Forum > GPU crunching

New Video Card

(1/2) > >>

TouchuvGrey:
it looks like my trusty GTX460 has passed away. i will be getting
a new card to sit beside my GTX560ti.

   i am considering a GTX760 but looking at the multibeam v7 application
i see that the cuda 50 is suggested for 6xxx and up Kepler cards and the
cuda 42 is for 4xxx and 5xxx Fermi cards.

   it seems to me that if i get a 760 it would be best to use the cuda 50
and get the most out of the more capable card.

     Am i right here ? and does anyone have any thoughts or suggestions ?

Claggy:
My trusty GTX460 passed away too (months ago through), Got myself a Gainward GTX760 Phantom, arrived today, and i've only fitted it in the last hour,
used the Lunatics 0.42 Installer selecting Cuda5, and got all my ATI Wu's resent as Nvidia Wu's,
over the next few days i'll do some Cuda Benches, my feeling is Cuda5 will be fastest.

Claggy

juan BFB:
To run 2 diferent families of GPU on the same host the best you could do is running 2 intances of Boinc on the host.  By doing this you could select run one GPU on each instance and choose the best configuration for each GPU. That includes the cuda version, number of simultaneus WU, best command line option, etc.

TouchuvGrey:
Thank you Juan, i had not even
considered that option.   :-[

juan BFB:
I use that to exactly to make a 580 (Fermi) work with a 670 (Keppler) on the same host since each one uses a diferent CUDA version. Works perfect.

If you run AP and the newer builds you have another good reason to do that since they use a diferent command line to optimize each GPU .

BTW The secondary "good" efect is your cache size limit will automaticaly rise from 200 for both GPU“s when you run 1 instance to 400 if you run 2 instances and you could control each cache size separately.

Navigation

[0] Message Index

[#] Next page