Message boards :
Number crunching :
Faster GPUs with Nvidia 310.70?
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5
Author | Message |
---|---|
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Now that I have been running the GTX660SC for over a month now using _x41zc cuda50 and nVidia 310.90, I'm anxious to try running AP tasks using opencl_nvidia_100. Could someone please provide me with an example for the app_info.xml file and any other suggestions that may be useful. Here is an app_info section for NV OpenCL AP, it is set up to reserve a core already: <app> <name>astropulse_v6</name> </app> <file_info> <name>AP6_win_x86_SSE2_OpenCL_NV_r1363.exe</name> <executable/> </file_info> <file_info> <name>ap_cmdline.txt</name> </file_info> <app_version> <app_name>astropulse_v6</app_name> <version_num>604</version_num> <platform>windows_intelx86</platform> <avg_ncpus>1.0</avg_ncpus> <max_ncpus>1.0</max_ncpus> <plan_class>cuda_opencl_100</plan_class> <coproc> <type>CUDA</type> <count>1</count> </coproc> <file_ref> <file_name>AP6_win_x86_SSE2_OpenCL_NV_r1363.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>ap_cmdline.txt</file_name> </file_ref> </app_version> <app_version> <app_name>astropulse_v6</app_name> <version_num>604</version_num> <platform>windows_intelx86</platform> <avg_ncpus>1.0</avg_ncpus> <max_ncpus>1.0</max_ncpus> <plan_class>cuda_fermi</plan_class> <coproc> <type>CUDA</type> <count>1</count> </coproc> <file_ref> <file_name>AP6_win_x86_SSE2_OpenCL_NV_r1363.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>ap_cmdline.txt</file_name> </file_ref> </app_version> Now make a text file called ap_cmdline.txt and put the following cmdline parameters in it (The advantage is you can change parameters on the fly, then suspend/resume GPU usage to make them take effect, rather than fully restarting Boinc): -unroll 10 -ffa_block 6144 -ffa_block_fetch 1536 The r1363 app is available in the Windows GPU section at Lunatics: AP6 r1363 for GPU Claggy |
Cliff Harding Send message Joined: 18 Aug 99 Posts: 1432 Credit: 110,967,840 RAC: 67 |
Many thanks Claggy!! Installed and ready for bear. Just waiting on new tasks to come down the pike. I don't buy computers, I build them!! |
ivan Send message Joined: 5 Mar 01 Posts: 783 Credit: 348,560,338 RAC: 223 |
1) Use the one that fits your processor, OS, and GPU. e.g. 64-bit, Win7, 200-series. Hmm, you're using geforce.com, not nvidia.com -- I wonder if that's the difference? (I usually get redirected to nvidia.co.uk, which is often much slower to download than from the US site. :-( ) |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
1) Use the one that fits your processor, OS, and GPU. e.g. 64-bit, Win7, 200-series. The url's for both 310.90 drivers are the same, i expect they just did two pages and split the supported GPUs across them, Claggy |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Many thanks Claggy!! Installed and ready for bear. Just waiting on new tasks to come down the pike. Strange that you haven't received any AP since you posted, while i've received three, Please make sure you added the app_info section before the </app_info> closing tag, and that you restarted Boinc afterwards, and that Boinc didn't complain about the app_info in any way. Claggy |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Many thanks Claggy!! Installed and ready for bear. Just waiting on new tasks to come down the pike. I was wondering how I could persuade BOINC to allow me to use the AP6_win_x86_SSE2_OpenCL_ATI_r1363.exe App from that same package. I managed to use the AP6_win_x86_SSE2_OpenCL_NV_r1363.exe App with both my machines' NV cards using BOINC 7.0.38, no problem. However, each time I tried to use the AP6_win_x86_SSE2_OpenCL_ATI_r1363.exe App BOINC Deleted the App without any Notices. I had left the 'stock' App in the app_info file below the r1363 entry as a 'backup' to try to keep from losing all my tasks in case the new entry didn't work. That works BTW, keep a backup below the entry you're trying to add and if the new entry is bad you keep your tasks. After about 6 tries, I removed my 'backup' to see if that was possibly causing BOINC to keep Vaporizing the AP6_win_x86_SSE2_OpenCL_ATI_r1363.exe App. The result was BOINC Vaporized the App and ALL my tasks since the backup was missing. Great. Still, nothing from BOINC about why it didn't want to use AP6_win_x86_SSE2_OpenCL_ATI_r1363.exe. I'm pretty sure I've seen other people using AP6_win_x86_SSE2_OpenCL_ATI_r1363.exe. I basically made a copy of my working astropulse_6.04_windows_intelx86__opencl_ati_100.exe app_info entry and replaced the needed lines with AP6_win_x86_SSE2_OpenCL_ATI_r1363.exe, and made a txt file I named AP6_win_x86_SSE2_OpenCL_ATI_r1363.txt. I wanted to keep my r1316 file unchanged. Again, BOINC never said there was anything wrong with the app_info file or the App, it just Vaporized the App, and at times, the txt file as well. After it also Vaporized ALL my tasks, I gave up and continued to use astropulse_6.04_windows_intelx86__opencl_ati_100.exe. Now, continuing with the Astropulse 'Computation Error' saga, I received another Error after 6 days using BOINC 7.0.38. I decided to try another version and went to BOINC 7.0.42. The very first full task after updating to 7.0.42 was an Error, and I've received 2 more since updating earlier today. I'd really like to try the AP6_win_x86_SSE2_OpenCL_ATI_r1363.exe App if someone can explain how to keep it from being Vaporized. I will keep my backup r1316 entry this time... |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
PM me your app_info, and i'll have a look at it, Claggy |
Cliff Harding Send message Joined: 18 Aug 99 Posts: 1432 Credit: 110,967,840 RAC: 67 |
Many thanks Claggy!! Installed and ready for bear. Just waiting on new tasks to come down the pike. I inserted everything at the end of the app_info.xml just before the last line which is the </app_info> line. There were no error msgs in the event log after restarting BOINC. The reason no NEW AP tasks have come down the pike as of yet is that I'm at the current machine max level. It will be a couple of days yet before I see any new ones, I still have 6 working Einstein tasks and 39 in the queue before I go back to pure SETI. I don't buy computers, I build them!! |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
PM me your app_info, and i'll have a look at it, That worked. I named the cmdline file ap_cmdline_ati_r1363.txt and now it is preforming normally. Now we sit back and see how BOINC 7.0.42 & AP6r1363 work on this machine. It won't be long, give it another 20 minutes. Thanks again. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
As I said, that didn't take long; Access Violation (0xc0000005) at address 0x0040A822 read attempt to address 0x003AA614 At least the address is different now. I could always try 7.0.44. Then back to 7.0.38 if that doesn't work. At least it was around a week between Errors with 7.0.38. I don't know, maybe it's just Windows being Windows. What's the old saying, Windows isn't done 'til Lotus won't run? I know XP knows it's running on a Mac... Or maybe 'cause the Mac has 6GB of Ram and XP32 only uses 2GB? It would really be nice if there were SETI GPU Apps in OSX. |
Cliff Harding Send message Joined: 18 Aug 99 Posts: 1432 Credit: 110,967,840 RAC: 67 |
As I said, that didn't take long; I'm do not mean to get anyone upset and if I do I aplogize in advance, but I have a couple of questions. 1) Have you tried to swap the cards out and have only one type in each machine? 2) Why do you have a type mix on 2 of your machines (AMD/ATI & nVidia)? Could it be that the OS has a problem with both driver types in place? It would seem to me that if you are going to have multiple devices on a machine, it would be a lot easier to have the same type of device. I don't buy computers, I build them!! |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
As I said, that didn't take long; There is an early list of steps taken here, http://setiathome.berkeley.edu/forum_thread.php?id=69924&postid=1303044#1303044 Pay close attention to the thread here; 2 AstroPulse v6 v6.04 (opencl_nvidia_100) tasks crashed with "Access Violation (0xc0000005)" There are Many machines getting this Error, some don't even have a 'Card', just an APU. My MacPro with ECC Ram appears to be more susceptible to this particular Error, it is certainly not the only machine. It would be nice to find the problem though. |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Try reducing the cmdline parameters to -unroll 10 -ffa_block 6144 -ffa_block_fetch 1536 from the -unroll 12 -ffa_block 6144 -ffa_block_fetch 3072 that you are running at the moment, Claggy |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
The Errors went away after updating to 7.0.44. However, there were 11 hours in between the first 2 Errors after updating to 7.0.42. We are approaching 11 hours since updating. If I start getting Errors, I'll change some settings before going backwards to .38. I'd rather not go backwards. |
Cliff Harding Send message Joined: 18 Aug 99 Posts: 1432 Credit: 110,967,840 RAC: 67 |
1st 2 opencl_nvidia_100 completed & awaiting for validation the i7/950. I don't buy computers, I build 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.