GTX780 Ti system

Message boards : Number crunching : GTX780 Ti system
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · Next

AuthorMessage
Ghia
Avatar

Send message
Joined: 7 Feb 17
Posts: 238
Credit: 28,911,438
RAC: 50
Norway
Message 1847849 - Posted: 10 Feb 2017, 19:44:42 UTC

Hi everyone....first let me say : I'm a total newbie, running Set@Home only since Feb.7. So please bear with me.. ;-)

My rig has a i7-4770K CPU on a MSI mb, with 16GB G-Skill 2133MHz memory, running Win7Pro x64.
I am very unsure of how to best configure everything. I have set my preferences to USE CPU+NVIDIA GPU, and to prefer AstroPulse. As it is, 7 tasks are running...6x CPU +1x cuda50. GPU Core cLock is at 1163 and memory clock 1750....memory used 655MB. Temps are fine, both GPU and CPU at 60C.

Running 7 tasks, is it "normal" to use nearly 2 hrs on a WU ? I haven't touched any config files...I wouldn't know where to begin and what to set... *sigh*

Also, I have 7 error tasks, they're all (opencl_nvidia_SoG)windows_intelx86. No tasks of this sort have been completed, and the're labelled "Error while computing". Could there be an error in my installation of BOINC ? Un-install and re-install ?
ID: 1847849 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1847874 - Posted: 10 Feb 2017, 20:59:02 UTC - in response to Message 1847849.  

Check your antivirus and firewall to see if it will allow the application OpenCl....SOG to run. Looks like it maybe stopped by them
ID: 1847874 · Report as offensive
Ghia
Avatar

Send message
Joined: 7 Feb 17
Posts: 238
Credit: 28,911,438
RAC: 50
Norway
Message 1847967 - Posted: 11 Feb 2017, 4:36:36 UTC

I looked at my Webroot SecureAnywhere logs, and found that the OpenCL...SoG application had been flagged as malicious. It claims to have removed the file, but it is still there. Filesize is 1004KB. Whitelisting applications seems to require opening a Webroot support task...arghh
Humans may rule the world...but bacteria run it...
ID: 1847967 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1847973 - Posted: 11 Feb 2017, 4:49:41 UTC - in response to Message 1847967.  

I looked at my Webroot SecureAnywhere logs, and found that the OpenCL...SoG application had been flagged as malicious. It claims to have removed the file, but it is still there. Filesize is 1004KB. Whitelisting applications seems to require opening a Webroot support task...arghh

It should be possible to exclude files or complete folders using the applications settings.
Grant
Darwin NT
ID: 1847973 · Report as offensive
Ghia
Avatar

Send message
Joined: 7 Feb 17
Posts: 238
Credit: 28,911,438
RAC: 50
Norway
Message 1847997 - Posted: 11 Feb 2017, 8:32:58 UTC - in response to Message 1847973.  

Was able to unblock the application by simply starting it and then taking it off the blacklist. So simple...tried everything else first.. ;-) Tnx !

Deadline for the OpenCL...SoG WU's has come and gone. Is it ok to purge them ?
Humans may rule the world...but bacteria run it...
ID: 1847997 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1848004 - Posted: 11 Feb 2017, 8:56:26 UTC - in response to Message 1847997.  

Deadline for the OpenCL...SoG WU's has come and gone. Is it ok to purge them ?

The deadlines are usually 1-2 months. If they've passed, the Manager will clear them on the next update.
Grant
Darwin NT
ID: 1848004 · Report as offensive
Ghia
Avatar

Send message
Joined: 7 Feb 17
Posts: 238
Credit: 28,911,438
RAC: 50
Norway
Message 1848027 - Posted: 11 Feb 2017, 13:21:20 UTC - in response to Message 1848004.  

The date here isn't deadline ?

[/img]
Humans may rule the world...but bacteria run it...
ID: 1848027 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22158
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1848030 - Posted: 11 Feb 2017, 13:51:37 UTC

In the case of your screenshot the date is the date the task was reported on. The five tasks you show all ended in errors, which occurred while performing the calculations. A quick look at the output from these tasks shows that you had a temporary issue with writing to the data directory, which has now been resolved.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1848030 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1848147 - Posted: 11 Feb 2017, 23:57:04 UTC - in response to Message 1848027.  

The date here isn't deadline ?

[/img]


. . Hi Ghia,

. . When the tasks which ended in errors on your computer are reported (when it uploads your results) they are reassigned to another host to complete. All tasks are issued to two hosts at once for comparison and validation. So when these tasks are completed by the third host they will be compared to the your original co-host (popularly called a wingman) and if validated will clear from the system and your error queue. Till then there is nothing you need to do but wait. If for some reason they fail to validate or error out again they will be issued to a fourth host and again to a fifth host if necessary to achieve a valid result. Any task which fails five attempts will be flagged as bad and removed from the system. So eventually they will disappear from your errors page.

. . Only the tasks displayed on your "In Progress" window are waiting to be processed or are being processed. The other windows display the current status of those tasks which you have already completed in one way or another but which have not yet been cleared from the SETI database.

. . I hope that helps make it clearer.

Stephen

:)
ID: 1848147 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1848159 - Posted: 12 Feb 2017, 0:38:31 UTC - in response to Message 1848027.  
Last modified: 12 Feb 2017, 0:39:42 UTC

Since you errored out the first series of OpenCl....SOG sent to you, the server has decided not to send you anymore at the moment. Looking at your in progress you have a series of Cuda 50 and Cuda 42 that your system is working thru. At some point, the server may send you new SOG work units to try but we won't know if you actually corrected the problem until that happens. Till then, just continue to crunch what you have. Of note. SOG work units by default use a 1 full CPU core for each work unit. If you have any CPU work units when a SOG starts and you don't have any free CPU cores, the GPU will get preference for that core and you may see the CPU work units paused. We also recommend that you not over task your computer when running SOG task as it can cause your system to become sluggish or even unresponsive. Just for your information if you decide to run more than 1 work units per GPU.
ID: 1848159 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1848169 - Posted: 12 Feb 2017, 1:32:50 UTC - in response to Message 1848159.  

SOG work units by default use a 1 full CPU core for each work unit. If you have any CPU work units when a SOG starts and you don't have any free CPU cores, the GPU will get preference for that core and you may see the CPU work units paused. We also recommend that you not over task your computer when running SOG task as it can cause your system to become sluggish or even unresponsive.

However the big advantage of SoG on more recent hardware, is it's much faster than the older CUDA applications. So you get more work done over the same time frame.
Grant
Darwin NT
ID: 1848169 · Report as offensive
Ghia
Avatar

Send message
Joined: 7 Feb 17
Posts: 238
Credit: 28,911,438
RAC: 50
Norway
Message 1848173 - Posted: 12 Feb 2017, 1:40:15 UTC

Thanks to everyone for your patience and help....useful info there, and things are getting clearer. Been working with computers since the Atari 800 (and built this machine myself), but this a new area of experience for me.

Yeah.. my machine is happily crunching away. I keep tabs on system values with the occasional GPU-Z logging session, and so far things look okay. No signs of GPU stress, GPU load is mostly in the nineties, but temp stays around 60C. Should be okay ?
Humans may rule the world...but bacteria run it...
ID: 1848173 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1848179 - Posted: 12 Feb 2017, 1:51:49 UTC - in response to Message 1848173.  

temp stays around 60C. Should be okay ?

Well and truly.
My previous GPUs spent a lot of their time in the high 60s, and that was with their fans running at around 75% (I think it was). I only retired them to replace them with some GTX 750Tis, and more recently some GTX 1070s.


As for GPU load, the best indicator is the power consumption. You can have 90% or higher GPU load indicated, but the Power consumption might only be 45% It's possible to have only 75% GPU load, with 80% or higher Memory Controller load and Power consumption up around 60%
The current applications aren't able to make full use of all the available Compute Units in current GPUs (it's being worked on), so often when you see 95% GPU load, that may be for just 1 particular Compute Unit, but many of the others may only be partially loaded, or even idle.
If you ever see the Power consumption at 100%, then you know the GPU is being utilised to it's fullest. That's a ways off yet.
Grant
Darwin NT
ID: 1848179 · Report as offensive
Ghia
Avatar

Send message
Joined: 7 Feb 17
Posts: 238
Credit: 28,911,438
RAC: 50
Norway
Message 1848288 - Posted: 12 Feb 2017, 9:07:49 UTC - in response to Message 1848179.  

Looking at the numbers...power consumption is max 27%, fan speed 44% and memory controller load low 30's. And that's with up to 99% GPU load. There may be room for some tweaking, methinks...but then again, frying my graphics card would be a setback.. ;-)
Also wondering how much there is to gain by upgrading to the 'latest and greatest' GPU ?
Humans may rule the world...but bacteria run it...
ID: 1848288 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22158
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1848292 - Posted: 12 Feb 2017, 9:34:08 UTC

The figures are "about right", so don't worry about them.
The difference between a 780 and a 980 is a big step, and the 1080 is even further ahead.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1848292 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1848294 - Posted: 12 Feb 2017, 9:50:35 UTC - in response to Message 1848288.  

Looking at the numbers...power consumption is max 27%, fan speed 44% and memory controller load low 30's. And that's with up to 99% GPU load.

When running the CUDA applications, you generally get more work done per hour running 2 WUs at a time (each WU takes longer to process, but you do get more WUs processed per hour).
With the SoG application (the one your AV programme had issues with), with some command line settings you can get even more work per hour just running 1 WU at a time, than with CUDA running 2, particularly with the Green Bank Telescope work (Guppies- eg blc05_2bit_guppi_57787_41023_HIP69152_0028.2073.818.24.47.120.vlar_0).

The BOINC Manager will download work, and allocate it to the different science applications to see which one is fastest, then go with that application. However, the different types of WUs can have drastically different run times. Depending on the available work mix, it is possible for the manager to actually select the slowest application.

I'd suggest leaving things for a few days, and see if it allocates some more work to SoG & if it runs OK and just see how the run times work out.
Grant
Darwin NT
ID: 1848294 · Report as offensive
Ghia
Avatar

Send message
Joined: 7 Feb 17
Posts: 238
Credit: 28,911,438
RAC: 50
Norway
Message 1848299 - Posted: 12 Feb 2017, 12:31:06 UTC - in response to Message 1848292.  

The figures are "about right", so don't worry about them.

I'm leaving everything as is for now.

The difference between a 780 and a 980 is a big step, and the 1080 is even further ahead.

Would probably wait for the 1080 Ti. Would be an expensive upgrade, though...new MB and of course a better CPU w. cooling, bigger PU...at least (:
Humans may rule the world...but bacteria run it...
ID: 1848299 · Report as offensive
Ghia
Avatar

Send message
Joined: 7 Feb 17
Posts: 238
Credit: 28,911,438
RAC: 50
Norway
Message 1848306 - Posted: 12 Feb 2017, 13:00:32 UTC - in response to Message 1848294.  

When running the CUDA applications, you generally get more work done per hour running 2 WUs at a time (each WU takes longer to process, but you do get more WUs processed per hour).

Looks like there's only one WU running on CUDA at any given time, right ? How/where can i change this ?

With the SoG application (the one your AV programme had issues with), with some command line settings you can get even more work per hour just running 1 WU at a time, than with CUDA running 2, particularly with the Green Bank Telescope work (Guppies- eg blc05_2bit_guppi_57787_41023_HIP69152_0028.2073.818.24.47.120.vlar_0).

The error units are just sitting there, and the SoG application hasn't been in use, afaik. Should the error units have been retried ? Guess I must just wait and see if the BOINC Manager decides to allocate something to SoG (to see if it works).

The BOINC Manager will download work, and allocate it to the different science applications to see which one is fastest, then go with that application. However, the different types of WUs can have drastically different run times. Depending on the available work mix, it is possible for the manager to actually select the slowest application.

Still unsure about how work allocation is carried out. But I see the widely different run times, with guppies running on CUDA50 seemingly being the fastest.
Humans may rule the world...but bacteria run it...
ID: 1848306 · Report as offensive
Ghia
Avatar

Send message
Joined: 7 Feb 17
Posts: 238
Credit: 28,911,438
RAC: 50
Norway
Message 1848307 - Posted: 12 Feb 2017, 13:05:28 UTC

Learning so much from you guys ! Hope I'm not bothering you to death with the obvious...but this old gal likes to have some idea what she's doing, hehe...
Humans may rule the world...but bacteria run it...
ID: 1848307 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1848368 - Posted: 12 Feb 2017, 17:42:02 UTC

Since you are new to this, I would suggest just leaving it as is.

For more advanced users there is a way increase the number of work units for a particular type of application (ie cuda 42, 50, SOG)

I would suggest that you not do that at this time. I've burned out 2 780Tis a while back by increasing the number of instances of work on them. They held out for a while but no one likes that smell of burnt electronics when they come home.

As you get more into Seti, you will start to learn more of the things we do to increase our productivity. Feel free to ask questions.
ID: 1848368 · Report as offensive
1 · 2 · 3 · Next

Message boards : Number crunching : GTX780 Ti system


 
©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.