Message boards :
Number crunching :
AstroPulse for Intel GPUs, open beta2
Message board moderation
Author | Message |
---|---|
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Here https://dl.dropboxusercontent.com/u/60381958/AP6_win_x86_SSE2_OpenCL_Intel_r2045.7z is new version of AstroPulse for Intel GPUs (CPUs with OpenCL-enabled GPU part). Improvements over previous version (http://setiathome.berkeley.edu/forum_thread.php?id=70717): 1) increased precision 2) no more 100% CPU usage, can be used with 100% CPU load (cause Intel GPUs not too fast such usage recommended). Please report found issues here. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Thanks a lot! I already had fun with version 1764 and a hit rate of more than 90% good. So i presume, this version will run fine! :) Aloha, Uli |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
So i presume, this version will run fine! :) Hope so :) Our inner "hard to pass" test WU was passed for this build so fingers crossed. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Just got the first WU, but see: 02-Nov-2013 11:15:00 [SETI@home] Started download of ap_21dc08ab_B0_P0_00124_20131102_07423.wu 02-Nov-2013 11:15:11 [SETI@home] Finished download of ap_21dc08ab_B0_P0_00124_20131102_07423.wu 02-Nov-2013 11:15:11 [SETI@home] Starting task ap_21dc08ab_B0_P0_00124_20131102_07423.wu_1 using astropulse_v6 version 604 (opencl_intel_100) in slot 3 02-Nov-2013 11:15:12 [SETI@home] Computation for task ap_21dc08ab_B0_P0_00124_20131102_07423.wu_1 finished 02-Nov-2013 11:15:12 [SETI@home] Output file ap_21dc08ab_B0_P0_00124_20131102_07423.wu_1_0 for task ap_21dc08ab_B0_P0_00124_20131102_07423.wu_1 absent :? Meanwhile it is up: http://setiathome.berkeley.edu/result.php?resultid=3222424019 It can't compile the binary: INFO: can't open binary kernel file: C:\ProgramData\BOINC/projects/setiathome.berkeley.edu\AstroPulse_Kernels_r2045.cl_IntelRHDGraphics4000.bin_V6, continue with recompile... Aloha, Uli |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Suspend new work fetch for now Loks like M_PI define missed and CL file fails to be built. Strange, will check posted binary again. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
And did this PC IntelGPU AP's before OK with very same driver ? #if USE_OPENCL_INTEL //R: Intel SDK includes <stdin> silently that contains Pi definition #else #define M_PI 3.14159265358979323846264338f #endif So, I intentionally disabled that definition in prev builds. Quite strange. EDIT: please try to locate posted part of code inside CL file (should be near line 79), remove all posted lines but #define M_PI 3.14159265358979323846264338f and try next task. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
And did this PC IntelGPU AP's before OK with very same driver ? The older build ran on Windows 8 pro 64 bit, but i stopped before the upgrade to Windows 8.1. So it is likely, that the driver changed meanwhile. The 1764 build ran out of the box on my machine. :? Aloha, Uli |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Ok, so now we have different Intel's approaches to Pi number definition, LoL :) Ok, try fix I proposed earlier and I will cook update for beta binaries. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Ok, so now we have different Intel's approaches to Pi number definition, LoL :) I made the changes to the *.cl-file according to your suggestion. Now let's hope, the server sends new AP-WUs to my laptop. ;) PS: Wow, i can program OpenCL... X) PS2: Meanwhile i got a new WU and seems to be running fine now. :) Aloha, Uli |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
:) Yeah, that's simple, most hard part is to start ;) I updated archive so hope next downloader will not meet this issue. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Yes, the #ifndef-approach seems a lot smarter, works both ways! :thumbsup: This version seems to be a little slower than version 1764, but maybe it is because of heavy blanking in the particular WU. Let's see, when it's done. Aloha, Uli |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Yes, by our offline tests it little slower but CPU load greatly reduced so it should compensate on fully committed CPU/GPU device. Anyway, one avoided inconclusive will overcome speed difference for many dozens if not hundreds of tasks. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Running since 2,5 hours, still going strong, but only at ~40% done. Version 1764 took around that time for a complete WU. Found a last one in my history: http://setiathome.berkeley.edu/result.php?resultid=3187634813 But i suspect, that one will be considered invalid. Aloha, Uli |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Meanwhile completed and resulted in a whoppin' 1044 cobblestones credit and a canonical result! :)) http://setiathome.berkeley.edu/workunit.php?wuid=1349795962 [edit] Blanking was around 7 percent. The next WU is running faster... ;) Aloha, Uli |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Hmm, looks like a one/two man show here... The new executable runs fine but i ran into another issue: The execution stops from time to time, when the computer blanks the screen after some time. When i reactivate the screen, it looks, as if everything is fine, but the GPU isn't crunching. The GPU-load is 0%. The only ways to get it back crunching is to restart BOINC or to suspend/resume the task with the intel GPU. Hmm, the old executable didn't had this issue... :? Aloha, Uli |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Most probably the old driver didn't have this issue. It's OS/driver level thing. To check just download older app and install it instead of current one. EDIT: I use Name: Intel(R) HD Graphics 2500 Vendor: Intel(R) Corporation Driver version: 9.18.10.3257 Version: OpenCL 1.2 And display switched off after 20 mins (now changed to after 5 mins). Crunching continues it seems. It crunches task in ~30k seconds withonly1,2k seconds of CPU time. CPU fully loaded with opt SETI v7 app (+2 CUDA tasks on GT9400 and GSO9600) SETI apps news We're not gonna fight them. We're gonna transcend them. |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Most probably the old driver didn't have this issue. It's OS/driver level thing. I did that, but thanks to BOINC trashing my whole supply of WUs for intel GPUs i'm now completely dry of WUs. Damn' f*ckin' sh*t! %(( It was the same error with missing declaration of M_PI. Why are they doin' this in driver programming? :/ Aloha, Uli |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
So, do same change (as I said they switched their approach with driver you use it seems). SETI apps news We're not gonna fight them. We're gonna transcend them. |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
So, do same change (as I said they switched their approach with driver you use it seems). Yes, i already did, but the WUs were already trashed and now i don't get new ones from the server... :/ Just look at this disaster: http://setiathome.berkeley.edu/results.php?hostid=7068094&offset=0&show_names=0&state=6&appid=12 Why can't i get them back... :..( Aloha, Uli |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
cause they already allocated to someone else. better to shutdown network and make backup BOINc data directory before testing. But sooner or later your host picks up some work anyway. SETI apps news We're not gonna fight them. We're gonna transcend them. |
©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.