Questions and Answers :
GPU applications :
problem running SETI on CUDA enabled IGP
Message board moderation
Author | Message |
---|---|
SolRad Send message Joined: 28 Nov 99 Posts: 4 Credit: 3,761,222 RAC: 1 |
I have just installed Seti@home on an Asus Nvidia nForce 750a (Athlon64X2+8GB) with the only video the IGP sharing 512MB of main memory. The IGP has CUDA 1.1 capability. I expected to see SETI running on it but it claims "insufficient memory". I have other system which runs CUDA optimized SETI on an 8600GTS with 256MB with no problems. I guess the problem resides in the "reserved" or "shared" nature of the (not really true) "local memory" but I really appreciate an illuminating answer. Please don't ask me why I am so silly to run SETI on CUDA IGP. Is just one of my few experiments with GPGPU (single/multi, IGP or not, ATi/nVidia). For example it would be interesting to have a developing platform (say a laptop with IGP only) to compile and test applications beforehand. http://www.boincstats.com/signature/user_209207.gif |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
Minimum memory needs to run CUDA on Seti is 255MB, your IGP shows up with 253MB of memory. Do know that it is already using up memory to show whatever is visible on screen. You may want to fiddle with screen resolution and bit depth, to see if your memory free on the IGP goes down. |
SolRad Send message Joined: 28 Nov 99 Posts: 4 Credit: 3,761,222 RAC: 1 |
Thanks a lot. I was aware that CUDA apps run in the memory left after substracting the display needs. I am also confused by the disagreement between nVidia 750a SLI with 256MB frame buffer reports [WinXP64 device manager (512MB!), GPUz (256MB), and CUDAz (253.688MB!)] compared with 8600GTS [WinXP32 device manager (256MB), GPUz (256MB), and CUDAz (255.688MB)]. Note the 2MB missing from IGP compared with pure 8600 (253 to 255). I thought to be due to some "paging" from a "pinned" 256MB to another 256MB "paged" (as WinXP64 reports 512MB even if I forced in BIOS 256MB frame buffer and disabled HybridSLI - there is no other video card anyway}. After fiddling some time with the BIOS IT FINALLY WORKED with frame buffer set on 512MB. NOW SETI CUDA ACCEPTS TO WORK ON THE IGP. But the 2MB are still missing (CUDAz reports 509.688MB) and as they created my problem I would be happy to know why. Thanks again for pointing me that there must be a solution. |
Emspilot Send message Joined: 18 May 99 Posts: 3 Credit: 6,028,377 RAC: 0 |
Does the GPU memory have to be 256MB dedicated, or can it be shared? I have a T61 notebook with a 128MG video card, sharing above that with the main memory. So, if I need 256 dedicated then I am short, but I can share up to far more than 256. Will CUDA work with shared memory? |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
Shared memory will only slow down the performance, but memory is memory and it will work. However, 256MB is just above the minimum and may cause problems for your everyday work. |
Emspilot Send message Joined: 18 May 99 Posts: 3 Credit: 6,028,377 RAC: 0 |
It does not seem to work for me. I'm running Vista x64 with 4GB of system memory and another 4GB in ReadyBoost. The video card has 128MB and then shares with the system above that. BOINC tells me I can't run CUDA work due to insufficient memory.... |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
I missed that part of 128MB onboard and 256MB shared. The onboard is the only part seen, the shared is only used as a last resort by some applications, but not by SETI-CUDA. |
Emspilot Send message Joined: 18 May 99 Posts: 3 Credit: 6,028,377 RAC: 0 |
(((sigh))) That's pretty much what I figured.... |
Bob Send message Joined: 5 Aug 01 Posts: 7 Credit: 7,241,828 RAC: 0 |
Running XP Home, video res 1680x1050, 16 bits Rest of info follows: Every CUDA work unit fails after about 12 seconds to 1+ minute (computation error). Tried removing ultraVNC, as that was listed as an additional video driver, but had no effect. Bought the 8400GS to increase Seti throughput on older system. Can not get any CUDA work completed. From Seti@home description of my system (seems fairly complete) Created 21 May 2005 4:23:09 UTC Total credit 97,936 Average credit 60.44 Cross project credit CPU type AuthenticAMD AMD Athlon(tm) Processor [x86 Family 6 Model 4 Stepping 2] Number of processors 1 Coprocessors NVIDIA GeForce 8400 GS (511MB) driver: 18250 Operating System Microsoft Windows XP Home x86 Edition, Service Pack 3, (05.01.2600.00) Memory 895.48 MB Cache 976.56 KB Swap space 2874.37 MB Total disk space 37.25 GB Free Disk Space 13.95 GB Measured floating point speed 1138.15 million ops/sec Measured integer speed 1940.78 million ops/sec Average upload rate 20.33 KB/sec Average download rate 213.85 KB/sec Average turnaround time 4.63 days Maximum daily WU quota per CPU 94/day Number of times client has contacted server 18162 Last time contacted server 18 Jul 2009 22:44:20 UTC % of time BOINC client is running 98.6616 % While BOINC running, % of time host has an Internet connection 99.9998 % While BOINC running, % of time work is allowed 99.8729 % Average CPU efficiency 0.919315 Task duration correction factor 1.042909 Log section showing part of startup and work units ending after just a few seconds. Log does not show the error, it only displays in the 'Tasks' discplay. 7/18/2009 8:06:41 PM Processor: 1 AuthenticAMD AMD Athlon(tm) Processor [x86 Family 6 Model 4 Stepping 2] 7/18/2009 8:06:41 PM Processor features: fpu tsc 3dnow mmx 7/18/2009 8:06:41 PM OS: Microsoft Windows XP: Home x86 Edition, Service Pack 3, (05.01.2600.00) 7/18/2009 8:06:41 PM Memory: 895.48 MB physical, 2.81 GB virtual 7/18/2009 8:06:41 PM Disk: 37.25 GB total, 13.95 GB free 7/18/2009 8:06:41 PM Local time is UTC -4 hours 7/18/2009 8:06:42 PM CUDA device: GeForce 8400 GS (driver version 18250, compute capability 1.1, 512MB, est. 4GFLOPS) 7/18/2009 8:06:42 PM Not using a proxy 7/18/2009 8:06:42 PM SETI@home URL: http://setiathome.berkeley.edu/; Computer ID: 921113; location: school; project prefs: school 7/18/2009 8:06:42 PM SETI@home General prefs: from SETI@home (last modified 31-Jan-2009 21:03:53) 7/18/2009 8:06:42 PM SETI@home Computer location: school 7/18/2009 8:06:42 PM General prefs: using separate prefs for school 7/18/2009 8:06:42 PM Reading preferences override file 7/18/2009 8:06:42 PM Preferences limit memory usage when active to 895.48MB 7/18/2009 8:06:42 PM Preferences limit memory usage when idle to 895.48MB 7/18/2009 8:06:42 PM Preferences limit disk usage to 14.00GB 7/18/2009 8:20:59 PM SETI@home Starting 31oc08ab.3244.11115.4.8.141_0 7/18/2009 8:20:59 PM SETI@home Starting task 31oc08ab.3244.11115.4.8.141_0 using setiathome_enhanced version 608 7/18/2009 8:21:09 PM SETI@home Sending scheduler request: To fetch work. 7/18/2009 8:21:09 PM SETI@home Requesting new tasks 7/18/2009 8:21:14 PM SETI@home Scheduler request completed: got 0 new tasks 7/18/2009 8:21:14 PM SETI@home Message from server: No work sent 7/18/2009 8:21:14 PM SETI@home Message from server: CUDA (GPU) jobs are available, but your preferences are set to not accept them 7/18/2009 8:21:32 PM SETI@home Computation for task 31oc08ab.3244.11115.4.8.141_0 finished 7/18/2009 8:21:32 PM SETI@home Starting 31oc08ab.3244.11115.4.8.156_0 7/18/2009 8:21:33 PM SETI@home Starting task 31oc08ab.3244.11115.4.8.156_0 using setiathome_enhanced version 608 7/18/2009 8:21:35 PM SETI@home Started upload of 31oc08ab.3244.11115.4.8.141_0_0 7/18/2009 8:21:40 PM Project communication failed: attempting access to reference site 7/18/2009 8:21:40 PM SETI@home Temporarily failed upload of 31oc08ab.3244.11115.4.8.141_0_0: HTTP error 7/18/2009 8:21:40 PM SETI@home Backing off 1 min 0 sec on upload of 31oc08ab.3244.11115.4.8.141_0_0 7/18/2009 8:21:41 PM Internet access OK - project servers may be temporarily down. 7/18/2009 8:22:01 PM SETI@home Computation for task 31oc08ab.3244.11115.4.8.156_0 finished 7/18/2009 8:22:01 PM SETI@home Starting 31oc08ab.3244.11115.4.8.147_0 7/18/2009 8:22:01 PM SETI@home Starting task 31oc08ab.3244.11115.4.8.147_0 using setiathome_enhanced version 608 7/18/2009 8:22:04 PM SETI@home Started upload of 31oc08ab.3244.11115.4.8.156_0_0 7/18/2009 8:22:05 PM Project communication failed: attempting access to reference site 7/18/2009 8:22:05 PM SETI@home Temporarily failed upload of 31oc08ab.3244.11115.4.8.156_0_0: HTTP error 7/18/2009 8:22:05 PM SETI@home Backing off 1 min 0 sec on upload of 31oc08ab.3244.11115.4.8.156_0_0 7/18/2009 8:22:06 PM Internet access OK - project servers may be temporarily down. 7/18/2009 8:22:17 PM SETI@home General prefs: from SETI@home (last modified 31-Jan-2009 21:03:53) 7/18/2009 8:22:17 PM SETI@home Computer location: school 7/18/2009 8:22:17 PM General prefs: using separate prefs for school 7/18/2009 8:22:17 PM Reading preferences override file 7/18/2009 8:22:17 PM Preferences limit memory usage when active to 895.48MB 7/18/2009 8:22:17 PM Preferences limit memory usage when idle to 895.48MB 7/18/2009 8:22:17 PM Preferences limit disk usage to 14.00GB 7/18/2009 8:22:32 PM SETI@home Computation for task 31oc08ab.3244.11115.4.8.147_0 finished 7/18/2009 8:22:34 PM SETI@home Started upload of 31oc08ab.3244.11115.4.8.147_0_0 Any suggestions? |
CElliott Send message Joined: 19 Jul 99 Posts: 178 Credit: 79,285,961 RAC: 0 |
You need at least the 185.85 Nvidia driver; 186.18 is the latest. |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
I don't see any computation errors. I do see upload problems, but we all have those. 7/18/2009 8:20:59 PM SETI@home Starting 31oc08ab.3244.11115.4.8.141_0 (stress mine) They show to finish normally. The HTTP error is an upload error, not a computation error. The project is still clogged up when it comes to uploading work. |
Bob Send message Joined: 5 Aug 01 Posts: 7 Credit: 7,241,828 RAC: 0 |
I understand about communication errors. There was not anything put into the log about a computation error. It only showed up in the 'Tasks' display. After less than two minutes, the task completed, showed 100%, but in the 'Status' column, it said computation error. They then uploaded. |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
I understand about communication errors. There was not anything put into the log about a computation error. It only showed up in the 'Tasks' display. After less than two minutes, the task completed, showed 100%, but in the 'Status' column, it said computation error. They then uploaded. Is it possible that they were "noisy" workunits that often terminate with errors? If so, this is perfectly normal to see. |
Bob Send message Joined: 5 Aug 01 Posts: 7 Credit: 7,241,828 RAC: 0 |
Uploaded and installed the latest driver from nVidia, 186.18. I'm not sure how to upload a .gif of the BOINC program showing the 'Tasks' window, with the CUDA tasks which ended quickly in the log below showing a status of "Computation error", but I do have it if anyone want to see it, I can e-mail it (343 KB). The tasks that errored out are named: 29au08ad.4618.1906878.14.8.254_0 14dc08aa.5464.203577.12.8.214_0 14dc08aa.5464.203577.12.8.215_0 31oc08ab.10240.9479.10.8.60_1 Basically, there was no change: 7/19/2009 11:42:13 AM Starting BOINC client version 6.6.28 for windows_intelx86 7/19/2009 11:42:13 AM log flags: task, file_xfer, sched_ops 7/19/2009 11:42:13 AM Libraries: libcurl/7.19.4 OpenSSL/0.9.8j zlib/1.2.3 7/19/2009 11:42:13 AM Data directory: T:\Documents and Settings\All Users\Application Data\BOINC 7/19/2009 11:42:13 AM Running under account Robert Hill 7/19/2009 11:42:14 AM Processor: 1 AuthenticAMD AMD Athlon(tm) Processor [x86 Family 6 Model 4 Stepping 2] 7/19/2009 11:42:14 AM Processor features: fpu tsc 3dnow mmx 7/19/2009 11:42:14 AM OS: Microsoft Windows XP: Home x86 Edition, Service Pack 3, (05.01.2600.00) 7/19/2009 11:42:14 AM Memory: 895.48 MB physical, 2.81 GB virtual 7/19/2009 11:42:14 AM Disk: 37.25 GB total, 13.57 GB free 7/19/2009 11:42:14 AM Local time is UTC -4 hours 7/19/2009 11:42:16 AM CUDA device: GeForce 8400 GS (driver version 18618, compute capability 1.1, 512MB, est. 4GFLOPS) 7/19/2009 11:42:16 AM Not using a proxy ... 7/19/2009 11:54:26 AM SETI@home Starting 29au08ad.4618.1906878.14.8.254_0 7/19/2009 11:54:27 AM SETI@home Starting task 29au08ad.4618.1906878.14.8.254_0 using setiathome_enhanced version 608 7/19/2009 11:54:29 AM SETI@home Finished download of 14dc08aa.5464.203577.12.8.215 7/19/2009 11:54:29 AM SETI@home Finished download of 31oc08ab.10240.9479.10.8.60 7/19/2009 11:54:29 AM SETI@home Started download of 29au08ad.4618.1906878.14.8.252 7/19/2009 11:54:29 AM SETI@home Started download of 29au08ad.4618.1906878.14.8.255 ... 7/19/2009 11:55:02 AM SETI@home Started download of 31oc08ab.10240.9479.10.8.231 7/19/2009 11:55:05 AM SETI@home Finished download of 31oc08ab.10240.9479.10.8.233 7/19/2009 11:55:05 AM SETI@home Finished download of 31oc08ab.10240.9479.10.8.231 7/19/2009 11:55:05 AM SETI@home Started download of 31oc08ab.10240.9479.10.8.151 7/19/2009 11:55:05 AM SETI@home Started download of 31oc08ab.10240.9479.10.8.167 7/19/2009 11:55:05 AM SETI@home Computation for task 29au08ad.4618.1906878.14.8.254_0 finished 7/19/2009 11:55:05 AM SETI@home Starting 14dc08aa.5464.203577.12.8.214_0 7/19/2009 11:55:05 AM SETI@home Starting task 14dc08aa.5464.203577.12.8.214_0 using setiathome_enhanced version 608 7/19/2009 11:55:08 AM SETI@home Finished download of 31oc08ab.10240.9479.10.8.151 ... 7/19/2009 11:55:14 AM SETI@home Finished download of 29au08ad.4618.1906878.14.8.253 7/19/2009 11:55:14 AM SETI@home Finished download of 14dc08aa.5464.203577.12.8.255 7/19/2009 11:55:36 AM SETI@home Computation for task 14dc08aa.5464.203577.12.8.214_0 finished 7/19/2009 11:55:36 AM SETI@home Starting 14dc08aa.5464.203577.12.8.215_0 7/19/2009 11:55:36 AM SETI@home Starting task 14dc08aa.5464.203577.12.8.215_0 using setiathome_enhanced version 608 7/19/2009 11:55:39 AM SETI@home Started upload of 14dc08aa.5464.203577.12.8.214_0_0 ... 7/19/2009 11:56:09 AM SETI@home Started upload of 29au08ad.4618.1906878.14.8.254_0_0 7/19/2009 11:56:09 AM SETI@home Computation for task 14dc08aa.5464.203577.12.8.215_0 finished 7/19/2009 11:56:09 AM SETI@home Starting 31oc08ab.10240.9479.10.8.60_1 7/19/2009 11:56:09 AM SETI@home Starting task 31oc08ab.10240.9479.10.8.60_1 using setiathome_enhanced version 608 |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
Since the replica database has been turned off, I cannot research those tasks for you to see if they are "normal" errors. |
Bob Send message Joined: 5 Aug 01 Posts: 7 Credit: 7,241,828 RAC: 0 |
I very much doubt that they are normal errors, because I have had no completed CUDA tasks. But I hope that you can take a look soon, and see. I've been in programming for about 32 years (IBM mainframe), built a couple of systems that I'm running Seti on from scratch, but I can't figure out what the heck is going. Probably some conflict with something or other that I'm running/loading at startup, but nothing seems out of the ordinary to me. Want a list of starup stuff, until they get the duplicate database going again? |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
It certainly can't hurt. The startup messages in BOINC Manager just might provide some clues. |
Bob Send message Joined: 5 Aug 01 Posts: 7 Credit: 7,241,828 RAC: 0 |
I gave up on that one old system, and moved the nVidia board to another system. It works fine there. Might be some old driver on the older system, non-compatable hardware or software, but at this point, I'm not going to worry about it. Thanks everyone for the help and suggestions. |
Bob Send message Joined: 5 Aug 01 Posts: 7 Credit: 7,241,828 RAC: 0 |
Well, I have two CUDA-enabled systems (running XP Prof SP3). Now, I'm trying a Linux system. System is described below in the startup message log. I bought a GeForce 9800 GTX+, 512MB GDDR3 memory mainly to run BOINC. Everything seems to be set up right, latest recommended Linux BOINC software (64 bit ver 6.4.5), latest nVidia driver for Linux from their site(NVIDIA-Linux-x86_64-185.18.31-pkg2.run). Boinc sees the nVidia board when it starts up. But no downloads of CUDA work units. Profile (home) is the same profile used by the XP systems which are working fine with CUDA work units being downloaded and processed. Here's the message log from starting BOINC. Any suggestions of what to try? Tue 04 Aug 2009 08:24:22 PM EDT||Starting BOINC client version 6.4.5 for x86_64-pc-linux-gnu Tue 04 Aug 2009 08:24:22 PM EDT||log flags: task, file_xfer, sched_ops Tue 04 Aug 2009 08:24:22 PM EDT||Libraries: libcurl/7.18.0 OpenSSL/0.9.8g zlib/1.2.3 c-ares/1.5.1 Tue 04 Aug 2009 08:24:22 PM EDT||Data directory: /home/rhill/BOINC Tue 04 Aug 2009 08:24:22 PM EDT||Processor: 2 AuthenticAMD AMD Athlon(tm) 64 X2 Dual Core Processor 6000+ [Family 15 Model 67 Stepping 3] Tue 04 Aug 2009 08:24:22 PM EDT||Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt rdtscp lm 3dnowext 3dnow rep_good nopl pni cx16 lahf_lm cmp_legacy svm extapic cr8_legacy Tue 04 Aug 2009 08:24:22 PM EDT||OS: Linux: 2.6.27.25-78.2.56.fc9.x86_64 Tue 04 Aug 2009 08:24:22 PM EDT||Memory: 1.96 GB physical, 1.94 GB virtual Tue 04 Aug 2009 08:24:22 PM EDT||Disk: 98.43 GB total, 77.88 GB free Tue 04 Aug 2009 08:24:22 PM EDT||Local time is UTC -4 hours Tue 04 Aug 2009 08:24:22 PM EDT||Not using a proxy Tue 04 Aug 2009 08:24:22 PM EDT||CUDA devices found Tue 04 Aug 2009 08:24:22 PM EDT||Coprocessor: GeForce 9800 GTX+ (1) Tue 04 Aug 2009 08:24:23 PM EDT|SETI@home|URL: http://setiathome.berkeley.edu/; Computer ID: 4637317; location: home; project prefs: home Tue 04 Aug 2009 08:24:23 PM EDT||General prefs: from SETI@home (last modified 01-Aug-2009 00:21:19) Tue 04 Aug 2009 08:24:23 PM EDT||Computer location: home Tue 04 Aug 2009 08:24:23 PM EDT||General prefs: using separate prefs for home Tue 04 Aug 2009 08:24:23 PM EDT||Reading preferences override file Tue 04 Aug 2009 08:24:23 PM EDT||Preferences limit memory usage when active to 1005.87MB Tue 04 Aug 2009 08:24:23 PM EDT||Preferences limit memory usage when idle to 1810.56MB Tue 04 Aug 2009 08:24:23 PM EDT||Preferences limit disk usage to 49.22GB Tue 04 Aug 2009 08:24:23 PM EDT|SETI@home|Restarting task 15oc08ac.23750.6616.4.10.201_1 using setiathome_enhanced version 528 Tue 04 Aug 2009 08:24:23 PM EDT|SETI@home|Starting 15mr09ac.2432.273563.6.10.165_0 Tue 04 Aug 2009 08:24:23 PM EDT|SETI@home|Starting task 15mr09ac.2432.273563.6.10.165_0 using setiathome_enhanced version 528 Tue 04 Aug 2009 08:24:23 PM EDT|SETI@home|Sending scheduler request: To report completed tasks. Requesting 0 seconds of work, reporting 3 completed tasks Tue 04 Aug 2009 08:26:01 PM EDT||Project communication failed: attempting access to reference site Tue 04 Aug 2009 08:26:02 PM EDT||Internet access OK - project servers may be temporarily down. Tue 04 Aug 2009 08:26:03 PM EDT|SETI@home|Scheduler request failed: Failed sending data to the peer Tue 04 Aug 2009 08:27:03 PM EDT|SETI@home|Sending scheduler request: To report completed tasks. Requesting 0 seconds of work, reporting 3 completed tasks Tue 04 Aug 2009 08:27:15 PM EDT||Project communication failed: attempting access to reference site Tue 04 Aug 2009 08:27:16 PM EDT|SETI@home|Scheduler request failed: Server returned nothing (no headers, no data) Tue 04 Aug 2009 08:27:17 PM EDT||Internet access OK - project servers may be temporarily down. Tue 04 Aug 2009 08:28:16 PM EDT|SETI@home|Sending scheduler request: To report completed tasks. Requesting 0 seconds of work, reporting 3 completed tasks Tue 04 Aug 2009 08:28:51 PM EDT|SETI@home|Scheduler request completed: got 0 new tasks |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
Well, I have two CUDA-enabled systems (running XP Prof SP3). Now, I'm trying a Linux system. ... Boinc sees the nVidia board when it starts up. But no downloads of CUDA work units. ... Any suggestions of what to try? There is no official CUDA Linux client yet. If you want to run CUDA under Linux, you must use a third party app to do so. |
©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.