Message boards :
Number crunching :
Replacing ap_6.01r546_sse3_linux32 with ap_6.01r546_avx_linux32
Message board moderation
Author | Message |
---|---|
OTS Send message Joined: 6 Jan 08 Posts: 369 Credit: 20,533,537 RAC: 0 |
Shortly after the ap_6.01r546_sse3_linux32 appeared on the scene I started using it and everything worked as it should. About three months ago I upgraded my CPU to a Xeon E3-1230 V3 and I continued to use the ap_6.01r546_sse3_linux32. Recently however I thought I would try the ap_6.01r546_avx_linux32 version now that I have a CPU to support it. The thought was that it might reduce that amount of time needed to process a WU. I read the install instructions and in a nut shell it said to stop boinc, replace the sse3 binary with the avx binary, replace the app_info.xml file, and then start boinc again. I did that and when boinc was started it appeared to pick up where it left off on the WUs in progress. It wasn’t long however before the first results were reported but they were all found to be inconclusive. This made me a little nervous but I attributed it to the fact it was a WU that was worked on first by the sse3 binary and finished off by the avx binary which probably was not a good idea. Eventually a new work that was run entirely under the avx binary also reported in and it too had results that were inconclusive. They were all later updated to invalid so I have gone back to using the sse3 binary. The question is, what went wrong? Am I mistaken about the E3-1230 v3 supporting avx? Did I do something wrong during the install? And most importantly, is it even worth pursuing? Does the avx binary really buy you shorter processing time over the sse3 binary? I assumed it would or the author would not have produced both an sse3 and avx version, but that could be an incorrect assumption on my part. |
petri33 Send message Joined: 6 Jun 02 Posts: 1668 Credit: 623,086,772 RAC: 156 |
I'm using the 64 bit version of the AVX version from Lunatics on my i7-3930K under linux. It works perfectly. My run times for AP tasks are about 13500 seconds per wu. I do six of them at a time and at the same time there are 6 GPU MB or 8 GPU AP tasks running. You'll find the CPU AP tasksk from my Host . To overcome Heisenbergs: "You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones |
OTS Send message Joined: 6 Jan 08 Posts: 369 Credit: 20,533,537 RAC: 0 |
Does anyone at least have a hint on the difference in performance between the 32 bit Linux SSE3 and AVX? There is no sense trying to figure out what went wrong if there is nothing to be gained. The converse is also true. Perhaps there is a good reason to try to figure it out. I just don't know. |
Urs Echternacht Send message Joined: 15 May 99 Posts: 692 Credit: 135,197,781 RAC: 211 |
dsh, did you delete the old "wisdom.dat" file in your projects/setiathome.berkely.edu/ directory? The old one was created by ap_6.01r546_sse3_linux32 without avx-support. The avx-app ap_6.01r546_avx_linux32 should create a new wisdom.dat file right away. If you open that new file after creation it should contain codelets with avx in the name (example: fftwf_codelet_t3fv_8_avx ...) _\|/_ U r s |
OTS Send message Joined: 6 Jan 08 Posts: 369 Credit: 20,533,537 RAC: 0 |
Thanks for taking the time to reply. I am however a little confused. I have never seen a wisdom.dat in /usr/sbin/BOINC/projects/setiathome.berkeley.edu and "slocate wisdom.dat" comes up empty for the whole server . It sounds a little like a Windows file. Should there be one in a Linux install as well? |
Urs Echternacht Send message Joined: 15 May 99 Posts: 692 Credit: 135,197,781 RAC: 211 |
Do you have a separate directory, where all your workunit-files/results and app-info.xml are stored during crunching ? The wisdom.dat file should be there and NO it is not a Windows file. _\|/_ U r s |
petri33 Send message Joined: 6 Jun 02 Posts: 1668 Credit: 623,086,772 RAC: 156 |
Do you have a separate directory, where all your workunit-files/results and app-info.xml are stored during crunching ? Pure wisdom, on the right ---> 16au13ac.11382.22971.438086664201.12.161 16my13ab.27104.242925.438086664195.12.67_1_0 ap_24jn13aa_B3_P1_00270_20140122_03115.wu 16au13ac.11382.22971.438086664201.12.161_0_0 16my13ab.27104.242925.438086664195.12.95 ap_24jn13aa_B3_P1_00270_20140122_03115.wu_3_0 16au13ac.11382.885.438086664201.12.211 16my13ab.27104.242925.438086664195.12.95_1_0 ap_6.01r546_avx_linux64 16au13ac.11382.885.438086664201.12.211_1_0 16my13ab.27104.9038.438086664195.12.245 ap_6.07r1952_avx_clGPU_x86_64-pc-linux-gnu 16au13ac.11382.885.438086664201.12.223 16my13ab.27104.9038.438086664195.12.245_0_0 AP_clFFTplan_GeForceGTX780_32768_r1844.bin 16au13ac.11382.885.438086664201.12.223_1_0 16my13ab.32198.4668.438086664196.12.223.vlar app_info.xml 16au13ac.12575.19290.438086664202.12.69 16my13ab.32198.4668.438086664196.12.223.vlar_1_0 app_info.xml~ 16au13ac.12575.19290.438086664202.12.69_1_0 16my13ab.7694.243845.438086664200.12.178 arecibo_181.png 16au13ac.12575.19290.438086664202.12.8 16my13ab.7694.243845.438086664200.12.178_0_0 AstroPulse_Kernels_r1844.cl 16au13ac.12575.19290.438086664202.12.8_1_0 16my13ab.7694.243845.438086664200.12.218 AstroPulse_Kernels_r1844.cl_GeForceGTX780.bin_V6 16au13ac.12575.21744.438086664202.12.212 16my13ab.7694.243845.438086664200.12.218_0_0 AstroPulse_Kernels_r1952.cl 16au13ac.12575.21744.438086664202.12.212_1_0 16my13ab.7694.243845.438086664200.12.220 AstroPulse_Kernels_r1952.cl~ 16au13ac.15991.20108.438086664198.12.194 16my13ab.7694.243845.438086664200.12.220_1_0 AstroPulse_Kernels_r1952.cl.bak 16au13ac.15991.20108.438086664198.12.194_1_0 16my13ab.7694.243845.438086664200.12.226 AstroPulse_Kernels_r1952.cl.bak.2 16au13ac.15991.20108.438086664198.12.207 16my13ab.7694.243845.438086664200.12.226_1_0 bak 16au13ac.15991.20108.438086664198.12.207_1_0 16my13ab.7694.249162.438086664200.12.9 boinc_lockfile 16au13ac.15991.20108.438086664198.12.218 16my13ab.7694.253661.438086664200.12.156 MBv7_7.05r2113_avx_x86_64-pc-linux-gnu 16au13ac.15991.20108.438086664198.12.218_1_0 16my13ab.7694.253661.438086664200.12.156_1_0 r1844_IntelRCoreTMi73930KCPU320GHz.wisdom 16au13ac.15991.20108.438086664198.12.224 16my13ab.7694.253661.438086664200.12.28 r2113_IntelRCoreTMi73930KCPU320GHz_x64.wisdom 16au13ac.15991.20108.438086664198.12.224_1_0 16my13ab.7694.253661.438086664200.12.28_1_0 sah_40.png 16au13ac.15991.20108.438086664198.12.230 17my13ab.11144.16427.438086664198.12.158 sah_banner_290.png 16au13ac.15991.20108.438086664198.12.230_1_0 17my13ab.11144.16427.438086664198.12.158_1_0 sah_ss_290.png 16au13ac.24161.11212.438086664200.12.151 17my13ab.11144.16427.438086664198.12.194 setiathome_x41g_x86_64-pc-linux-gnu_cuda32 16au13ac.24161.11212.438086664200.12.151_1_0 17my13ab.11144.16427.438086664198.12.194_1_0 setiathome_x41zc_x86_64-pc-linux-gnu_cuda55 16au13ac.24161.11212.438086664200.12.154 17my13ab.11144.1703.438086664198.12.104 slideshow_setiathome_enhanced_00 16au13ac.24161.11212.438086664200.12.154_0_0 17my13ab.11144.1703.438086664198.12.104_0_0 slideshow_setiathome_enhanced_01 16au13ac.24161.11212.438086664200.12.63 17my13ab.11144.1703.438086664198.12.110 slideshow_setiathome_enhanced_02 16au13ac.24161.11212.438086664200.12.63_1_0 17my13ab.11144.1703.438086664198.12.110_1_0 stat_icon 16au13ac.24161.11212.438086664200.12.76 17my13ab.11144.1703.438086664198.12.150 16au13ac.24161.11212.438086664200.12.76_1_0 17my13ab.11144.1703.438086664198.12.203 linux1 petri # To overcome Heisenbergs: "You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones |
OTS Send message Joined: 6 Jan 08 Posts: 369 Credit: 20,533,537 RAC: 0 |
Do you have a separate directory, where all your workunit-files/results and app-info.xml are stored during crunching ? My apologies if I offended you or anyone else in the way I answered previously. It is just that I have seen a lot more .dat files on Windows machines so it seemed strange particularly when I didn't see one in my working directory where I was looking. This is what I have currently in my BOINC/projects/setiathome.berkelley.edu directory. ap_01au13aa_B2_P1_00308_20140214_31533.wu ap_01au13aa_B2_P1_00308_20140214_31533.wu_2_0 <snip> ap_31mr13ad_B2_P1_00360_20140130_09219.wu ap_31mr13ad_B6_P1_00150_20140130_32100.wu ap_6.01r546_sse3_linux32 app_info.xml arecibo_181.png sah_40.png sah_banner_290.png sah_ss_290.png slideshow_setiathome_enhanced_00 slideshow_setiathome_enhanced_01 slideshow_setiathome_enhanced_02 stat_icon The good news is that I must have typed wisdom.dat in wrong the first time I used slocate as I tried it again and this time I did find it, or actually, them. In my setup they are located one each in /usr/sbin/BOINC/slots/x where x is 0 through 7. This is a typical listing for each one. ap_6.01r546_sse3_linux32 ap_state.dat0 ap_state.dat1 boinc_lockfile fold.dat0 fold.dat1 in.dat indices.txt init_data.xml pulse.out pulse.out0 pulse.out1 stderr.txt wisdom.dat Now that I have found them my best guess is that I should wait until all my WUs are finished and then delete everything, install the avx binary, and then let the rest be recreated. I will wait however until I hear if that is really the case. Seems that it might now be a good time to do that since there won't be any further APs for a while :(. |
Urs Echternacht Send message Joined: 15 May 99 Posts: 692 Credit: 135,197,781 RAC: 211 |
Don't feel offended in any way here. The way you seem to have optimized AstroPulse working is rather unusual and was not meant in that way. "Your way" means : recreating wisdom.dat at start of every new workunit in the slot directory, wasting time. Check if "wisdom.dat" eventually is a link to the real file. If not : Is your BOINC version a service install ? Which Linux distribution are you using ? Eventually i can reproduce your situation and see if the (old, 6.2.14) service install is the reason for no reusable "wisdom.dat" in the projects directory. _\|/_ U r s |
OTS Send message Joined: 6 Jan 08 Posts: 369 Credit: 20,533,537 RAC: 0 |
<snip> Don't feel offended in any way here. Listed below are the files with what I found in them when I used “catâ€. It looks to me that wisdom.dat is not a soft link but holds the codelets as you described in your first posting. I am running Slackware and I think the version is 14.0 since that was the current version when I did the OS install in August of last year. The kernel is definitely version 3.10.7. As for when and how the original boinc install was done I am afraid I do not recall. This is at least the third server the installation has been in as a result of upgrades to the motherboard and CPU along with the OS. In each case I moved a lot of the old configuration files from the old server to the new so I didn’t have to reinvent the wheel. This includes the complete /usr/bin/BOINC directory, rc.local which launches boinc, and some bash script files I cobbled together to make sure SETI was running and updating. I can say that I do not remember making the directory structure so there is a good chance some sort of install created it. I just cannot say that with 100% certainty. ap_6.01r546_sse3_linux32 soft link to binary ap_state.dat0 non text data ap_state.dat1 non text data boinc_lockfile empty file fold.dat0 non text data fold.dat1 non text data in.dat softlink to WU indices.txt text data init_data.xml text data pulse.out softlink to WU intermediate results pulse.out0 softlink to WU intermediate results pulse.out1 softlink to WU intermediate results stderr.txt text data wisdom.dat text data |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
I think it depends on the executable where it creates/looks for the 'wisdom' file (some look in directory of executable, some in current (slot) directory) Optimized apps for Windows look in directory of executable. Stock apps for Windows seem to look in current (slot) directory. So on my old Windows 98 computer (where I have to use the stock apps but with app_info.xml to supply different DLL) I use: <file_ref> <file_name>wisdom.sah</file_name> <copy_file/> </file_ref> Full app_info.xml is (just as an example, near the end is <copy_file/>): <app_info> <app> <name>setiathome_v7</name> </app> <file_info> <name>setiathome_7.00_windows_intelx86.exe</name> <executable/> </file_info> <file_info> <name>setigraphics_7.00_windows_intelx86.exe</name> <executable/> </file_info> <file_info> <name>libfftw3f-3-3_upx.dll</name> <executable/> </file_info> <file_info> <name>Psapi.dll</name> <executable/> </file_info> <file_info> <name>wisdom.sah</name> </file_info> <app_version> <app_name>setiathome_v7</app_name> <version_num>700</version_num> <file_ref> <file_name>setiathome_7.00_windows_intelx86.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>setigraphics_7.00_windows_intelx86.exe</file_name> </file_ref> <file_ref> <file_name>libfftw3f-3-3_upx.dll</file_name> </file_ref> <file_ref> <file_name>Psapi.dll</file_name> <copy_file/> </file_ref> <file_ref> <file_name>wisdom.sah</file_name> <copy_file/> </file_ref> </app_version> </app_info> So (for faster app startup) you need: - copy wisdom.dat from slot to where ap_6.01r546_sse3_linux32 binary is - add to your app_info.xml in the appropriate place: <file_ref> <file_name>wisdom.dat</file_name> <copy_file/> </file_ref> Do the same again when you change the binary. For BOINC <copy_file/> means: "copy this file from the directory where app_info.xml is to the working slot directory" Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
OTS Send message Joined: 6 Jan 08 Posts: 369 Credit: 20,533,537 RAC: 0 |
Thanks. I will give your suggestions a try, but I hope you will forgive me if I wait until APs are being issued again. The thought doing something wrong and ending up with my meager current queue of APs being abandoned does not thrill me. I will report back on my success, or lack of if that is the case, when I do try it. Thanks again. |
©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.