Message boards :
Number crunching :
ATI GPU app 0.19f fixes the ps_sgr_208_3s errors
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 . . . 7 · Next
Author | Message |
---|---|
Send message Joined: 26 Jul 08 Posts: 627 Credit: 94,940,203 RAC: 0 |
it's a great step forward in stability this new app! But I still have problems. I begin the Milkyway adventure with an ATI HD 4870, Catalyst 9.3 and 0.19d on Windows XP 32bit. Then I tried catalyst 9.4 and 9.5 and 0.19e and this new release. I must say that now I can crunch many WU...but it's not still stable 100% because I get VPU recovery when I do other things while 4870 is working. Even if I scroll a webpage I get VPU and all WU stop working. Previously I always had reboot and I could manage working only a few WU. I see today that server is feeding good WU..finally :). Investigate please on this problem. Maybe it's a Catalyst/Brook+ problem...that will be fixed in Cat 9.6 ? It's a driver problem under WinXP. The latest stable driver for XP (32 and 64Bit) was Cat9.2 or so (had it run for a week without problems). Later driver releases have serious stability issues with the GPGPU stuff. AMD/ATI admitted this problem with the Cat 9.3 and claimed they fixed it with a hotfix and in 9.4, but I would claim otherwise. I was not able to get it to run even only somewhat stable with Cat 9.3 (hotfix) or 9.4 under XP64 as well as XP32. I would recommend you to install the Cat 8.12. From the reports I got there are the least problems with that version. Btw., under Vista/Win7 all Catalyst releases run if the appropriate dlls are copied and renamed as described in the readme. |
Send message Joined: 26 Jan 09 Posts: 589 Credit: 497,834,261 RAC: 0 |
|
Send message Joined: 18 Nov 07 Posts: 280 Credit: 2,442,757 RAC: 0 |
By the way, shouldn't support for the previous version be dropped Server-side to ensure everyone upgrades? (though I think the GPU people are pretty avid, not everyone is guaranteed to notice) |
Send message Joined: 30 Aug 07 Posts: 2046 Credit: 26,480 RAC: 0 |
By the way, shouldn't support for the previous version be dropped Server-side to ensure everyone upgrades? (though I think the GPU people are pretty avid, not everyone is guaranteed to notice) There's no way for us to force users to upgrade an unofficial application. However, having it repeatedly crash and not grant credit is pretty good incentive (which is what's happening now :P ). |
Send message Joined: 16 Mar 08 Posts: 12 Credit: 732,092,849 RAC: 0 |
Travis, when did you stop granting credit? Just wondering how many days I was crunching for nothing. I must say very good job on this update. I've been running it since this morning and I'm already impressed with the improvement. My system no longer seem to be brought to a halt when crunching. Maybe the next big step is to see if it will work with the newer driver (9.5) or maybe I shouldn't fix what isn't broken :) |
Send message Joined: 18 Nov 07 Posts: 280 Credit: 2,442,757 RAC: 0 |
There's no way for us to force users to upgrade an unofficial application. However, having it repeatedly crash and not grant credit is pretty good incentive (which is what's happening now :P ). Really? I thought you were already checking the version string ... there's nothing saying users can't deliberately change their version string to match an official app, but why would they do that? Surely you could do a server-side check of the version string and only allow officially supported ones? (i.e. send a message 'this application version is unsupported, please upgrade' upon WU requests) Your dialogue with Cluster Physik seems pretty good nowadays. (I have no real knowledge of BOINC server code, so if it can't be done, I guess that's that) MoDifiy: I'm pretty sure Travis just meant that if the WU errors out as it would with a previous version app, you don't get credit. Not that he disabled credit system-wide. |
Send message Joined: 26 Jul 08 Posts: 627 Credit: 94,940,203 RAC: 0 |
I must say very good job on this update. I've been running it since this morning and I'm already impressed with the improvement. My system no longer seem to be brought to a halt when crunching. Maybe the next big step is to see if it will work with the newer driver (9.5) or maybe I shouldn't fix what isn't broken :) Thanks! By the way, the improvement of the system responsiveness was already done with the version released sometimes in march (0.19e) ;) As you have a Vista system, you can use also newer Catalyst drivers. But be sure to have a look to the readme.txt (within the zip file) as you have to copy and rename some dll files. |
Send message Joined: 30 Nov 07 Posts: 9 Credit: 165,873,750 RAC: 0 |
Hello i have some little problems with the 3s sérial on the 19f version somes wu running for ever arount 10 % from all only with s3-5 the s3-6 are working goud |
Send message Joined: 16 Mar 08 Posts: 12 Credit: 732,092,849 RAC: 0 |
ahh yes that's right. I also remember why I wasn't using it. I ended up speaking too soon on this update or jinxed it. It blue screened on me 3 times tonight and I remember having that same problem with 19e. I'm not sure why it ran so nicely all day with no problems only to then get home and have problems. I've updated to Cat9.5 and I will see if that helps. I saw in the release notes that it does state fixed some stability with stream. I'll post how it runs. Thanks. |
Send message Joined: 14 Jul 08 Posts: 5 Credit: 44,130,280 RAC: 0 |
Help! How do I change from 3 WU(default) to 8? I was running 8 on a 4850 with a quad 6600. I have read the “read me” file, have tried editing <cmdline>n8</cmdline> + many variations but no change still only 3. I’m obviously doing something wrong, can any one help PS Many thanks CP, the improvements are fantastic! |
Send message Joined: 2 Jan 08 Posts: 123 Credit: 69,756,860 RAC: 1,237 |
Help! Do you really need 8 ?? I have left mine at the default 3 and my output (thanks to Travis and more work) has gone from an average of 900 to over 8,800 in one day on just the one card. I also remember Cluster Physik saying that running a heap of threads does not mean you will do more work as it slows all the threads down, and doing 1 to 4 threads will get you the same amount of output as doing many more. Cluster will correct me if I am wrong I am sure. |
Send message Joined: 14 Jul 08 Posts: 5 Credit: 44,130,280 RAC: 0 |
I would like to see for myself. ATM I’m running 3 x MW & 4 x ABC, would like to even it out a little. So if anyone can help, I'll be grateful. |
Send message Joined: 21 Feb 09 Posts: 180 Credit: 27,806,824 RAC: 0 |
Help! Did you shut down BOINC, make the change, save, then reload boinc? Personally I use <cmdline>n1 w1.1</cmdline> |
Send message Joined: 14 Jul 08 Posts: 5 Credit: 44,130,280 RAC: 0 |
Yes shut down & saved. reloaded. tried your <cmdline>n1 w1.1</cmdline> only 3 WU? I thought that n= number of WU crunching, obviously its not. So how would I get it to crunch 4, 6 or 8 WU simultaneously? |
Send message Joined: 25 Mar 09 Posts: 11 Credit: 10,178,231 RAC: 0 |
_3s WUs still get compute error. When BOINC client start to crunch a _3s WU, the GPU app stop crunching any other WU indefinitely (GPU time move but progress remain at 0%). I'll have to suspend all _3s in the list, close BOINC and restart to get GPU app to work on _2s WUs. Also, after installation of 019f, my quad system will run only 2 other CPU WUs plus 2 GPU WUs or 4 CPU WUs and no GPU task. I'm using BOINC 6.6.36 on W7 and catalyst v9.5. This catalyst package install atical...64.dll files on my system and I make a copy and rename them to amdcal...64.dll and place them back on system32 folder, will this be sufficient? |
Send message Joined: 26 Jul 08 Posts: 627 Credit: 94,940,203 RAC: 0 |
_3s WUs still get compute error. When BOINC client start to crunch a _3s WU, the GPU app stop crunching any other WU indefinitely (GPU time move but progress remain at 0%). I'll have to suspend all _3s in the list, close BOINC and restart to get GPU app to work on _2s WUs. After having a look to your host list (the Win7 hosts), the only compute errors showing up on your computers are caused by using the old 0.19e version. The Core2Quad still runs 0.19e. After you updated the Core2Duo to 0.19f, all appears to be okay. And if you specify the options "n2 w1.1 f20" on that C2D, you don't have to wonder why 2 WUs run. But that is competely okay as running more won't help the performance. And why the BOINC-Client 6.6.36 is not starting the GPU tasks in parallel to CPU tasks, I simply don't now. You can play around with the avg_ncpu value in the app_info.xml or raise the number of processed WUs in the c_config.xml. I'm using the client version 5.5.45, 6.2.19 and 6.6.21 and it runs quite okay there. |
Send message Joined: 19 Feb 08 Posts: 350 Credit: 141,284,369 RAC: 0 |
Hi, today i have updated to 0.19f and boinc to 6.6.36. It works fine now, only a few wu's produce compute errors (~2/hr). But i see the same thing, my quad core cpu works only on two standard wu's and ~20 milkyway wu's are marked as active and three of them show progress. This didn't happen prior to the update. |
Send message Joined: 21 Feb 09 Posts: 180 Credit: 27,806,824 RAC: 0 |
Yes shut down & saved. reloaded. n is the number of WUs to be crunched at once - for some reason your app_info.xml isn't being read properly. There may be an error in it. Here's mine for the Win32_SSE2 : <app_info> <app> <name>milkyway</name> </app> <file_info> <name>astronomy_0.19_ATI_SSE2f.exe</name> <executable/> </file_info> <file_info> <name>brook.dll</name> <executable/> </file_info> <app_version> <app_name>milkyway</app_name> <version_num>19</version_num> <flops>1.0e11</flops> <avg_ncpus>0.1</avg_ncpus> <max_ncpus>1</max_ncpus> <cmdline>n1 w1.1</cmdline> <file_ref> <file_name>astronomy_0.19_ATI_SSE2f.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>brook.dll</file_name> </file_ref> </app_version> </app_info> Just change n1 to n8 if you want 8. I choose 1, then my cache of WUs is at its peak. 31s for the 2s WUs, 47s for the 3s WUs. Just under 1 credit per second. |
Send message Joined: 21 Feb 09 Posts: 180 Credit: 27,806,824 RAC: 0 |
Hi, If you did it for the older versions, you will have to re-edit your app_info.xml file. All the WUs will be marked as active, and if each reserve 0.1 cpu, and you have 20, then it'll take away two CPU cores from computing. Reduce 0.1 to 0.01 if you want to free up your cores, though it may make WUs run a little (2%?) slower when they do need the CPU. The three WUs at a time is an automatic feature - again edit your app_info.xml to <cmdline>n1</cmdline> if you want 1 at a time (which, if you read a few posts before hand, I wrote not too long ago in this very thread). |
Send message Joined: 14 Jul 08 Posts: 5 Credit: 44,130,280 RAC: 0 |
Thanks BORANDI. Mine is exactly the same with the exception that its running x64 so it reads <name>astronomy_0.19_ATI_x64f.exe</name>. Might be an error in the x64 version? But knowing my lack of programing skills it must be me! |
©2024 Astroinformatics Group