Message boards :
Number crunching :
5970 Help.....
Message board moderation
Author | Message |
---|---|
Send message Joined: 6 Nov 09 Posts: 12 Credit: 348,876,876 RAC: 0 |
I was running 2 5970's on MW a while back i changed to other projects that needed newer opencl drivers. I now want to move back to MW but all i get are compute errors no mater what drivers i use, i have retried the cards on other projects no problems but on MW after 2 seconds error. Running windows 7x64 Stderr output <core_client_version>6.12.33</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 SSE3 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 11:36:18 (3240): called boinc_finish </stderr_txt> ]]> Think it must be somthing simple i missed but i have a 4770 and 4850X2 running without problems:( |
Send message Joined: 30 Dec 07 Posts: 311 Credit: 149,490,184 RAC: 0 |
Yes some older drivers are now not compatible with recent ATI MilkyWay applications and with newer drivers you may be getting "maximum time limit elapsed" errors. The Stderr output in your post indicates the driver isn't being recognised by the MilkyWay application. This was possibly an older driver you were using for that task. You could try Catalyst 11.5 or 11.3 and install the optimised application or alter the <duration_correction_factor> in client_state.xml file. Details in maximum time limit elapsed bug thread. This is just a suggestion, it has worked for some but others continue to get errors. |
Send message Joined: 6 Nov 09 Posts: 12 Credit: 348,876,876 RAC: 0 |
Thanks it looks like as you say the older drivers just dont work and with the later drivers i get 18/07/2011 13:55:19 | Milkyway@home | Aborting task ps_separation_13_3s_free_2_1135434_1: exceeded elapsed time limit 50.37 (2960409.21G/58777.69G) will give the thread a look thanks again. |
Send message Joined: 15 Jul 08 Posts: 383 Credit: 729,293,740 RAC: 0 |
Yes some older drivers are now not compatible with recent ATI MilkyWay applications and with newer drivers you may be getting "maximum time limit elapsed" errors. The Stderr output in your post indicates the driver isn't being recognised by the MilkyWay application. I'm still using 10.12 on all machines, works fine. Any compelling reasons to change? |
Send message Joined: 30 Dec 07 Posts: 311 Credit: 149,490,184 RAC: 0 |
I'm still using 10.12 on all machines, works fine. Any compelling reasons to change? No reason to change if it is working for you. I had seen posts from some using 10.12 and maybe 10.11 successfully with Windows 7 on MilkyWay ATI. Ba however was using 10.5 I believe at the time of that Stderr output task. I suggested Catalyst 11.5 or 11.3 as he said he was also contributing to other ATI projects and I had seen those versions recommended for PrimeGrid ATI. Catalyst 11.6 gives errors for some/many on PrimeGrid. |
Send message Joined: 4 Oct 08 Posts: 1734 Credit: 64,228,409 RAC: 0 |
I am currently using 11.3 with APP and these work OK on DNETC and PrimeGrid. I have yet to come over to Milkway, and hope I don't hit crunching snags. I should mention that one of my ATI cards is an HD5970. Go away, I was asleep |
Send message Joined: 15 Jul 08 Posts: 383 Credit: 729,293,740 RAC: 0 |
[quote]I suggested Catalyst 11.5 or 11.3 as he said he was also contributing to other ATI projects and I had seen those versions recommended for PrimeGrid ATI. Catalyst 11.6 gives errors for some/many on PrimeGrid. v10.12 also works fine here on Collatz, Moo! & PrimeGrid. I saw somewhere that 11.6 is using a new version of OpenCL that's faster on PrimeGrid, but as you indicate maybe has problems? I pretty much concentrate on NVidia over there though and use the ATIs where they run best. |
Send message Joined: 30 Dec 07 Posts: 311 Credit: 149,490,184 RAC: 0 |
I did not know that 11.6 was faster on PrimeGrid. I only knew 11.6 produced errors with HD 4xxx cards on PrimeGrid. As Ba also has HD 4xxx cards I thought I would suggest the most project compatible Catalyst versions that were suitable for all his computers, not just his computers with 5970 cards. Now that he knows however, he can use the most project compatible/efficient driver for each card if he wishes. He'll have swapped drivers so many times after all this that he'll probably have downloaded just about all the different versions by now. :) Working now, Ba? May need to reduce the core clock a bit from 900Mhz on a 5970 as the recent MilkyWay tasks may be more stressful than the ones in the past and certainly much more demanding on hardware than PrimeGrid. With the very fast purging of tasks here you need to watch very carefully to notice occasional tasks that appear to complete and report successfully but do not validate due to hardware errors. |
Send message Joined: 6 Nov 09 Posts: 12 Credit: 348,876,876 RAC: 0 |
Thanks guys ive gone to 11.3 but had to go for the app from arkayn to get it to work, without it all fail at around 60ish seconds. Also gone back to stock for now will up the speed later, as all my 5970's have accelero coolers this should not be a problem:) |
Send message Joined: 15 Jul 08 Posts: 383 Credit: 729,293,740 RAC: 0 |
May need to reduce the core clock a bit from 900Mhz on a 5970 as the recent MilkyWay tasks may be more stressful than the ones in the past and certainly much more demanding on hardware than PrimeGrid. With the very fast purging of tasks here you need to watch very carefully to notice occasional tasks that appear to complete and report successfully but do not validate due to hardware errors. I've noticed this too and have had to dial back the clocks on all my boxes. Still, there's more validation errors than there used to be. |
Send message Joined: 9 May 11 Posts: 15 Credit: 22,575,796 RAC: 0 |
May need to reduce the core clock a bit from 900Mhz on a 5970 as the recent MilkyWay tasks may be more stressful than the ones in the past and certainly much more demanding on hardware than PrimeGrid. With the very fast purging of tasks here you need to watch very carefully to notice occasional tasks that appear to complete and report successfully but do not validate due to hardware errors. Could that be why my Windows UI has started lagging when I run Milkyway? I used to be able to do anything unintensive, things like browsing the Internet or typing documents, with MW in the background, but now everything lags. To fix it I'll hit "Snooze GPU" on BOINC and I can watch all my lagging actions fast forward for about three seconds until I'm back to normal, crisp responsiveness. |
Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 |
Thanks guys ive gone to 11.3 but had to go for the app from arkayn to get it to work, without it all fail at around 60ish seconds. I am running twin 5970s with a 1090T, no issues with the normal GPU WUs so far. I am not regularly crunching here at the moment, tend to be a bit of a wondering beast :) I did some a couple of days ago just to check all was well, it was then. I'll switch as soon as I posted this for two batches of four, see what happens. I did have a couple of time elapsed errors some time ago, but its not a issue with me at present. Back in a few .. :) Regards Zy |
Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 |
Put around 25 through in the end. So far all went through, including some that were retreads from previous cruncher attempts. Not seen an error yet. About half have been through validation so far, both single and with wingment. Not sure what that contributes ... but for what its worth there it is. My settings for that run were: 765/300 using 11.6 Regards Zy |
Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 |
...... Also gone back to stock for now will up the speed later, as all my 5970's have accelero coolers this should not be a problem:) Ended up triple post ... sorry. Ba, What temperatures are you running at with the VRMs on the 5970(s) ? Regards Zy |
Send message Joined: 6 Nov 09 Posts: 12 Credit: 348,876,876 RAC: 0 |
Thats the only problem with the accelero's even with paste to replace the pads i still get over 100 on the vrms when overclocked running milkyway. |
Send message Joined: 6 Nov 09 Posts: 12 Credit: 348,876,876 RAC: 0 |
Just checked the vrm temps on one of my 5970's and down at 800(current setting) the highest is only 80:) I was getting over 100 at 900+ earlier in the year when the sun was out. |
Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 |
Thats the only problem with the accelero's even with paste Same effect on mine without the pads. The AMD formal limit for VRMs is 120 degrees C, anything below 100 not even worth thinking about. 100-105 is fine as a normal runner. 105-110 means life is getting a little warm, keep the eyes open, otherwise lose no sleep. 110-115 is time to do something get it down to below 110. 115-120 is sleepness nights time, stop GPU crunching until you sort out whats happening. 120+ the card is now melting, just a case of when not if - usually takes 1-3 months then it will go, but even if pulled back the degrading will have started reducing life. VRMs are a design fault on the 5970s - they were badly positioned causing them to heat up more than was designed for. However it only happens when overclocking, hence there is zero chance of RMA due to VRM burnout. In the last month alone I've seen at least five cruncher 5970s go South due to this. VRMs have to be watched very carefully, GPU temp is not the issue, VRM temperature is. Whilst the extreme end of this is happening, WUs can start to fall over, hence my question. So apologies if you were aware of the 5970 VRM saga already, but its an expensive *opps* to suffer :) EDIT: GPUs have a thermal cutout as a last ditch protection - VRMs do not, they will happily fry away without anyone knowing unless they are monitored. Its the lack of thermal cutout that is not realised by many - *poof* goes the card. Regards Zy |
Send message Joined: 1 Sep 08 Posts: 204 Credit: 219,354,537 RAC: 0 |
Lowering the memory clock is good to keep VRM temps in check while crunching MW. If I lower them from 1250 to 625 MHz on my OC'ed HD6950@6970 the VRM temp instantaneously drops by 10°C. I could go lower with the memory clock, but there'll be diminishing returns (and my backup projects suffer heavily, when kicked in..). MrS Scanning for our furry friends since Jan 2002 |
©2024 Astroinformatics Group