Forum > Windows
GTX460 and GTX295 combination.
Pizzadude:
I have just installed a GTX460 alongside my trusty GTX295 and what I would like to do is have the GTX460 crunching at least two workunits at once. I have gone in to the App_info and reduced the Cuda limit from 1 to 0.5 and this does indeed allow the GTX460 to crunch to units, however the downside is that it also causes the GTX295 to attempt the same and this slows the whole system to being unusable.
Can you specify somehow that only the GTX 460 multitasks ?
Ghost0210:
Unfortunately not. The <count> field will apply to both cards
Many have tried to find ways round this, but my understanding is that Boinc just isn't sophisticated enough to be able to do this yet
Pizzadude:
Another strange thing I have noticed is that using one work unit per GPU the GTX460 crunches faster on its own, crunching with the GTX295 also slows down the GTX460 by as much as 50%.
Any suggestions ?
Ghost0210:
--- Quote from: Pizzadude on 01 Jan 2011, 04:18:06 am ---Another strange thing I have noticed is that using one work unit per GPU the GTX460 crunches faster on its own, crunching with the GTX295 also slows down the GTX460 by as much as 50%.
Any suggestions ?
--- End quote ---
Not seen this before, but have noticed that when crunching 2 tasks on my 465, when a task loads, and there is an existing task running on the 465; the existing task will pause crunching untill the new task has loaded to the GPU and starts crunching. So for example :
Task A finishes
Task B is 50% complete
Task C starts to load
Task B pauses crunching (or time to completion and progress bar pause) for a few seconds (5-10 seconds max)
Task B continues normallly
Task C fininshes loading and starts to crunch
Is this similar to what you are seeing? Can't remember if my 240 was slowing my 260 when I had that setup, I'm currently running an ATI and 465 setup so can't really compare your setup with mine
Raistmer:
did you try to use some tools to set different cuda tasks to different cpu cores?
Navigation
[0] Message Index
[#] Next page
Go to full version