Message boards :
Number crunching :
astronomy_020.ATI_x64 problems
Message board moderation
Author | Message |
---|---|
Send message Joined: 23 Feb 09 Posts: 28 Credit: 10,775,220 RAC: 0 |
Hi, I installed astronomy_020.ATI_x64 on my machine but got error: Protokollname: Application Quelle: Application Error Datum: 22.09.2009 17:21:44 Ereignis-ID: 1000 Aufgabenkategorie:(100) Ebene: Fehler Schlüsselwörter:Klassisch Benutzer: Nicht zutreffend Computer: V8-SK01 Beschreibung: Fehlerhafte Anwendung astronomy_0.20_ATI_x64.exe, Version 0.0.0.0, Zeitstempel 0x4aac1a09, fehlerhaftes Modul amdcalrt64.dll, Version 6.0.6002.18005, Zeitstempel 0x49e0421d, Ausnahmecode 0xc0000135, Fehleroffset 0x00000000000b8fb8, Prozess-ID 0x1e64, Anwendungsstartzeit 01ca3b9865000a55. Ereignis-XML: <Event xmlns="http://schemas.microsoft.com/win/2004/08/events/event"> <System> <Provider Name="Application Error" /> <EventID Qualifiers="0">1000</EventID> <Level>2</Level> <Task>100</Task> <Keywords>0x80000000000000</Keywords> <TimeCreated SystemTime="2009-09-22T15:21:44.000Z" /> <EventRecordID>35786</EventRecordID> <Channel>Application</Channel> <Computer>V8-SK01</Computer> <Security /> </System> <EventData> <Data>astronomy_0.20_ATI_x64.exe</Data> <Data>0.0.0.0</Data> <Data>4aac1a09</Data> <Data>amdcalrt64.dll</Data> <Data>6.0.6002.18005</Data> <Data>49e0421d</Data> <Data>c0000135</Data> <Data>00000000000b8fb8</Data> <Data>1e64</Data> <Data>01ca3b9865000a55</Data> </EventData> </Event> --------------------------- Hardware: Sapphire 3870 Ultimate, driver 9.2 ATI GPU 0: RV670 (CAL version 1.3.186, 512MB, 497GFLOPS) 8core Xeon, 16GB RAM OS:Vista64 ------------------------ running BOINC 6.10.6 any ideas what todo ? |
Send message Joined: 19 Feb 09 Posts: 32 Credit: 32,843,308 RAC: 0 |
Ausnahmecode 0xc0000135 Hi, this indicates that the application could not be initialized. Did you copy and(!) rename the CAL-DLLs (this is necessary for Catalyst 9.2 and newer) in the Windows-System directories - simply renaming them is not enough. If not do the following: 1. Open command prompt as Administrator 2. execute "cd %systemroot%\system32" 3. execute "copy atical*.dll amdcal*.dll" 4. execute "cd %systemroot%\SysWOW64" 5. execute "copy atical*.dll amdcal*.dll" Now you should have the following files: System32: amdcalrt64.dll,amdcalcl64.dll,amdcaldd64.dll,aticalrt64.dll, aticalcl64.dll,aticaldd64.dll SysWOW64: amdcalrt.dll,amdcalcl.dll,amdcaldd.dll,aticalrt.dll, aticalcl.dll,aticaldd.dll You've to repeat this procedure anytime you update the Catalyst drivers. If you already did this, please check if you have the Visual Studio C++ runtimes (2005 and maybe 2008) installed. Starfire |
Send message Joined: 23 Feb 09 Posts: 28 Credit: 10,775,220 RAC: 0 |
Hi Starfire, I copied all as you said, now it works. Thank you :-) but have next question: 24.09.2009 20:02:43 Milkyway@home Message from server: No work sent 24.09.2009 20:02:43 Milkyway@home Message from server: (reached limit of 48 tasks) how can avoid this limit of 48 tasks ? some later edit: now I see the next wu's are downloading and working. looks like it's OK |
Send message Joined: 19 Feb 09 Posts: 32 Credit: 32,843,308 RAC: 0 |
Hi, glad I could help. This limit is set by the project - you can only have 6 WUs per CPU core downloaded at any time. New tasks will be downloaded when you complete and upload some of them. |
Send message Joined: 23 Feb 09 Posts: 28 Credit: 10,775,220 RAC: 0 |
Hi, some of my results have errors like this: Couldn't find input file have I a uncorrect app_info.xml ? <app_info> <app> <name>milkyway</name> </app> <file_info> <name>astronomy_0.20_ATI_x64.exe</name> <executable/> </file_info> <file_info> <name>brook64.dll</name> <executable/> </file_info> <app_version> <app_name>milkyway</app_name> <version_num>19</version_num> <flops>1.0e11</flops> <avg_ncpus>0.05</avg_ncpus> <max_ncpus>1</max_ncpus> <coproc> <type>ATI</type> <count>1</count> </coproc> <cmdline></cmdline> <file_ref> <file_name>astronomy_0.20_ATI_x64.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>brook64.dll</file_name> </file_ref> </app_version> <app_version> <app_name>milkyway</app_name> <version_num>20</version_num> <flops>1.0e11</flops> <avg_ncpus>0.05</avg_ncpus> <max_ncpus>1</max_ncpus> <coproc> <type>ATI</type> <count>1</count> </coproc> <cmdline></cmdline> <file_ref> <file_name>astronomy_0.20_ATI_x64.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>brook64.dll</file_name> </file_ref> </app_version> </app_info> -------------------------------- can I delete the 0.19 part ? |
Send message Joined: 19 Feb 09 Posts: 32 Credit: 32,843,308 RAC: 0 |
Hi, Your app_info.xml is OK - these errors sometimes occur because of a bug in the 6.10.* Boinc clients. The new version 6.10.7 - available here - should fix this. |
Send message Joined: 23 Feb 09 Posts: 28 Credit: 10,775,220 RAC: 0 |
thanks, I installed BOINC 6.10.7 no error occured since then. I have a 8-core Xeon, what change is recommended for optimal use in the app_info.xml ? have now: <avg_ncpus>0.05</avg_ncpus> <max_ncpus>1</max_ncpus> <core_client_version>6.10.7</core_client_version> <![CDATA[ <stderr_txt> Running Milkyway@home ATI GPU application version 0.20 (Win64) by Gipsel ignoring unknown input argument in app_info.xml: --device ignoring unknown input argument in app_info.xml: 0 CPU: Intel(R) Xeon(R) CPU E5405 @ 2.00GHz (8 cores/threads) 2.39827 GHz (273ms) CAL Runtime: 1.3.186 Found 1 CAL device Device 0: ATI Radeon HD 3800 (RV670) 512 MB local RAM (remote 2047 MB cached + 2047 MB uncached) GPU core clock: 776 MHz, memory clock: 1126 MHz 320 shader units organized in 4 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads supporting double precision 0 WUs already running on GPU 0 Starting WU on GPU 0 main integral, 320 iterations predicted runtime per iteration is 341 ms (33.3333 ms are allowed), dividing each iteration in 11 parts borders of the domains at 0 152 296 440 584 728 872 1024 1168 1312 1456 1600 Calculated about 8.22242e+012 floatingpoint ops on GPU, 1.23583e+008 on FPU. Approximate GPU time 130.876 seconds. probability calculation (stars) Calculated about 3.34818e+009 floatingpoint ops on FPU. WU completed. CPU time: 22.2813 seconds, GPU time: 130.876 seconds, wall clock time: 132.497 seconds, CPU frequency: 2.3983 GHz </stderr_txt> ]]> the machine need 130 sec for each wu. Is this a good value ? it uses the GPU and CPU 7 Node 0 with 2,14% temps are moderate: GPU Diode 61 °C (142 °F) Temperatur 1 43 °C (109 °F) Temperatur 2 51 °C (124 °F) Temperatur 3 50 °C (122 °F) The HD3870 Ultimate has passive cooling unit. I built in a additional 10cm fan which blows on the passive cooling unit. The GPU is not overclocked. If I will lokup with GPU-Z.0.3.3 to see the load, the driver crashs. your hints are welcome |
Send message Joined: 26 Jul 08 Posts: 627 Credit: 94,940,203 RAC: 0 |
thanks, I installed BOINC 6.10.7 130 seconds for a HD3870 is the expected value. The temperatures look great (other are running in the 80-90°C range). Guess the cooler is quite large (designed for passive cooling after all) and the additional fan helps a lot. You can try to optimize the settings to get an even lower CPU load, shave a single second off the calculation time, and save a few Watts for the GPU. One after the other. HD3800 series cards can use a slightly higher wait factor than HD4700/4800 cards without loosing performance. This lowers the CPU load and can be achieved with setting a wait factor slightly higher than one (1.15 to 1.2 works good without loosing performance, maybe even up to 1.25 if multiple WUs are running). Running multiple WUs concurrently on one GPU can hide the time the CPU does something (depending on the CPU about one or two seconds per WU). This doesn't change the GPU time you see in the task list, but effectively raises the throughput slightly. The time the manager shows will roughly double (the two WUs have to share the GPU). But running two WUs at once should take slightly less than double the time of a single WU. This works best if you let it run a bit so the two concurrent WUs don't need the CPU at the exact same time ;). Using more than 2 WUs won't bring any significant gain. Be warned that some report that is has a negative impact on stability, but generally it makes no difference if the system (with the driver version one uses) runs the app stable. To get the client to start more than one instance of the app per GPU, just reduce the <count> value in the <coproc> section of the app_info.xml (0.5 means it uses a half of the GPU, so the client starts 2 WUs). Summed up, I would suggest that you try these settings: <coproc> <type>ATI</type> <count>0.5</count> </coproc> <cmdline>w1.15</cmdline> To save a a bit on the power consumption, you can reduce the memory frequency, as MW doesn't use a lot. You have to see how important this aspect is for you, but going down to 600MHz or so shouldn't show much of a difference to the runtimes. When you go further to let's say 300MHz (if your card can run that stable), you will start to loose a few seconds calculation time. But you could compensate that with a raised GPU core clock of course ;) And that GPU-Z crashes the driver when MW runs (also the diagnostic tab of the CCC) is a known problem. I have no idea what these diagnostic tools does different than the ATI Tray tools for instance which doesn't show this behaviour. But only some OS driver combinations are affected. |
Send message Joined: 23 Feb 09 Posts: 28 Credit: 10,775,220 RAC: 0 |
[quote]
Thank you Custer Physik tried above settings as you recommended. 600MHZ elapsed 5:26 776MHz elapsed 4:23 both when 2 wu's are running on GPU. GPU temp shows now 75 grd celsius. running now <cmdline>w1.2</cmdline> cpu usage is now 0,19% till 0,39% very low, mostly on CPU7 Node 0 with w1.2 the cpu usage is going against zero, really great sucess the machine shows 0 or 1% cpu usage !! very impressive :-) |
Send message Joined: 26 Jul 08 Posts: 627 Credit: 94,940,203 RAC: 0 |
tried above settings as you recommended. Uhmm, but you changed the core drequency, not the memory frequency. Just look at your WUs: Device 0: ATI Radeon HD 3800 (RV670) 512 MB local RAM (remote 2047 MB cached + 2047 MB uncached) It's no wonder you got shorter times with higher core clocks. I spoke of reducing the memory frequency (now on 1126MHz) to something like 600MHz. You will see much less of an impact (if you see any change at all) than changing the core frequency. And it saves some power too. |
Send message Joined: 23 Feb 09 Posts: 28 Credit: 10,775,220 RAC: 0 |
if I look at Catalyst Control Center lowest possible memory frequency is 1126 highest 1387 I can not reduce it lower than 1387, thats factory standard or do I missing something ? |
Send message Joined: 26 Jul 08 Posts: 627 Credit: 94,940,203 RAC: 0 |
if I look at Catalyst Control Center lowest possible memory frequency is 1126 highest 1387 You can't lower it? That's strange. The CCC limits the possible range, okay. But I've never seen that one can't lower it at all. But the range also depends on the manufacturer and the graphics card BIOS (the range is deteremined by the BIOS), so it is entirely possible. If you still want to try it, 3rd party tools (like ATI Tray Tools) don't have this restriction. |
Send message Joined: 9 Feb 09 Posts: 166 Credit: 27,520,813 RAC: 0 |
Isn't it ddr2 memory on the these cards If so its 1126 meaning 563 Mhz for real and is damn low :D Normal should be around 600 or am i mistaken Its new, its relative fast... my new bicycle |
Send message Joined: 26 Jul 08 Posts: 627 Credit: 94,940,203 RAC: 0 |
Isn't it ddr2 memory on the these cards HD3870 is more or less the only card using GDDR4 ;) |
Send message Joined: 23 Feb 09 Posts: 28 Credit: 10,775,220 RAC: 0 |
|
Send message Joined: 5 Sep 09 Posts: 2 Credit: 1,866,545,469 RAC: 0 |
Hello, I installed astronomy_020.ATI_x64 and now I can only crunch 1 unit at a time. When I was running .19f it would work on 3 but I set it to "n2" for stability. I do have 18 units waiting. Any suggestions. |
Send message Joined: 14 Feb 09 Posts: 999 Credit: 74,932,619 RAC: 0 |
Hello, Yep, look for this area in the app_info <coproc> <type>ATI</type> <count>1</count> </coproc> Change the count to either .50 for 2 at a time or .33 for 3 at a time. I personally use the .50 setting. |
Send message Joined: 5 Sep 09 Posts: 2 Credit: 1,866,545,469 RAC: 0 |
Thanks works great now. |
Send message Joined: 30 Nov 07 Posts: 9 Credit: 165,873,750 RAC: 0 |
yes but crunching one unit by one its 25 % more fasters |
Send message Joined: 1 Sep 08 Posts: 204 Credit: 219,354,537 RAC: 0 |
yes but crunching one unit by one its 25 % more fasters No, it's not. Your 4850 mostly need 60s for a "long" WU, one needs 55s and at least 2 need much longer (are you running cpu projects on all cores of these?). If I take the 55s as a baseline and calculate the speed for 800 MHz instead of 625 MHz I get 55 * 625/800 = 43s. That's just what I get with my 4870 at 800 MHz core, running 2 WUs in parallel to keep the GPU busy and avoid some idle seconds (see app readme). Of course this assumes you're only running 1 WU at a time. MrS Scanning for our furry friends since Jan 2002 |
©2024 Astroinformatics Group