Message boards :
Number crunching :
New ATI OpenCL-based AstroPulse (rev516) released
Message board moderation
Previous · 1 · 2 · 3 · 4 · Next
Author | Message |
---|---|
Gundolf Jahn Send message Joined: 19 Sep 00 Posts: 3184 Credit: 446,358 RAC: 0 |
Any reason why the Pendings from the MB WU's on my rig with this app don't appear in Pendings but the ones from my Dual-Core do? Normal behaviour here with 5.10.45 and 5.8.16: pending list is empty though there are pending tasks for both hosts. Gruß, Gundolf [edit]@Blurf: are all pending tasks shown or only very old ones?[/edit] |
aad Send message Joined: 3 Apr 99 Posts: 101 Credit: 204,131,099 RAC: 26 |
My oldest pending is this one; http://setiathome.berkeley.edu/result.php?resultid=1763087319 Pending credit: 15,982.68 |
dskagcommunity Send message Joined: 24 Feb 11 Posts: 43 Credit: 2,901,049 RAC: 0 |
Im a little noob in BOINC and config things with it ^^ But i cant get any work on my WorkPC for my 4350 Card with Cat11.3 with SDK(OpenCL). I checked my preferences to accept other applications but BOINC says that there is no GPU work for me. Hmm. Celeron E3200 2*2,4Ghz 2GB RAM WinXP SP3 32bit HD4350 512MB BOINC 6.10.58 In a thread i saw that i possibly change something to GPU_MAX_HEAP_SIZE=256 but where? O.o Thanks for any halps ^^ |
terencewee* Send message Joined: 10 Oct 09 Posts: 53 Credit: 7,022,510 RAC: 0 |
For a couple of days I left 1 of my cruncher (TREX, C2Duo E6600g0,8GB, ATI5850) running 2 instances after updated from r456 to r516 - then I noticed slow report/credits at 2x CPU-projects (PG+AQUA-fp). I also noticed that after running 2 AP instances from start, BOINC will only run 1 AP instance some time later. I took a look at Tasks-Manager, ap_5.06_win_x86_SSE2_OpenCL_ATI_r516.exe is taking ~80mb RAM, 30-45% CPU. I'm experiencing high CPU load, no wonder! So far, I've tried: 1] Switching -hp on and off 2] Changing -ffa block & fetch sizes 3] Running only 1 instance 4] Initially was using Catalyst APP 11.2, updated to 11.3 - same problem 5] Rebooting 6] Ran Collatz & MilkyWay WU - both ran fine with valid WUs, 0-4% CPU load Processed AP ran and reported without problem. The cruncher is running BOINC 6.12.15. There was 1 validated AP (already removed from DB) and there are 20+ completed AP waiting validation. I've checked another cruncher, 0-4% CPU load - lots of validated/pending AP. Normal... Please help me to resolve this problem - the cruncher has got some 60+ AP WUs to process. Failing which, I will rename r456.exe as r516.exe and replace back the old .cl file to trick BOINC to use r456 instead - won't want to lose these AP WUs. :P Thank you in advance. Edit: 7] Installing SDK 2.3 does not help. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Im a little noob in BOINC and config things with it ^^ I think you saw that in the Lunatics release notes. Also in the release notes is the location of where to change the value. With AP work you may not get tasks every time you request them. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
terencewee* Send message Joined: 10 Oct 09 Posts: 53 Credit: 7,022,510 RAC: 0 |
Im a little noob in BOINC and config things with it ^^ GPU_MAX_HEAP_SIZE is a Windows system environment variable, I think - off the top of my head, in XP - right click on the Computer icon on your desktop > Properties > Advanced > Environment Variables. Create a new variable in your user variable section. Also, go thru this check list - may yield some AP: 1] Look at BOINC-Manage Event Log a+ Message: ATI GPU 0: ATI Radeon HD 4xxxx xxx xxx xxx b+ Message: SETI@home | Found app_info.xml; using anonymous platform Post your app_info.xml file if you don't see 1b. 2] Set *only* SETI@Home to accept more work 3] Set cache size to 10 days 4] Hit update every 5+mins (or automate it) 5] You'll catch an AP (hopefully), let it process and report 6] You'll start to get more AP units after that. Remember to set your cache to 2days after catching enough AP WUs. I have 60+ AP WUs each on 2 crunchers. Best of luck! |
dskagcommunity Send message Joined: 24 Feb 11 Posts: 43 Credit: 2,901,049 RAC: 0 |
Thx for first Infos. I know where i can set variables..but..öhm i need to give the variable a name. How is it named? and the value is then this GPU HEAP size thing? Second problem i dont know where to find this app file. loked at project order, boinc main dir and User directory where boinc is also with his projects. BOINC finds the Graphiccard, so point A is correct. Point B is negative. dont find any app file. I created now a file with this content (from the link from threadstarter) but where i must save it?: <app> <name>astropulse_v505</name> </app> <file_info> <name>ap_5.06_win_x86_SSE2_OpenCL_ATI_r516.exe</name> <executable/> </file_info> <file_info> <name>AstroPulse_Kernels.cl</name> <executable/> </file_info> <app_version> <app_name>astropulse_v505</app_name> <version_num>506</version_num> <platform>windows_intelx86</platform> <avg_ncpus>0.04</avg_ncpus> <max_ncpus>0.20</max_ncpus> <plan_class>ati13ati</plan_class> <cmdline>-instances_per_device 1 -hp -unroll 10 -ffa_block 4096 -ffa_block_fetch 2048</cmdline> <flops>30987654321</flops> <file_ref> <file_name>ap_5.06_win_x86_SSE2_OpenCL_ATI_r516.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>AstroPulse_Kernels.cl</file_name> <copy_file/> </file_ref> <coproc> <type>ATI</type> <count>1</count> </coproc> </app_version> |
terencewee* Send message Joined: 10 Oct 09 Posts: 53 Credit: 7,022,510 RAC: 0 |
Thx for first Infos. Aso-deska... GPU_MAX_HEAP_SIZE is the variable name, the value is 256. 0] Exit from BOINC or stop the BOINC service 1] Unpack the RAR file you downloaded from lunatic site 2] Copy/move those 2 files (.exe and .cl) into (YOUR-BOINC-DATA-FOLDER)\projects\setiathome.berkeley.edu 3] Open NotePad, copy and paste below code <app_info> <app> <name>astropulse_v505</name> </app> <file_info> <name>ap_5.06_win_x86_SSE2_OpenCL_ATI_r516.exe</name> <executable/> </file_info> <file_info> <name>AstroPulse_Kernels.cl</name> <executable/> </file_info> <app_version> <app_name>astropulse_v505</app_name> <version_num>506</version_num> <flops>48000000000</flops> <avg_ncpus>0.04</avg_ncpus> <max_ncpus>0.20</max_ncpus> <plan_class>ati13ati</plan_class> <cmdline>-instances_per_device 1 -hp -unroll 10 -ffa_block 4096 -ffa_block_fetch 2048</cmdline> <-- Other suggested options posted by Claggy -ffa_block 6144 -ffa_block_fetch 1536 -ffa_block 4096 -ffa_block_fetch 1024 -ffa_block 2048 -ffa_block_fetch 512 --> <coproc> <type>ATI</type> <count>1</count> </coproc> <file_ref> <file_name>ap_5.06_win_x86_SSE2_OpenCL_ATI_r516.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>AstroPulse_Kernels.cl</file_name> <copy_file/> </file_ref> </app_version> </app_info> 4] Save as app_info.xml into (YOUR-BOINC-DATA-FOLDER)\projects\setiathome.berkeley.edu ++ Ensure you are using ANSI encoding and the file extension is XML not txt 5] Start BOINC-manager/service AP-go-go-go! :) |
dskagcommunity Send message Joined: 24 Feb 11 Posts: 43 Credit: 2,901,049 RAC: 0 |
Thx!!!!!!! :) Seems ok, need to rename the exe filename in app config file cos he dont find it with your past.. perhaps 1 letter is wrong. so now he founds everything. Now i know what anybody means with this app info thing and custom exe´s :) Got no work to do, so i restarted boinc twice and now i got work!! veryyyyyyyyyyyyyyyyyyyy slow download but doesnt matter ^^ Again very big thx to ya!! Happy crunching and a good afternoon :) *continous looking nasaTV while downloading :)* |
terencewee* Send message Joined: 10 Oct 09 Posts: 53 Credit: 7,022,510 RAC: 0 |
I'm very glad to know S@H got another AP GPU cruncher; you are welcome. Now, back to my problem of high CPU-load - anybody (using r516) having the same problem? Is there a solution? Thanks in advance. Edit: AP WUs by the cruncher - validated successfully. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Did you check AP timings graph I posted recently? Do you see higher CPU times than on graph? |
aad Send message Joined: 3 Apr 99 Posts: 101 Credit: 204,131,099 RAC: 26 |
Results of my HD6970 Win 7 64 cat 11.4 With 3 wu's together on the GPU http://setiathome.berkeley.edu/results.php?hostid=5750222&offset=0&show_names=0&state=0&appid=5 |
Cruncher-American Send message Joined: 25 Mar 02 Posts: 1513 Credit: 370,893,186 RAC: 340 |
Sorry - where is this graph? |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
EDIT: for the record OS is Vista x86 http://setiathome.berkeley.edu/forum_reply.php?thread=63230&post=1097555#1097555 |
Cruncher-American Send message Joined: 25 Mar 02 Posts: 1513 Credit: 370,893,186 RAC: 340 |
Raistmer: I don't see any CPU utilization on these graphs; am I nuts? As I mentioned in another thread, I, like the fellow up above here, was seeing very high CPU utilization for my ATI rig (4 HD 5550s). |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Raistmer: CPU means CPU-only time while elapsed means wall-clock time spent for both CPU and GPU processing. [For GPU apps CPU time usually open symbol while elapsed is filled symbol of same color, only early versions don't follow this rule. But in any case there is legend on graph] |
terencewee* Send message Joined: 10 Oct 09 Posts: 53 Credit: 7,022,510 RAC: 0 |
*red-faced* No, I turn off image in my community preferences - but since corrected. Base on 4 validated AP: http://setiathome.berkeley.edu/result.php?resultid=1872783622 FFA thread block override value:8192 FFA thread fetchblock override value:4096 Run time 4,607.49 CPU time 1,046.74 percent blanked: 7.18 http://setiathome.berkeley.edu/result.php?resultid=1872769917 FFA thread block override value:8192 FFA thread fetchblock override value:4096 Run time 2,848.05 CPU time 689.87 percent blanked: 2.78 http://setiathome.berkeley.edu/result.php?resultid=1872135847 FFA thread block override value:8192 FFA thread fetchblock override value:128 Run time 4,403.89 CPU time 710.10 percent blanked: 2.54 http://setiathome.berkeley.edu/result.php?resultid=1871594876 FFA thread block override value:4096 FFA thread fetchblock override value:2048 Run time 9,880.01 CPU time 643.08 percent blanked: 0.00 I presume I should look at (CPU512x2 - green triagle, not-filled) as there is no 516 in the apgraph.png. Is this correct? A guesstimate is the CPU time is almost similar. Update: I've since tested with config of: <cmdline>-instances_per_device 2 -unroll 12 -ffa_block 16384 -ffa_block_fetch 8192</cmdline> There is no lag at these values. I notice CPU-load is down to a more acceptable level - there is a tick-tock of 0-4%, then 8-20% (most of the time). Some tocks will shoot up to 36%. Previously CPU-load is in the 30%-45% range (constantly). I surmise having higher ffa_blocks actually help in my case. :) ... and having lower ffa-blocks gives higher CPU-load?? Interesting, no? :) <cmdline>-instances_per_device 2 -unroll 12 -ffa_block 32768 -ffa_block_fetch 8192</cmdline> Crashes the app - so no go. |
terencewee* Send message Joined: 10 Oct 09 Posts: 53 Credit: 7,022,510 RAC: 0 |
Raistmer: by "the fellow up above here" you mean moi? :) (and not Him) |
terencewee* Send message Joined: 10 Oct 09 Posts: 53 Credit: 7,022,510 RAC: 0 |
@ Raistmer: I got a new validated AP with the new (higher) ffa-block: http://setiathome.berkeley.edu/result.php?resultid=1872783764 Run time 3,721.21 CPU time 541.29 percent blanked: 0.00 Number of app instances per device setted to:2 DATA_CHUNK_UNROLL setted to:12 FFA thread block override value:16384 FFA thread fetchblock override value:8192 Compared to the previous 0.00 blanked AP, the CPU time is down - but could just mean because Wall-time is down too. Edit: CPU down by 1min 42secs GPU down by 1hour 42mins |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Dependance non-linear and not monotonic so each GPU probably has its own sweet spot in parameter space (that's why I introduced param tuning, in earlier versions all params were hardwired with defaults measured as good for my own GPU). But too low block values mean too many kernel calls so CPU overhead should increase. From other side, too big values mean waste of thread load (cause many thread will process unused data, different periods require different array lengths and threads block always process rectangular data block). So, at too high values overhead (not CPU this time but GPU one, elapsed time) should increase again. Best value somewhere in between. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.