Message boards :
Number crunching :
Lunatics Windows Installer v0.40 release notes
Message board moderation
Previous · 1 . . . 8 · 9 · 10 · 11 · 12 · 13 · 14 · Next
Author | Message |
---|---|
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
Uhh... Fred's wonderful tool allows for setting the priority of tasks. On my crunch only rigs, I set it to maximum. On my daily driver, I set it to minimum, to avoid lag of the usual tasks. Works just grand. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
I have noticed huge blanking in almost all AP wu... Radar is certainly the prime offender, and there are two separate ways of dealing with its effects. First, before the data is split it goes through an examination which combines channels and looks for known radar patterns. A signal is put on a 15th channel indicating where the patterns are seen, then as the data is being loaded for splitting those sections are replaced with pseudo random data shaped to fit the frequency profile of the good parts of the data. There's no indication in the WUs produced during the splitting what sections are real and what replaced, though. Second, if there's a <remove_radar>1</remove_radar> element in an AP WU header, the application does an analysis which senses RFI by its tendency to add a DC component. That analysis makes a list of where in the 32 Mebisamples there are 128 sample sections with DC above a threshold. (While an AP task is being crunched, that list is in the slot directory as indices.txt.) For each such section, the application considers the preceding and following 400000 samples bad. In cases where there are enough to effectively mark all the data as bad, the application exits with the "...Blanking too much RFI?" message. Otherwise, searching for pulses commences, but for any data chunk being analyzed which is considered bad, that chunk is replaced with the same kind of pseudo random shaped data as used server-side. The application keeps track of how much data it has replaced and writes the <fraction_blanked> in the result file, Lunatics apps also put it in stderr as percent blanked. Joe |
Michael W.F. Miles Send message Joined: 24 Mar 07 Posts: 268 Credit: 34,410,870 RAC: 0 |
Radar is certainly the prime offender, and there are two separate ways of dealing with its effects. First, before the data is split it goes through an examination which combines channels and looks for known radar patterns. A signal is put on a 15th channel indicating where the patterns are seen, then as the data is being loaded for splitting those sections are replaced with pseudo random data shaped to fit the frequency profile of the good parts of the data. There's no indication in the WUs produced during the splitting what sections are real and what replaced, though. Second, if there's a <remove_radar>1</remove_radar> element in an AP WU header, the application does an analysis which senses RFI by its tendency to add a DC component. That analysis makes a list of where in the 32 Mebisamples there are 128 sample sections with DC above a threshold. (While an AP task is being crunched, that list is in the slot directory as indices.txt.) For each such section, the application considers the preceding and following 400000 samples bad. In cases where there are enough to effectively mark all the data as bad, the application exits with the "...Blanking too much RFI?" message. Otherwise, searching for pulses commences, but for any data chunk being analyzed which is considered bad, that chunk is replaced with the same kind of pseudo random shaped data as used server-side. The application keeps track of how much data it has replaced and writes the <fraction_blanked> in the result file, Lunatics apps also put it in stderr as percent blanked. Thanks for the very detailed answer. Have a great day Michael Miles |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
If you have a CPU with SSE2 and above but not AVX the r557 MAY be faster. I don't know if anybody needs this info but in case somebody wonders: ap_6.01r557_SSE2_331_AVX.exe seems to work OK and is faster on my AMD Athlon(tm) 64 Processor 3500+ http://setiathome.berkeley.edu/show_host_detail.php?hostid=4832843 Till now 2 AstroPulse v6 tasks are finished using ap_6.01r557_SSE2_331_AVX.exe (but not yet validated): http://setiathome.berkeley.edu/result.php?resultid=2374705716 http://setiathome.berkeley.edu/result.php?resultid=2377424264 (they took 33-34 hours CPU time while the older ap_5.05r409_SSE.exe was using 35-38 hours CPU time for AstroPulse v505 tasks) To test that on my CPU both r555 SSE and r557 SSE2 give the same results and which of them is faster (for me) I used these files: "AP bench 2.10" and "Shortened AP test WU for offline speed tests" from here: http://lunatics.kwsn.net/index.php?module=Downloads;catd=44 and 2 apps extracted by 7-Zip from Lunatics_Win32_v0.40_setup.exe I put the files this way (folder/files): ...\APbench\Science_apps\Reference\ AP6_win_x86_SSE_CPU_r555.exe libfftw3f-3.dll ...\APbench\Science_apps\ ap_6.01r557_SSE2_331_AVX.exe ...\APbench\TestWUs\ ap_18se08aa_B6_P1_00046_1LC25.wu Then I started APbench210.cmd The results from ...\APbench\Testdatas\{computer-name}-20120401-0608-benchAP.txt (shows that the results match and r557 is about 7% faster than r555 on this CPU using this WU) Quick timetable WU : ap_18se08aa_B6_P1_00046_1LC25.wu AP6_win_x86_SSE_CPU_r555.exe -verbose : Elapsed 1248.953 secs CPU 1198.750 secs ap_6.01r557_SSE2_331_AVX.exe -instances_per_device 1 -hp -unroll 16 -ffa_block 4096 -ffa_block_fetch 2048 : Elapsed 1164.703 secs, speedup: 6.75% ratio: 1.07x CPU 1120.031 secs, speedup: 6.57% ratio: 1.07x ------------ CPU: Number of processors 1 Number of cores 1 (max 1) Specification AMD Athlon(tm) 64 Processor 3500+ Codename Orleans Core Stepping DH-F2 Technology 90 nm Instructions sets MMX (+), 3DNow! (+), SSE, SSE2, SSE3, x86-64, AMD-V L1 Data cache 64 KBytes, 2-way set associative, 64-byte line size L1 Instruction cache 64 KBytes, 2-way set associative, 64-byte line size L2 cache 512 KBytes, 16-way set associative, 64-byte line size Core Speed 2325.1 MHz Multiplier x FSB 11.0 x 211.4 MHz HT Link speed 845.5 MHz Stock frequency 2200 MHz ------------ Chipset: Northbridge NVIDIA MCP61 rev. A2 Southbridge NVIDIA MCP61 rev. A2 ------------ RAM: Memory Type DDR2 Memory Size 1024 MBytes Channels Dual Memory Frequency 387.5 MHz (CPU/6) Max bandwidth PC2-6400 (400 MHz) CAS# latency (CL) 5.0 RAS# to CAS# delay (tRCD) 5 RAS# Precharge (tRP) 5 Cycle Time (tRAS) 18 Bank Cycle Time (tRC) 24 Command Rate (CR) 2T ------------ ref-AP6_win_x86_SSE_CPU_r555.exe-ap_18se08aa_B6_P1_00046_1LC25.wu.res: <ap_signal>11,<pulses>1,<best_pulses>10 result-ap_6.01r557_SSE2_331_AVX.exe-ap_18se08aa_B6_P1_00046_1LC25.wu.res: <ap_signal>11,<pulses>1,<best_pulses>10 All Signals: Checked 11, 11 , Strongly Similar Pulses: Checked 1, 1 , Strongly Similar Best Pulses: Checked 10, 10 , Strongly Similar -(.\testDatas\ref\ref-AP6_win_x86_SSE_CPU_r555.exe-ap_18se08aa_B6_P1_00046_1LC25.wu.res)- Reportable Single Pulses: 0 [OK], 0 above threshold*THRESHOLD_FUDGE Reportable Repeating Pulses: 1 [OK] Single Pulses (Best): 10 [OK], 0 above threshold*THRESHOLD_FUDGE -(.\testDatas\result-ap_6.01r557_SSE2_331_AVX.exe-ap_18se08aa_B6_P1_00046_1LC25.wu.res)- Reportable Single Pulses: 0 [OK], 0 above threshold*THRESHOLD_FUDGE Reportable Repeating Pulses: 1 [OK] Single Pulses (Best): 10 [OK], 0 above threshold*THRESHOLD_FUDGE Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
skildude Send message Joined: 4 Oct 00 Posts: 9541 Credit: 50,759,529 RAC: 60 |
ummm The only reason that it is faster is the SSE2. I'm willing to be that since your CPU doesn't have AVX you are basically just running it off the SSE2 portion of the app In a rich man's house there is no place to spit but his face. Diogenes Of Sinope |
skildude Send message Joined: 4 Oct 00 Posts: 9541 Credit: 50,759,529 RAC: 60 |
Radar is certainly the prime offender, and there are two separate ways of dealing with its effects. Please actually quote someone if you intend to quote someone. Copy and pasting their posts gets a bit annoying when I have to pick out your post from anothers. You notice on my post you can clearly see that I quoted you. On your, not so much In a rich man's house there is no place to spit but his face. Diogenes Of Sinope |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
ummm The only reason that it is faster is the SSE2. I'm willing to be that since your CPU doesn't have AVX you are basically just running it off the SSE2 portion of the app I mentioned this already. r557 is faster on AMDs without using AVX. With each crime and every kindness we birth our future. |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
ummm The only reason that it is faster is the SSE2. I'm willing to be that since your CPU doesn't have AVX you are basically just running it off the SSE2 portion of the app There's no "SSE2 portion of the app" as such. Jason needed SSE2 compiler settings to get the statically linked FFTW with AVX to build r557 as he wanted, IIRC. It's a different source branch than Raistmer's r555 build, too, so there are other minor differences. Our testing indicated Jason's build clearly better on systems which can use AVX, Raistmer's build obviously the only possibility for SSE systems, and mixed results for all the rest with SSE2 or above but not AVX. YMMV definitely applies. Joe |
achim.huber Send message Joined: 29 Apr 01 Posts: 1 Credit: 6,132,100 RAC: 0 |
The new ATI AP app doesn't work on my HD 4670. The program crashes after a few seconds. The previous AP app r516 didn't make any problems. http://setiathome.berkeley.edu/result.php?resultid=2381249092 http://setiathome.berkeley.edu/result.php?resultid=2380377224 http://setiathome.berkeley.edu/result.php?resultid=2380585788 |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
You have a 512 MB card. One of the messages say out of memory. Have to check how much its consuming. Edit: With your settings the app only uses 150 MB so shouldn´t be a problem. Did you try newer drivers ? 11.5. wasn´t the best one. Edit 2: No need to change drivers. With each crime and every kindness we birth our future. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
The new ATI AP app doesn't work on my HD 4670. The program crashes after a few seconds. The previous AP app r516 didn't make any problems. If anyone uses r555 AP with any of HD4xxx please make report if it work for you. For now we have reports from WG128 cards where different problems can occur. I would like to see report from HD4870 card ... |
David S Send message Joined: 4 Oct 99 Posts: 18352 Credit: 27,761,924 RAC: 12 |
How many people have posted in this thread? Without counting, I'd guess at fewer than 20. What do the other 660 think? I'm coming into this fray fresh and probably somewhat naive. I've never used optimized apps before, but I started reading this thread because I'm ready (I think) to take the plunge. Also, I have not read all the way to the end of this thread, but I wanted to respond to something at this point. Many software installers start the user off with a screen where they can choose a standard or express install that sets them up with the most common options; or an advanced install where one or more additional screens walk them through all the available options and let them choose. Could this installer implement something like that? The advanced screens would include choices for setting aside a CPU core or not, whether to leave some GPU performance available for regular use, etc. (If this installer already does that, just tell me so and I'll shut up. Like I said, I'm just getting ready to do this for the first time.) David Sitting on my butt while others boldly go, Waiting for a message from a small furry creature from Alpha Centauri. |
skildude Send message Joined: 4 Oct 00 Posts: 9541 Credit: 50,759,529 RAC: 60 |
First get GPUz and CPUz if you are unsure what gpu or CPU instruction set your CPU can run. Next just try to run the installer. it takes several steps before it actually starts installing anything. You get to choose what apps you want and it creates an app_info for your install. Once it is done it will restart your BOINC manager and you should see different apps running in your windows task manager. In a rich man's house there is no place to spit but his face. Diogenes Of Sinope |
Stick Send message Joined: 26 Feb 00 Posts: 100 Credit: 5,283,449 RAC: 5 |
I am having problems with SETI7, r390 on my 6200/6300/7300 series (Wrestler). Seems to work OK if the task is never interrupted but crashes if BOINC switches the GPU to a different project. The previous version Lunatics app worked just fine. http://setiathome.berkeley.edu/result.php?resultid=2381264891 http://setiathome.berkeley.edu/result.php?resultid=2380811374 http://setiathome.berkeley.edu/result.php?resultid=2380564418 http://setiathome.berkeley.edu/result.php?resultid=2380564416 EDIT: I updated to BOINC 7.0.22 after installing r390. |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
I am having problems with SETI7, r390 on my 6200/6300/7300 series (Wrestler). Seems to work OK if the task is never interrupted but crashes if BOINC switches the GPU to a different project. The previous version Lunatics app worked just fine. It's a known bug in 7.0.22 and 7.0.23 interacting with a slightly delayed exit time for r390, see changeset [trac]changeset:25498[/trac] for when the bug was fixed. Back off to 7.0.20 (or earlier) or upgrade to 7.0.24 (or later). {Edit:} The r390 application is not new, the v0.39 installer had the same S@H Enhanced applications as v0.40 does. Only Astropulse v6 is new. Joe |
skildude Send message Joined: 4 Oct 00 Posts: 9541 Credit: 50,759,529 RAC: 60 |
just dont back down to the 6.12.34 or you'll lose all your work. In a rich man's house there is no place to spit but his face. Diogenes Of Sinope |
B-Man Send message Joined: 11 Feb 01 Posts: 253 Credit: 147,366 RAC: 0 |
just dont back down to the 6.12.34 or you'll lose all your work. Isn't it don't back down to any 6.x.x version from any recent 7.0.x version. Sort of like once you go version 7.0.x you can't go back to version 6.x BOINC. |
Stick Send message Joined: 26 Feb 00 Posts: 100 Credit: 5,283,449 RAC: 5 |
It's a known bug in 7.0.22 and 7.0.23 interacting with a slightly delayed exit time for r390, see changeset [trac]changeset:25498[/trac] for when the bug was fixed. Back off to 7.0.20 (or earlier) or upgrade to 7.0.24 (or later). Thanks Joe, et. al. for the replies! After I posted, I realized that the Lunatics SETI7 app hadn't changed with v0.40 and that it had to be a BOINC issue - hence my edit regarding 7.0.22. But I wasn't aware that the issue had already been identified and fixed. Again, thanks to all for the quick replies and advice. |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
Not going well lately with the ATI app for AP v6. Over the past couple of days, I've seen the time the ATI GPU uses on the APs increase, now even so much that the default app on an i5-2500 CPU is faster than the r555 app on my ATI HD6850. I'll be shutting it down, if at least that one AP that's been going at it for the past 17 hours (!!) will ever finish. The last long one: 2378675057 6065655 1 Apr 2012 | 1:24:32 UTC 2 Apr 2012 | 2:49:35 UTC Completed and validated 78,139.64 173.27 473.17 AstroPulse v6 Anonymous platform (ATI GPU) 2378681005 6376207 1 Apr 2012 | 1:29:54 UTC 4 Apr 2012 | 15:53:57 UTC Completed and validated 59,607.90 57,617.44 473.17 AstroPulse v6 v6.01 |
Wiggo Send message Joined: 24 Jan 00 Posts: 34744 Credit: 261,360,520 RAC: 489 |
Not going well lately with the ATI app for AP v6. You just have to be amazed at how well a 2500K chews through those v6 AP's don't you? ATM my average for them is around 5.75hrs but it will be interesting to see what happens when I clock the thing up when the cool/cold weather settles in here. Cheers. |
©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.