Is this setup ok ?
21 Feb 2013 8:45:25 UTC
Topic 196815
(moderation:
The reason i ask is because i can not seem to get 2 threads to run on each GPU .
I have 2 video cards a 570 and a 560 installed . I would like to get this proper before i install the other 2 570's .
http://www.worldcommunitygrid.org/download.php?xml=1
http://www.worldcommunitygrid.org/download.php
http://www.ibm.com/
1
120
einsteinbinary_BRP4
4
.5
1
Is this setup ok ?
)
Why not just use the web preferences page? That's what it was set up for.
Hello, You generally need
)
Hello,
You generally need the use_all_gpus flag in a cc_config if you have two cards, so that's good. For making two simultaneous tasks run on each card, at E@H you use the E@H preferences in your online account, by setting the "GPU Utilization Factor" value to 0.5 (your system will do the inverse of the value specified, so 2 simultaneous for 0.5, 3 for 0.33, 4 for 0.25, etc). This value should be set conservatively at first, at 0.5; many cards cannot handle more. And you will only start seeing two tasks per GPU at a time after you have set this value and your client has successfully contacted E@H after making the change AND received new work units; if you are already maxed out you can up your "maximum work buffer" a bit in your client's computing preferences.
OK I went to the E@A Pref
)
OK I went to the E@A Pref page and set "GPU Utilization Factor" value to 0.5 now I am just waiting for what is in my buffer to run out .
From what i understand all these that I have will have to be run before I can get new tasks that will run 2 per GPU ?
That is if my GPU's will run 2 each . That is my other question .
Will a GT 570 run 2 tasks ? There water cooled so heat shouldent be a problem .
The system there runing in is a i7 950 on a eVGA classified board (171)X58
with 12gig's of ram .
oops another question , I have 3 banks with 4 gigs of ram in each.
I have also got 6 sticks of 1 gig each avalable to me , would I be better off
useing the 6 x 1 gig sticks in terms of proformance ?
I can watercool the 1 gig sticks to keep it cooler .
I just thought I would
)
I just thought I would mention that i am useing stock cooling on the 560 and 570 i have in my system right now and the best fan control software i have found is eVGA's precisionX software . It allows you to set a custom fan curve ,
it also alows you to go manuel with them .
OK here it goes
I now have 2 tasks runing on each the 560 and the 570 as well as 8 on the processer .
I did as you told me and all is good, one thing i also did is to disable runing tasks that have GPU versions as that took 24 hrs . it makes me wonder why that is set on as a default . I had to reset because of that and I realy did not want to dump tasks like that. I hope I dident mess up the project by dumping the long runing tasks , sorry .
I've found it can take quite
)
I've found it can take quite some time (hours) for changes to the web preferences to take effect on the local system (even using 'update'). Leaving it overnight seems to mean it's working as expected by the morning!
For best performance it's generally reckoned you need to leave one cpu core free to feed the GPUs. Not sure how this works with hyperthreading, so I set my CPU count to 3 (on a quad core hyperthreaded) rather than 7.
As to doing BRP4 work on the CPU when the GPUs are much faster, AIUI it lets people who don't have GPUs set up contribute too. I think the points for CPU work are roughly the same whatever work the CPU is doing.
It would have been better
)
It would have been better ,for me ,if the BRP4 stuff wasent loaded to the CPU by default .
But I guess I should have looked at the preffrences page 1st . Not that it would have done much good as I probly wouldent have spotted it .
I-R-SteamFitter and way green to this sort of thing.
It took me at least 3 days to spot it because i was looking for something else .
I guess I should have read the manuel .;p
If you know enough about your
)
If you know enough about your computer to care about airflow and cooling, you are probably OK. Even running the maximum number of work units does not seem to stress video cards as much as running graphics benchmarking software or a new 3D game at high resolution.
I also don't do BRP4 on my CPU because of the excess heat and electricity usage. GPU computing is much more efficient.
I am running BPR4 tasks both
)
I am running BPR4 tasks both on my laptop and my SUN WS,
no GPU. They both run 24/7, with no ill effect.The laptop has a SSD.
Tullio
RE: I've found it can take
)
at first i was under the impression that a change in GPU utilization factor would not take effect until all work in the queue (at the time i made the change) finished, and that only work downloaded after i made the change would crunch using the new GPU utilization factor...but then i was told otherwise by a well known and very knowledgeable forum member (it may have been Richard Hasslegrove, but i can't remember for certain).
last night swapped a GTX 560 Ti for a GTX 580 in one of my machines. with the intentions of testing various numbers of simultaneous tasks at once and then documenting the run times and any changes in efficiency, i changed my GPU utilization factor from 0.33 to 1. while the next 6 or so tasks continued to run 3 at a time, the switchover to 1 task at a time happened pretty quickly (within a few hours of having changed the GPU utilization factor via my project web preferences). having collected enough data to now know the average run time of BRP tasks run one at a time on my GTX 580, i was ready to switch to 2 tasks at a time. i made that change earlier this morning, and still they're only running 1 at a time. while i know that there are improvements in efficiency to be had by choosing to run 2 or 3 simultaneous tasks over a single task at a time, running 2 or 3 simultaneous tasks are still going to generally take longer to finish than a single task takes to finish. but i would have expected the switchover to 2 tasks at a time to have taken place by now. granted, i know it'll probably happen by the time i wake up in the morning, but its kind of annoying that it doesn't happen faster than that. i'd love to test 2 at a time now as opposed to having to wait for the change. maybe i'll just go back to using the good old app_info.xml file to override the web preferences temporarily (at least until i get my testing done).[/i]
my card does 1 task in about
)
my card does 1 task in about 30 mins while running 2 gets done around 53 mins... so 1 task at a time 2 an hour or 2 tasks at 1 time and get them done before the hour... hhhmmmm.... I thinks 2 task at a time... not much of a gain for me to do 3 task at 1 time and since I use this desktop it's best for me to only do 2 at 1 time...
PC setup MSI-970A-G46 AMD FX-8350 8 core OC'd 4.45GHz 16GB ram PC3-10700 Geforce GTX 650Ti Windows 7 x64 Einstein@Home