Message boards :
Number crunching :
The Highest Ranked SETI AMD Host is a MAC: Time for a STOCK MAC APP?
Message board moderation
Previous · 1 . . . 18 · 19 · 20 · 21
Author | Message |
---|---|
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I will crunch that work unit shortly. It seems like my RAC is only about 10,000 crunching MB units where I could hit 20,000 crunching AP units. Of that number, about 5,000 credits come from crunching MB units on the CPU. Yes, you won't break any records crunching MBs, especially with an ATI/AMD card. The general difference is about 25% for crunching MBs on ATIs verses APs. Look at the run times, about 23 minutes for 110 MB credits verses 27 minutes for 550 AP credits on the same card...that's using the old math. After compiling my first CPU App I looked around at different machines. I noticed some machines did much better than others with the stock App and even the Optimized r2549 at Lunatics. I really don't have a clue why there's such a difference and there's not much you can do about it. One thing you can try is to install the FFTW package with the SIMD options. The options are listed in section 10 of the fftw3.pdf manual. Basically you want a custom configure line, for an AVX CPU something like; ./configure --enable-float --enable-avx --enable-threads --with-combined-threads It might not have any effect at all, in which case, you can always try my AVX App at Crunchers Anonymous. The one person that left feedback has had positive results over the previous App. This is one machine using the App, Intel(R) Core(TM) i7-2675QM CPU @ 2.20GHz Hope that helps. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
So....has anyone been able to have their new Mac's CPUs use AVX rather than SSEx? The best I can tell, just about All of them are using SSEx on MBs and APs. They would be much faster using AVX... |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I'm still not having any luck compiling the AstroPulse CPU App in any OSX. It's always the same Error. This Time I used the new OSX 10.10.3 with the OS minimum set for 10.10, the same as the AP GPU App I compiled earlier. The GPU App compiled and runs without any problem, http://setiathome.berkeley.edu/result.php?resultid=4083492725. The CPU App Ends with; 8 warnings generated. mv -f .deps/ap_client-ap_remove_radar.Tpo .deps/ap_client-ap_remove_radar.Po g++ -I/opt/local/include -mmacosx-version-min=10.10 -O3 -march=core2 -mtune=core2 -DUSE_I386_OPTIMIZATIONS -DUSE_I386_CORE2 -DUSE_I386_XEON -DHALF_STRIDE -I./ -I./Users/Tom/sah_v7_opt/src -I./Users/Tom/sah_v7_opt/AKv8 -msse4.1 -mmacosx-version-min=10.10 -O3 -march=core2 -mtune=core2 -DUSE_I386_OPTIMIZATIONS -DUSE_I386_CORE2 -DUSE_I386_XEON -DHALF_STRIDE -I./ -I./Users/Tom/sah_v7_opt/src -I./Users/Tom/sah_v7_opt/AKv8 -ffast-math -DHAVE_CONFIG_H -DTEXT_UI -DNDEBUG -DCLIENT -I../server/db -I/Users/Tom/sah_v7_opt/AKv8/../src -I/Users/Tom/sah_v7_opt/AKv8 -I/Users/Tom/sah_v7_opt/AKv8/db -I/Users/Tom/sah_v7_opt/AKv8/client -I/Users/Tom/boinc -I/Users/Tom/boinc/api -I/Users/Tom/boinc/lib -I/Users/Tom/boinc/sched -I/Users/Tom/boinc/db -D_THREAD_SAFE -D_THREAD_SAFE -mmacosx-version-min=10.10 -march=core2 -lfftw3f -ffast-math -ldl -lz -lpthread -framework Carbon -lm -L/opt/local/lib -mmacosx-version-min=10.10 -march=core2 -lfftw3f -ffast-math -ldl -lz -lpthread -framework Carbon -o ap_client ap_client-sqlblob.o ap_client-sqlrow.o ap_client-xml_util.o ap_client-lcgamm.o ap_client-ap_schema.o ap_client-ap_client_main.o ap_client-ap_science.o ap_client-ap_fileio.o ap_client-ap_fold.o ap_client-ap_timer.o ap_client-ap_debug.o ap_client-mtrand.o ap_client-ap_version.o ap_client-malloc_a.o ap_client-GPU_lock.o ap_client-sbtf.o ap_client-ap_shmem.o ap_client-ap_remove_radar.o -L/Users/Tom/boinc/api -L/Users/Tom/boinc/api/.libs -lboinc_api -L/Users/Tom/boinc/lib -L/Users/Tom/boinc/lib/.libs -lboinc -lfftw3f ../../lib/OSX64/libfftw3f.a -lz -lstdc++ -lm -lpthread -ldl /Users/Tom/sah_v7_opt/lib/OSX64/libfftw3f.a -lz -lstdc++ -lm -lpthread -ldl /Users/Tom/sah_v7_opt/lib/OSX64/libfftw3f.a ld: warning: directory not found for option '-L/opt/local/lib' Undefined symbols for architecture x86_64: "Astropulse::number_of_logged_signals", referenced from: Astropulse::add_signal(std::__1::vector<ap_signal, std::__1::allocator<ap_signal> >&, ap_signal) in ap_client-ap_fileio.o ld: symbol(s) not found for architecture x86_64 clang: error: linker command failed with exit code 1 (use -v to see invocation) make: [ap_client] Error 1 (ignored) /bin/cp ap_client ap_7.01r2873_sse41_x86_64-apple-darwin cp: ap_client: No such file or directory make: [ap_7.01r2873_sse41_x86_64-apple-darwin] Error 1 (ignored) strip ap_7.01r2873_sse41_x86_64-apple-darwin error: /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/strip: can't open file: ap_7.01r2873_sse41_x86_64-apple-darwin (No such file or directory) make: [ap_7.01r2873_sse41_x86_64-apple-darwin] Error 1 (ignored) /bin/rm -f astropulse-7.01_x86_64-apple-darwin.debug /bin/ln ap_client astropulse-7.01_x86_64-apple-darwin.debug ln: ap_client: No such file or directory make: [astropulse-7.01_x86_64-apple-darwin.debug] Error 1 (ignored) /Applications/Xcode.app/Contents/Developer/usr/bin/make all-am Along with successfully compiling the AP GPU App, I can also successfully compile the AP CPU App 6.09. But the APv7 CPU compile fails...always with the same Error no matter what configure lines are used or which OS is used. This machine used to complete an unblanked APv6 task in 8 hours, the current unblanked v7 task is close to 10 hours. I have an older Conroe Core2 Xeon that completes a v7 AP in under 8 hours in XP or Linux, http://setiathome.berkeley.edu/result.php?resultid=4082160019. You would think this Harpertown Core2 Xeon would be just as fast...it was in APv6. I had success with the MBv7 CPU App, it would be nice if I could compile the APv7 CPU App. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
I'm not sure you have right set of defines. To build SSE2 AP on Windows one should use: BLANKIT;SMALL_CHIRP_TABLE;USE_LRINT;TWINDECHIRP;USE_VIRTUAL_ALLOC;AP_CLIENT;WIN32;_MT;NDEBUG;_WINDOWS;_CONSOLE;HAVE_STD_MAX;HAVE_STD_MIN;HAVE_STD_TRANSFORM;CLIENT;USE_CONVERSION_OPT;USE_SSE;USE_FFTW;USE_INCREASED_PRECISION |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I started with the lines here, https://setisvn.ssl.berkeley.edu/trac/browser/branches/sah_v7_opt/AP_BLANKIT/configure_line_AP_OSX64_sse3_OpenCL.txt I've added a few items with different attempts. There have been many different attempts. If you could post any changes here to the above line I will give it a try. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
I started with the lines here, https://setisvn.ssl.berkeley.edu/trac/browser/branches/sah_v7_opt/AP_BLANKIT/configure_line_AP_OSX64_sse3_OpenCL.txt It's Urs' domain, to build under OS X :) I can only suggest something that definitely works on Windows. If specific lines present I suppose app was built OK with them before line was committed. Hence real issue lies somewhere else. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Well, I still haven't been able to Compile an APv7 CPU App in Any OSX. I Can Compile a APv6 CPU App though, and an APv7 GPU App. However, The MBv7 CPU Apps have been added to Beta for Testing. So, anyone that wants to see a New MBv7 CPU App on Main should go to Beta and Test the MBv7 CPU Apps. Since all the trouble with the MB tasks at Beta it appears most people there have turned Off MB CPU work. The problems with the Beta MB tasks appear to be much better, so, you need to turn MB CPU tasks back on and test the the New MB Apps. There are New Mac MB Apps across the board, just waiting to be tested; Mac OS X/64-bit Intel 7.05 (avx) 21 May 2015, 1:01:24 UTC 0 GigaFLOPS Mac OS X/64-bit Intel 7.05 (opencl_ati5zc_mac) 6 May 2015, 0:51:12 UTC 28 GigaFLOPS Mac OS X/64-bit Intel 7.05 (opencl_ati5_mac) 23 Apr 2015, 23:04:48 UTC 55 GigaFLOPS Mac OS X/64-bit Intel 7.05 (opencl_ati_mac) 23 Apr 2015, 23:04:48 UTC 38 GigaFLOPS Mac OS X/64-bit Intel 7.05 (sse3) 21 May 2015, 1:01:24 UTC 0 GigaFLOPS Mac OS X/64-bit Intel 7.06 (opencl_intel_gpu_sah) 20 May 2015, 0:34:24 UTC 6 GigaFLOPS Mac OS X/64-bit Intel 7.06 (opencl_nvidia_mac) 20 May 2015, 0:34:24 UTC 64 GigaFLOPS http://setiweb.ssl.berkeley.edu/beta/apps.php All we need are Testers. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
It's always something. Seems there is another BOINC Bug hampering the rollout of the New Mac CPU Apps at Beta. So, after 4 days there are very few results to consider. In fact, there are basically only 2 Hosts to analyze. Since one is mine, let's consider the other; Computer 60533 Intel(R) Core(TM) i7-2620M CPU @ 2.70GHz SETI@home v7 7.00 i686-apple-darwin: Average processing rate = 22.01 GFLOPS SETI@home v7 7.05 x86_64-apple-darwin (sse3): Average processing rate = 35.59 GFLOPS SETI@home v7 7.05 x86_64-apple-darwin (avx): Average processing rate = 40.06 GFLOPS Well, just as with my Host, that is basically *Twice as Fastâ„¢* There was another host which ran a couple of the sse3s and they were twice as fast also. So, for now if you want to run the New *Twice as Fastâ„¢* Mac Apps at Beta you have to set the preferences to run CPU MB and then ABORT EVERY SETI@home v7 7.00 CPU task the Server sends you. After about 20 or 30 Abortions, the silly server will send you the new Apps. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Thanks for the info. Got it to send me a couple of the new apps after following your advice Zalster |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Not bad. It appears to have gone from; 4 hours 3 min 46 sec to 2 hours 16 min 54 sec on the angle ranges around 0.408657 Not quite *Twice as Fastâ„¢*, but close enough. Unfortunately you have to keep aborting the old v7.00 tasks until you reach 11 completions on the New versions. Once you reach 11, the server starts using the new APR and will automatically send the New/Faster tasks. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Good to know, just abort about a dozen until I got some of the new ones. Looks like there is a mix of sse3 and avx that I got Zalster |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
TBar are we looking at the AVX or the SSE3 MB? I stopped getting the SSE3 but have been crunching alot of AVX. Zalster |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
You're receiving the AVX version, the plan classes are working, All tasks for computer 72386 A couple of others are also receiving the AVX version; http://setiweb.ssl.berkeley.edu/beta/results.php?hostid=60533&offset=20 http://setiweb.ssl.berkeley.edu/beta/results.php?hostid=75255 A couple are receiving the SSE3 version; http://setiweb.ssl.berkeley.edu/beta/results.php?hostid=39415 http://setiweb.ssl.berkeley.edu/beta/results.php?hostid=59702 It's a shame that after Two weeks only a handful of people have been issued a new App that's about twice as fast as the old one. You'd think people would be excited about basically Doubling the number of Mac CPUs at SETI. That's about the same as being twice as fast, at the same speed it would equate to doubling the number of CPUs available. That's a lot of CPUs. I finally gave up and attacked the Linux NoCAL problem at Beta myself. By disconnecting the Cayman only the Juniper was seen and since the Juniper has CAL, the Beta server sent the New ATI MB App. It sent the plain Jane App, but, I was able to just add the 5 to ATI and download the ATI5 version. Since reconnecting the Cayman the Server once again is refusing to send Work. Unfortunately, the new MB r2910 doesn't seem any faster than the old r1844 I was using. Oh well, at least it didn't trash the work it sent after reconnecting the Cayman...and I do have the New App. |
©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.