have a 4870 but I do have three 4850s.
For the two systems with the 4850s installed I use Cat 10.10. One system is XP Pro x64 box and the other is running Win7 Ultimate x64. I had to many issues with the later versions of Catalyst and could never figure out why.
If you want to give Cat. 10.10 a shot, try the procedure below.
This procedure will require the following programs...
Driver Sweeper: Version 2.1 can be DL'd here. Version 3.1 can be DL'd here.
Note: It is reported that the versions in-between have spyware. Also I have read one report on download.com that version 3.1 also has spyware but Softpedia.com says it is spyware free.
Catalyst 10.10: For Windows XP 32 bit --- For Windows XP 64 bit
CCleaner: The latest version can be DL'd here.
Procedure:
1) While BOINC Manager (BM) is running, set MW to No New Work.
2) Abort any MW wu's you may have and then Reset MW.
3) Next, from the BM Main Menu click Advanced > Options and in the popup window click the General Tab and uncheck Run Manager at Login?, then click the OK button.
4) Shutdown BM.
5) Uninstall your current ATI/AMD Catalyst Suite. When the uninstaller is finished it will say a reboot is required, click yes/OK.
*For XP: Click the Windows Start Button > Settings > Control Panel > Add or Remove Programs and double-click the ATI Catalyst Install Manger to begin the uninstall process.
6) After reboot install Driver Sweeper and CCleaner.
7) Start Driver Sweeper. In the Main Window select AMD display which should be the first item on the list and click the Clean button in the bottom right hand corner of the app window. If when finished the program requests a reboot, click yes.
8) After reboot right-click your desktop Recycle Bin icon and select Open CCleaner. In the left most column of the main window click the Registry icon and select all the check-boxes in the adjacent column to the right. Next click the Scan for Issues button (Bottom Left). After scan is complete click the Fix Selected Issues button (Bottom Right). In the pop-up window that appears click yes to backup registry changes and save the backup (Do not forget where you save it). On the next pop-up window that appears click the Fix All Selected Issues button.
9) Next, back in the left column of the main window, click the Cleaner icon and in the adjacent column to the right put a check in the boxes for the items you wish to be cleaned. (If you are unsure what effect cleaning any particular item in the list will have on your system, please leave it un-checked). Now click the Analyze button (Lower left) and preview the list that is displayed in the main window. Make sure there is nothing in the list that you need and then click the Clean button (Lower Right). When complete close CCleaner.
10) Now install the Catalyst 10.10 driver (Or whichever version you prefer). The installer should request that your system be rebooted when it is finished. Click yes.
11) After reboot start BOINC manager and test the project you had trouble with. After testing is complete and all is well, undo the changes made in step #3.
I've tried the card in both a win7 64 and an xp pro 32 with the same results. I'm using the newest drivers. I'm working today but I'll give the older drivers a test this evening and let you know what happens.
Thanks for the suggestions.
Al
Here's one of the messages:
<core_client_version>6.10.60</core_client_version>
<![CDATA[
<message>
Incorrect function. (0x1) - exit code 1 (0x1)
</message>
<stderr_txt>
Error loading Lua script 'astronomy_parameters.txt': [string "number_parameters: 4..."]:1: '<name>' expected near '4'
Error reading astronomy parameters from file 'astronomy_parameters.txt'
Trying old parameters file
Using SSE2 path
Failed to get CAL device attributes: Parameter passed in is invalid (CAL_RESULT_INVALID_PARAMETER)
Error getting device information: Parameter passed in is invalid (CAL_RESULT_INVALID_PARAMETER)
Failed to get CAL info: Parameter passed in is invalid (CAL_RESULT_INVALID_PARAMETER)
Failed to setup CAL
10:10:53 (3296): called boinc_finish
</stderr_txt>
]]>
After a brief Google search and a little Google Translate, I found the following possible issues.Quote:
Originally Posted by Al McAdams
Driver version. Might be too old or too new. :-?? That's probably not it.
Did you reboot after installing the driver?
Antivirus?
That just may be the answer. The driver is too new for the 4xxx series. That is very similar to the problem I had which is why I reverted back to Catalyst 10.10. The 4xxx series and lower, have only a partial (As in Beta) implementation of OpenCL. Somewhere around Cat. 10.12 AMD started a full implementation of OpenCL which started playing havoc with a lot of 4xxx series owners.Quote:
Originally Posted by trigggl
No antivirus.. I did reboot. I'm about to head downstairs to install the older drivers. I'll let you know.
Al
I just noticed something.
http://milkyway.cs.rpi.edu/milkyway/...?id=2489#49692Quote:
Error loading Lua script 'astronomy_parameters.txt': [string "number_parameters: 4..."]:1: '<name>' expected near '4'
Try a CPU only WU if you haven't already. It may have nothing to do with the card.
If that is the problem then all that is required is to reset the project. That will force BOINC to DL the new file.Quote:
Originally Posted by trigggl
When Kevint and I were doing this in the beginning I think it was a .NET issue. I know your issue is different, just throwing that out there. In the end it was 2.0 matched to the right driver and the right ccc and Boinc version. I'm currently having the exact issue with 5830's. Going to start rolling back tomorrow. Let you know what I find. I did get success with the latest Boinc, 11.3 CCC for Primegrid.
CPU only seem to run fine, though I didn't run all variations yet. I'll read through the complete post at MW forum and see if anything there helps. I followed Fire$torm's directions to the letter and all tasks continue to fail at about 127 seconds at 80+% done. Here's the error message:
Stderr output
<core_client_version>6.10.60</core_client_version>
<![CDATA[
<message>
Maximum elapsed time exceeded
</message>
<stderr_txt>
Error loading Lua script 'astronomy_parameters.txt': [string "number_parameters: 4..."]:1: '<name>' expected near '4'
Error reading astronomy parameters from file 'astronomy_parameters.txt'
Trying old parameters file
Using SSE2 path
Found 1 CAL devices
Chose device 0
Device target: CAL_TARGET_770
Revision: 2
CAL Version: 1.4.838
Engine clock: 750 Mhz
Memory clock: 900 Mhz
GPU RAM: 1024
Wavefront size: 64
Double precision: CAL_TRUE
Compute shader: CAL_TRUE
Number SIMD: 10
Number shader engines: 1
Pitch alignment: 256
Surface alignment: 256
Max size 2D: { 8192, 8192 }
Estimated iteration time 330.481667 ms
Target frequency 30.000000 Hz, polling mode 1
Dividing into 9 chunks, initially sleeping for 0 ms
Integration range: { nu_steps = 640, mu_steps = 1600, r_steps = 1400 }
Using 9 chunk(s) with sizes: 176 176 176 176 176 176 176 176 192
Integration time = 200.021940 s, average per iteration = 312.534281 ms
Integral 0 time = 202.951483 s
Likelihood time = 3.230086 s
<background_integral> 0.000928726019059 </background_integral>
<stream_integral> 353.723081457166980 130.799939675965280 1464.654321262977600 </stream_integral>
<background_likelihood> -3.602249639427543 </background_likelihood>
<stream_only_likelihood> -17.295920317294978 -4.388825045143967 -4.522211712778050 </stream_only_likelihood>
<search_likelihood> -3.096088347341647 </search_likelihood>
<search_application> milkywayathome_client separation 0.82 Windows x86 double CAL++ </search_application>
00:20:30 (3416): called boinc_finish
</stderr_txt>
]]>
On the upside, PrimeGrid now seems to work without errors...Al Oh, and I have reset the project.
Sorry it didn't work for you. For the moment I do not have anything else I can think of. Maybe take a break from it for a day or so, just to help dissipate some of the frustration.
It would help if we knew someone with a 4870 that is currently running MW.
Oh, a thought..... OK, you said Collatz crunched fine. Have you tried a couple of Moo wu's? Again, just to see.
Edit: Not that you need to do this now but when you feel like it run GPU-Z. This utility just gives detailed info on GPUs.
Mine was working fine last month, driver ver 6.14.10.7131 (cat 10.10). Just tried some new work, all error out with the majority of them being a "exceeded elapsed time limit" error after a couple seconds and a couple with the "Error loading Lua script 'astronomy_parameters.txt': [string "number_parameters: 4..."]:1: '<name>' expected near '4' "... bla bla bla output.Quote:
Originally Posted by Fire$torm
Maybe it's not me? PG fine, Collatz good to go, Moo no problem...MW - can't run anything but cpu, all gpu wu continue to fail. Guess if I want to crunch MW I'll have to do it on my Nvidia cards.
Al
Thanks for the input. It is what I was hoping to hear.Quote:
Originally Posted by DrBob
That was my line of thinking. Which is why I suggested Moo. It has to be a batch of bad wu's. So good news for you. Bad news for MW.Quote:
Originally Posted by Al McAdams
Probably too late to get in on the double credit thing, but I had checked my machines out that used to run MW. All errored out with same message. Looks like there is a fix. I was able to get mt 2 XP64 machines crunching using the ap and ap_info file from Arkayn. Wasn't able to get my dual core XP32 running though. If you are running XP32, I think this would be the DL for you.
http://www.arkayn.us/forum/index.php...downfile;id=75
Hopefully that will work for you. There is a directory in there called files to copy. If you know where your MW folder is, then shut down BOINC. Run the task manager to make sure that boinc.exe is not running. Copy the files into your folder and restart BOINC. You might get a couple of errors at first, but it should straighten out. Let us know if this works.
I just re-attached to MW yesterday and am having the same problem. I'm 99% certain it's a server side issue. I was getting errors and BSOD's.
All four of my ATI machines produce nothing but errors for MW. They used to work just fine. No app_info or anything. And they continue to work fine on dnetc/moo/etc.
This is weird. I'm am not having any issues at all. Not with my 5830, nor the three 4850. That's four cards across three systems, Win7 Home premium, WinXP Pro and Win7 Ultimate (all are x64).
BOINC Manager v6.10.58 (All 3 systems)
Milkyway App app 0.82
HD 4850s - ATI Catalyst 10.10 (w/o OpenCL)
HD 5830 - AMD Catalyst 11.5 (w/ OpenCL)
Edit: Just to be clear, the ATI CAtalyst 10.10 driver comes in two versions for each OS platform.
Example:
Without OpenCL: 10-10_vista64_win7_64_dd_ccc_enu.exe
With OpenCL: 10-10_vista64_win7_64_dd_ccc_ocl.exe
The recent aps posted at Arkayn seem to address this issue. I couldn't get my XP32 going because I didn't have the time to screw with it. Check out their aps for your OS and DL the one that fits your chip instructions (SSE1, SSE2, SSE3...etc........) The WU's I have been sent, work for the 0.82 files.
http://www.arkayn.us/forum/index.php...wnloads;cat=11
Doesn't it figure that when a possible solution is in reach, mw goes down. I've downloaded the files and copied them to the mw directory. When it comes up again we'll find out if it worked on the Win32. Just glad I had the Nvidia cards to at least get some of the double credits. Thanks for all the suggestions and I'll let you know what happens.
Al
app_info worked. the 4870 is now crunching without errors. Better late than never. Thanks.
Maybe just to help the admin(s) at MW you could post your original problem on their forum. See this ---> http://www.setiusa.us/showthread.php...ll=1#post20497Quote:
Originally Posted by Al McAdams
Done!
Glad to hear you got it running. These stupid problems really make running their project frustrating!
+1 to that...Thatīs the main reason I still havenīt got to the 1 MILL mark on this project....And now I canīt since I have no DP GPU....:|Quote:
Originally Posted by Crazybob
Hey Beer, Who told you that you didn't have any Double Precision ATI's? From where I'm sitting YOU have 3 HD 4850s and 1 HD 5830 just waiting to crunch some Milkyway!!!! :DQuote:
Originally Posted by Beerdrinker
Man, thats very tempting!! That is a really generous offer!...I will however feel better if you just crunch on our own. I donīt wanna take anyones crunchers. Also - since the reducing credit thingy at PG, I am looking for DP ATI next time..:cool:Quote:
Originally Posted by Fire$torm