| Author |
Message |
|---|
|
|
|
Hello!
I have installed the latest drivers for GTS 450 (version 266.58) but i'm getting error "Fermi class GPU not supported by cuda 2.2" & "reached limit of 2GPU tasks in progress". The point is that at the time being i can see only one WU processed by "ACEMD2 (cuda 3.1)". Are everything all right? Moreover is the total elapse time of 16 hours precise? It seems to me too much but you know better! |
|
|
 skgivenVolunteer moderator
Volunteer tester
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
Scientific publications
      |
|
The limit is 2 tasks per GPU.
16h is slow.
Look at your NVidia Power settings:
Right click on desktop, NVidia Control Panel, Manage 3D settings, Global Settings, Power Management Mode.
Make sure it is set to Prefer maximum performance
Good luck,
|
|
|
|
|
|
I inserted the settings you suggested but i didn't saw any significant decrease in elapse time at all; the numbers are still the same. Is it possible that by pointing out for my case scenario that 16H are too slow that you had a specific GTS 450 configuration (overclocked or something) in mind? Is there anything else that I can try as it comes to changing settings? But if the point gets to overclocking my VGA I have to tell you that i have no experience in this area on a VGA and would like to know what exactly would be safe to adjust and in what percentage (do you want to increase the memory frequency or GPU's ?). And what about the errors i'm getting throw BOINC's messages tab? |
|
|
skgivenVolunteer moderator
Volunteer tester
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
Scientific publications
|
|
Read this, FAQ: Best configurations for GPUGRID
I suggest the usual:
Use swan_sync=0, leave a CPU core free, keep a low cache and report tasks immediately.
|
|
|
|
|
|
Hello again!
I created the file "cc_config.xml" you suggested but i don't see any improvement in elapsed time tab; still total time is 16h. Can you explain why in your second message you said that 16h for 2 WUs is too slow? Is there something else to try? |
|
|
skgivenVolunteer moderator
Volunteer tester
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
Scientific publications
|
|
I did not quite say that. Perhaps something was lost in translation?
16h is slow for one task on a GTS450. One task is run at a time, not two. So downloading two tasks at a time means one task waits for 16h to run.
It is important to return tasks quickly, so making them wait to run slows them down.
Windows 7 x86 is slower than XP and Linux, so that explains part of the slowness.
IBUCH tasks are also slower than KASHIF_HIVPR tasks.
If you create the environmental variable swan_sync=0, and leave a CPU core free, this will increase task speed. Did you do this?
If you report tasks immediately this will increase your credit, well prevent late reporting and reduced credit to be more precise.
If you keep a low cache this will mean you download a second task just before you finish the first, which means the second task will start a few minutes or hours after downloading, and therefore finish sooner after download and therefore return faster on average. Better for the project and better for your credit.
While your KASHIF_HIVPR task will run faster than the IBUCH task, because it sat in the queue waiting to run you may not return it within 24h and therefore may not get full credit for it. If you implement my suggestions you will avoid this in the future. |
|
|
|
|
|
It has come to my attention that my GPU is processing simultaneous two WUs (you can see that in task board for my account) and this is why the elapsed time doesn't get smaller that 14h (the most). This is something that happens only to myself or it is a global setting in powerful GPUs 2 WUs to be processed? |
|
|
skgivenVolunteer moderator
Volunteer tester
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
Scientific publications
|
|
Reduce your cache to about 0.2days or less. I use 0.01 on most systems.
Do you report tasks immediately?
It might also be because you inadvisably edited your app_info.xml file to run two tasks simultaneously. Did you? If so, select no new tasks, finish any running tasks, delete the files contents and reboot. |
|
|
|
|
|
The only thing I did is to create the file cc_config.xml you wanted and asked for in section "FAQ: Best configurations for GPUGRID". I'm [osting underneath what exactly i pasted from that page into the file i created:
<cc_config>
<options>
<report_results_immediately>1</report_results_immediately>
</options>
</cc_config>
swan_sync=0
Also as for swan_sync, i tried the system variable you mentioned in the above section. I don't know how to set the cache size.. |
|
|
|
|
|
Now I am trying Linux (FatDog)and BOINC just to check if elapsed time will be different but it's exactly the same. Without having changed anything and using default preferences with all programs, elapsed time is @ 14h. In my previous message i didn't answer to one of your questions regarding the fact that my GPU is running two WUs at the same time; because i'm a beginner user in GPUGrid program, the only thing i changed in BOINC's folder is the creation of the file cc_config.xml. The other file I didn't created it! Sorry for the pressure or the inconvenience but I bought my GPU for this purpose only and I'm wondering if my choice was correct or if I have to sell this one and get another one VGA with better GPU.
Thanks for your time anyway! |
|
|
skgivenVolunteer moderator
Volunteer tester
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
Scientific publications
|
|
My GTX260 completes tasks in about 6h. So if setup correctly a GTS450, on Windows or Linux, the same task should take around 9h, not 14 to 16h:
My GTX260's clock rate is 1.35GHz and has 216 Cores, your GTS450 has 192 cores and a clock rate of 1.57GHz. While in theory the performance should be about the same, within 5%, we know the GTS450 only uses 2/3 of the cuda cores, but even at that it should have a performance of about 2/3 my stock GTX260. Of course this would require you to use swan_sync=0, free up one CPU core, report tasks immediately and keep a low cache.
To set a low cache,
open Boinc Manager, click Advanced, Preferences, network usage.
Set the Additional network buffer to 0.10 days, or similar.
Good luck,
|
|
|
|
|
|
Sorry for asking but you said in your previous message that "we know the GTS450 only uses 2/3 of the cuda cores" and I'm wondering if there is a way to change this. Do you have anything in mind (e.g. NVidia Bios Editor, overclocking or else)? I have done all the other settings you proposed (aka swan_sync, one CPU core is free, report of tasks is immediately and kept a low cache).
|
|
|
skgivenVolunteer moderator
Volunteer tester
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
Scientific publications
|
|
Unfortunately I dont know of a way and I think it might require a new version of ACEMD/major application changes. If it was easy it would have been done by now.
I see you are running two malaria tasks. Are you running both at the same time?
Have you set Boinc to only use one CPU core?
If not, Open Boinc Manager,
Advanced Preferences, Processor Usage Tab, select to use 99% of the processors (On multiprocessor systems use at most).
Good luck, |
|
|
|
|
|
I don't know how but this is true. I am running two tasks at the same time. I haven't modified anything; when I entered the program this started too (just like a default preference). I don't know how to change this. Is it possible to run only one task or does the program gets to decide by itself what i must run based on my GPU? I made your changes and have to tell you that in "Tasks" tab I can see that the tasks are using "0.07 CPUs and 1 NVidia GPUs". Bellow i will paste my messages tab (I don't know if they'll help):
26/1/2011 5:58:40 μμ Starting BOINC client version 6.10.58 for windows_intelx86
26/1/2011 5:58:40 μμ log flags: file_xfer, sched_ops, task
26/1/2011 5:58:40 μμ Libraries: libcurl/7.19.7 OpenSSL/0.9.8l zlib/1.2.3
26/1/2011 5:58:40 μμ Data directory: D:\ProgramData\BOINC
26/1/2011 5:58:40 μμ Running under account Stefanos
26/1/2011 5:58:40 μμ Processor: 2 GenuineIntel Intel(R) Core(TM)2 Duo CPU E7600 @ 3.06GHz [Family 6 Model 23 Stepping 10]
26/1/2011 5:58:40 μμ Processor: 3.00 MB cache
26/1/2011 5:58:40 μμ Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 sse4_1 lm vmx tm2 pbe
26/1/2011 5:58:40 μμ OS: Microsoft Windows 7: Ultimate x86 Edition, (06.01.7600.00)
26/1/2011 5:58:40 μμ Memory: 3.25 GB physical, 6.49 GB virtual
26/1/2011 5:58:40 μμ Disk: 268.56 GB total, 199.03 GB free
26/1/2011 5:58:40 μμ Local time is UTC +2 hours
26/1/2011 5:58:40 μμ NVIDIA GPU 0: GeForce GTS 450 (driver version 26658, CUDA version 3020, compute capability 2.1, 993MB, 461 GFLOPS peak)
26/1/2011 5:58:40 μμ GPUGRID URL http://www.gpugrid.net/; Computer ID 92114; resource share 100
26/1/2011 5:58:40 μμ GPUGRID General prefs: from GPUGRID (last modified 25-Jan-2011 19:30:44)
26/1/2011 5:58:40 μμ GPUGRID Host location: none
26/1/2011 5:58:40 μμ GPUGRID General prefs: using your defaults
26/1/2011 5:58:40 μμ Reading preferences override file
26/1/2011 5:58:40 μμ Preferences:
26/1/2011 5:58:40 μμ max memory usage when active: 665.46MB
26/1/2011 5:58:40 μμ max memory usage when idle: 2994.57MB
26/1/2011 5:58:40 μμ max disk usage: 100.00GB
26/1/2011 5:58:40 μμ suspend work if non-BOINC CPU load exceeds 85 %
26/1/2011 5:58:40 μμ (to change preferences, visit the web site of an attached project, or select Preferences in the Manager)
26/1/2011 5:58:40 μμ Not using a proxy
26/1/2011 5:58:40 μμ GPUGRID Restarting task 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1 using acemd2 version 613
26/1/2011 5:58:40 μμ GPUGRID Sending scheduler request: To fetch work.
26/1/2011 5:58:40 μμ GPUGRID Requesting new tasks for CPU
26/1/2011 5:58:41 μμ GPUGRID Scheduler request completed: got 0 new tasks
26/1/2011 5:58:41 μμ GPUGRID Message from server: No work sent
26/1/2011 5:58:41 μμ GPUGRID Message from server: Fermi-class GPU not supported by cuda2.2
26/1/2011 8:25:20 μμ GPUGRID Task 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1 exited with zero status but no 'finished' file
26/1/2011 8:25:20 μμ GPUGRID If this happens repeatedly you may need to reset the project.
26/1/2011 8:25:20 μμ GPUGRID Restarting task 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1 using acemd2 version 613
26/1/2011 9:07:33 μμ GPUGRID Sending scheduler request: To fetch work.
26/1/2011 9:07:33 μμ GPUGRID Requesting new tasks for GPU
26/1/2011 9:07:35 μμ GPUGRID Scheduler request completed: got 1 new tasks
26/1/2011 9:07:37 μμ GPUGRID Started download of p45-IBUCH_10_wtEGFR_110121-4-LICENSE
26/1/2011 9:07:37 μμ GPUGRID Started download of p45-IBUCH_10_wtEGFR_110121-4-COPYRIGHT
26/1/2011 9:07:39 μμ GPUGRID Finished download of p45-IBUCH_10_wtEGFR_110121-4-LICENSE
26/1/2011 9:07:39 μμ GPUGRID Finished download of p45-IBUCH_10_wtEGFR_110121-4-COPYRIGHT
26/1/2011 9:07:39 μμ GPUGRID Started download of p45-IBUCH_10_wtEGFR_110121-4-p45-IBUCH_10_wtEGFR_110121-3-20-RND4268_1
26/1/2011 9:07:39 μμ GPUGRID Started download of p45-IBUCH_10_wtEGFR_110121-4-p45-IBUCH_10_wtEGFR_110121-3-20-RND4268_2
26/1/2011 9:07:49 μμ GPUGRID Finished download of p45-IBUCH_10_wtEGFR_110121-4-p45-IBUCH_10_wtEGFR_110121-3-20-RND4268_1
26/1/2011 9:07:49 μμ GPUGRID Started download of p45-IBUCH_10_wtEGFR_110121-4-p45-IBUCH_10_wtEGFR_110121-3-20-RND4268_3
26/1/2011 9:07:51 μμ GPUGRID Finished download of p45-IBUCH_10_wtEGFR_110121-4-p45-IBUCH_10_wtEGFR_110121-3-20-RND4268_2
26/1/2011 9:07:51 μμ GPUGRID Started download of p45-IBUCH_10_wtEGFR_110121-4-pdb_file
26/1/2011 9:07:54 μμ GPUGRID Finished download of p45-IBUCH_10_wtEGFR_110121-4-p45-IBUCH_10_wtEGFR_110121-3-20-RND4268_3
26/1/2011 9:07:54 μμ GPUGRID Started download of p45-IBUCH_10_wtEGFR_110121-4-psf_file
26/1/2011 9:08:21 μμ GPUGRID Finished download of p45-IBUCH_10_wtEGFR_110121-4-pdb_file
26/1/2011 9:08:21 μμ GPUGRID Started download of p45-IBUCH_10_wtEGFR_110121-4-par_file
26/1/2011 9:08:23 μμ GPUGRID Finished download of p45-IBUCH_10_wtEGFR_110121-4-psf_file
26/1/2011 9:08:23 μμ GPUGRID Started download of p45-IBUCH_10_wtEGFR_110121-4-conf_file_enc
26/1/2011 9:08:24 μμ GPUGRID Finished download of p45-IBUCH_10_wtEGFR_110121-4-par_file
26/1/2011 9:08:24 μμ GPUGRID Finished download of p45-IBUCH_10_wtEGFR_110121-4-conf_file_enc
26/1/2011 9:08:24 μμ GPUGRID Started download of p45-IBUCH_10_wtEGFR_110121-4-metainp_file
26/1/2011 9:08:24 μμ GPUGRID Started download of p45-IBUCH_10_wtEGFR_110121-4-p45-IBUCH_10_wtEGFR_110121-3-20-RND4268_7
26/1/2011 9:08:25 μμ GPUGRID Finished download of p45-IBUCH_10_wtEGFR_110121-4-metainp_file
26/1/2011 9:08:25 μμ GPUGRID Finished download of p45-IBUCH_10_wtEGFR_110121-4-p45-IBUCH_10_wtEGFR_110121-3-20-RND4268_7
26/1/2011 10:23:54 μμ GPUGRID Task 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1 exited with zero status but no 'finished' file
26/1/2011 10:23:54 μμ GPUGRID If this happens repeatedly you may need to reset the project.
26/1/2011 10:23:54 μμ GPUGRID Restarting task 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1 using acemd2 version 613
26/1/2011 11:21:19 μμ GPUGRID Computation for task 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1 finished
26/1/2011 11:21:19 μμ GPUGRID Starting p45-IBUCH_10_wtEGFR_110121-4-20-RND4268_1
26/1/2011 11:21:19 μμ GPUGRID Starting task p45-IBUCH_10_wtEGFR_110121-4-20-RND4268_1 using acemd2 version 613
26/1/2011 11:21:20 μμ GPUGRID Started upload of 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1_0
26/1/2011 11:21:20 μμ GPUGRID Started upload of 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1_1
26/1/2011 11:21:24 μμ GPUGRID Finished upload of 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1_0
26/1/2011 11:21:24 μμ GPUGRID Started upload of 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1_2
26/1/2011 11:21:47 μμ GPUGRID Finished upload of 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1_1
26/1/2011 11:21:47 μμ GPUGRID Finished upload of 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1_2
26/1/2011 11:21:47 μμ GPUGRID Started upload of 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1_3
26/1/2011 11:21:47 μμ GPUGRID Started upload of 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1_4
26/1/2011 11:22:00 μμ GPUGRID Finished upload of 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1_3
26/1/2011 11:22:00 μμ GPUGRID Started upload of 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1_7
26/1/2011 11:22:01 μμ GPUGRID Finished upload of 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1_7
26/1/2011 11:27:03 μμ GPUGRID Finished upload of 205-KASHIF_HIVPR_n1_bound_so_ba2-73-100-RND0342_1_4
26/1/2011 11:48:32 μμ GPUGRID General prefs: from GPUGRID (last modified 25-Jan-2011 19:30:44)
26/1/2011 11:48:32 μμ GPUGRID Host location: none
26/1/2011 11:48:32 μμ GPUGRID General prefs: using your defaults
26/1/2011 11:48:32 μμ Reading preferences override file
26/1/2011 11:48:32 μμ Preferences:
26/1/2011 11:48:32 μμ max memory usage when active: 665.46MB
26/1/2011 11:48:32 μμ max memory usage when idle: 2994.57MB
26/1/2011 11:48:32 μμ max disk usage: 100.00GB
26/1/2011 11:48:32 μμ Number of usable CPUs has changed from 2 to 1. Running benchmarks.
26/1/2011 11:48:32 μμ max CPUs used: 1
26/1/2011 11:48:32 μμ suspend work if non-BOINC CPU load exceeds 85 %
26/1/2011 11:48:32 μμ (to change preferences, visit the web site of an attached project, or select Preferences in the Manager)
26/1/2011 11:48:32 μμ Running CPU benchmarks
26/1/2011 11:48:32 μμ Suspending computation - running CPU benchmarks
26/1/2011 11:49:03 μμ Benchmark results:
26/1/2011 11:49:03 μμ Number of CPUs: 1
26/1/2011 11:49:03 μμ 3229 floating point MIPS (Whetstone) per CPU
26/1/2011 11:49:03 μμ 6631 integer MIPS (Dhrystone) per CPU
26/1/2011 11:49:04 μμ Resuming computation
26/1/2011 11:58:39 μμ GPUGRID General prefs: from GPUGRID (last modified 25-Jan-2011 19:30:44)
26/1/2011 11:58:39 μμ GPUGRID Host location: none
26/1/2011 11:58:39 μμ GPUGRID General prefs: using your defaults
26/1/2011 11:58:39 μμ Reading preferences override file
26/1/2011 11:58:39 μμ Preferences:
26/1/2011 11:58:39 μμ max memory usage when active: 665.46MB
26/1/2011 11:58:39 μμ max memory usage when idle: 2994.57MB
26/1/2011 11:58:39 μμ max disk usage: 100.00GB
26/1/2011 11:58:39 μμ max CPUs used: 1
26/1/2011 11:58:39 μμ suspend work if non-BOINC CPU load exceeds 85 %
26/1/2011 11:58:39 μμ (to change preferences, visit the web site of an attached project, or select Preferences in the Manager) |
|
|
skgivenVolunteer moderator
Volunteer tester
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
Scientific publications
|
|
Hopefully we will see more positive improvement by tomorrow.
Your presently running IBUCH task might take longer than your last KASHIF task, but your next KASHIF task should be faster than the last one.
There is some improvement between these two KASHIF tasks, the second one being returned this morning:
23 Jan 2011 23:34:13 UTC 24 Jan 2011 18:06:07 UTC Completed and validated 48,674.16 3,090.81 8,011.54 12,017.31
26 Jan 2011 6:07:49 UTC 27 Jan 2011 0:07:34 UTC Completed and validated 45,330.75 44,675.10 8,011.54 12,017.31
I guess some changes you made happened during the second tasks run, but the improvement is about 7%.
|
|
|
|
|
|
The difference in the perfomance is a result from overclocking a "little" my GPU. But I have to say that i discovered that my GPU's low rate deliveries are cause by my motherboard's lack of a PCIe v 2.0 x 16 slot connection but v1.0(because it is ment for the GPU to be connected to v2.0 PCIe slot).
Take care ;)! |
|
|
/ 
Reply
Quote
