gtx680 lightning

Message boards : Number crunching : gtx680 lightning

To post messages, you must log in.

AuthorMessage
Profile trader
Volunteer tester

Send message
Joined: 25 Jun 00
Posts: 126
Credit: 4,968,173
RAC: 0
United States
Message 1347405 - Posted: 16 Mar 2013, 19:41:32 UTC

on monday my video card will arrive. before that happens i would like to know a few things. i hear that you can crunch more then one wu on the card at the same time, how is this achieved? also i have been reading that if you use a gpu to crunch you need to leave 1 core free if so is this just done by changing use at most x processors in muti processor systems to equal in my case 2 less then what my system shows since i have hyperthreading? and if i do crunch more then 1 wu on the card do i have to free up one core per wu running?


I RTFM and it was WYSIWYG then i found out it was a PEBKAC error

ID: 1347405 · Report as offensive
Profile ivan
Volunteer tester
Avatar

Send message
Joined: 5 Mar 01
Posts: 780
Credit: 232,268,443
RAC: 84,302
United Kingdom
Message 1347416 - Posted: 16 Mar 2013, 20:04:08 UTC - in response to Message 1347405.

on monday my video card will arrive. before that happens i would like to know a few things. i hear that you can crunch more then one wu on the card at the same time, how is this achieved?

In your projects\setiathome.berkely.edu\app_info.xml file, wherever you see
<coproc>
  <type>CUDA</type> 
  <count>0.4999</count> 
  </coproc>
make sure that the count is .4999 (or .3333 if you want to try three at a time)

also i have been reading that if you use a gpu to crunch you need to leave 1 core free if so is this just done by changing use at most x processors in muti processor systems to equal in my case 2 less then what my system shows since i have hyperthreading? and if i do crunch more then 1 wu on the card do i have to free up one core per wu running?

I don't know where this myth comes from -- it seems it might be necessary with some AMD/ATI co-processors, but I've never seen the need with my Nvidia rigs.

ID: 1347416 · Report as offensive
kittymanProject Donor
Volunteer tester
Avatar

Send message
Joined: 9 Jul 00
Posts: 45862
Credit: 814,617,599
RAC: 121,624
United States
Message 1347418 - Posted: 16 Mar 2013, 20:08:31 UTC - in response to Message 1347416.


I don't know where this myth comes from -- it seems it might be necessary with some AMD/ATI co-processors, but I've never seen the need with my Nvidia rigs.

It's not a myth, but it's not a requirement either. At least not on Intel/nVidia rigs.
Leaving a processor idle can slightly improve GPU output by reducing the preprocessing and loading of WUs into the card, especially on rigs that have several GPU cores running.
I am not sure it would provide much benefit on a rig with only a single GPU.
Kitties make wonderful traveling companions on your journey through life.

Have made a few friends in this life.
Most were cats.

ID: 1347418 · Report as offensive
Tom*

Send message
Joined: 12 Aug 11
Posts: 126
Credit: 12,004,781
RAC: 4,841
United States
Message 1347421 - Posted: 16 Mar 2013, 20:13:12 UTC
Last modified: 16 Mar 2013, 20:15:03 UTC

Cuda MB's do not need a CPU free but AP's usually do especially if they are the current OpenCL app.

One way to tell if they need a cpu free is suspend everything but one AP running
and watch the CPU usage. on my system I determined this as my CPU time did not drop below 100% for one cpu during this test.

Also my AP runtime went from 80 minutes per task to 40 minutes per task
when I allocated a 1 CPU per AP task, never did try two AP's at once to see if I needed two free or not though.

I used Claggy's advice and used the app_config.xml to easily do this with 7.042 or later BOINC.

Good Luck that 680 ought to perform well for you\

Edit - I have a NVIDIA GTX660

ID: 1347421 · Report as offensive
Profile ivan
Volunteer tester
Avatar

Send message
Joined: 5 Mar 01
Posts: 780
Credit: 232,268,443
RAC: 84,302
United Kingdom
Message 1347428 - Posted: 16 Mar 2013, 20:45:14 UTC - in response to Message 1347418.


I don't know where this myth comes from -- it seems it might be necessary with some AMD/ATI co-processors, but I've never seen the need with my Nvidia rigs.

It's not a myth, but it's not a requirement either. At least not on Intel/nVidia rigs.
Leaving a processor idle can slightly improve GPU output by reducing the preprocessing and loading of WUs into the card, especially on rigs that have several GPU cores running.
I am not sure it would provide much benefit on a rig with only a single GPU.

"myth" may not have been exactly the right word, but in my career I have come across many cases where some workaround had become embedded in the mythos (there's that word again...) of a workgroup, long after the need had been remedied. E.g. one case where I was working at UCL and I found my workstation grinding to a halt; on examination I found several instances of grad students running ROOT on the machine, exhausting CPU time and virtual memory -- they'd been told that "the fastest ROOT machine is xxx".

ID: 1347428 · Report as offensive
Profile trader
Volunteer tester

Send message
Joined: 25 Jun 00
Posts: 126
Credit: 4,968,173
RAC: 0
United States
Message 1347458 - Posted: 16 Mar 2013, 22:27:14 UTC - in response to Message 1347421.

Cuda MB's do not need a CPU free but AP's usually do especially if they are the current OpenCL app.

One way to tell if they need a cpu free is suspend everything but one AP running
and watch the CPU usage. on my system I determined this as my CPU time did not drop below 100% for one cpu during this test.

Also my AP runtime went from 80 minutes per task to 40 minutes per task
when I allocated a 1 CPU per AP task, never did try two AP's at once to see if I needed two free or not though.

I used Claggy's advice and used the app_config.xml to easily do this with 7.042 or later BOINC.

Good Luck that 680 ought to perform well for you\

Edit - I have a NVIDIA GTX660


hmmm wonder how fast the 680 is if your doing ap's in 40min. my rac i think with out the card doing standard wu's on this rig as is will even out to somewhere around 12k i7 3770k oc'd to 4.2ghz with the card i am hoping to get at least another 5k
I RTFM and it was WYSIWYG then i found out it was a PEBKAC error

ID: 1347458 · Report as offensive
Horacio

Send message
Joined: 14 Jan 00
Posts: 536
Credit: 75,967,266
RAC: 0
Argentina
Message 1347459 - Posted: 16 Mar 2013, 22:31:13 UTC - in response to Message 1347458.

My I7-2600 + GTX680 is currently doing 33K RAC and its stll rising...


ID: 1347459 · Report as offensive
kittymanProject Donor
Volunteer tester
Avatar

Send message
Joined: 9 Jul 00
Posts: 45862
Credit: 814,617,599
RAC: 121,624
United States
Message 1347462 - Posted: 16 Mar 2013, 22:38:39 UTC - in response to Message 1347458.



hmmm wonder how fast the 680 is if your doing ap's in 40min. my rac i think with out the card doing standard wu's on this rig as is will even out to somewhere around 12k i7 3770k oc'd to 4.2ghz with the card i am hoping to get at least another 5k

I believe a good running 680 shall be worth a lot more than 5k RAC.
Kitties make wonderful traveling companions on your journey through life.

Have made a few friends in this life.
Most were cats.

ID: 1347462 · Report as offensive
Profile trader
Volunteer tester

Send message
Joined: 25 Jun 00
Posts: 126
Credit: 4,968,173
RAC: 0
United States
Message 1347475 - Posted: 16 Mar 2013, 23:20:16 UTC - in response to Message 1347459.

My I7-2600 + GTX680 is currently doing 33K RAC and its stll rising...


woohoo!!! 30k+ rac here i come :)
I RTFM and it was WYSIWYG then i found out it was a PEBKAC error

ID: 1347475 · Report as offensive
Profile trader
Volunteer tester

Send message
Joined: 25 Jun 00
Posts: 126
Credit: 4,968,173
RAC: 0
United States
Message 1348300 - Posted: 18 Mar 2013, 22:59:33 UTC - in response to Message 1347475.

Thank you everyone for all the answers to my questions over the last few days because of it my transition to the video card and 7.0.56 went smoothly with no issues or problems. ty ty ty

ID: 1348300 · Report as offensive

Message boards : Number crunching : gtx680 lightning


 
©2016 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.