GT 630 as a dedicated CUDA card for Mercury Playback

[Geek55]

Hello all
I currently have a 6970 in my PC and although it's a great card for gaming, it is lacking in the GPU acceleration department. I'm thinking about purchasing Premiere Pro CS6 and I really want to be able to use GPU acceleration, but it only works with CUDA, not OpenCL, and I don't have the money to replace my 6970. So I was wondering if I could use a GT 630 or another card of the price point along side my 6970 purely for CUDA rendering, and if so, if it would be much of an improvement over simply using my CPU.
Any help would be appreciated

 

[mayankleoboy1]

dont know if gtx630 would be faster than the cpu or not but one thing i do know :

nvidia has three cards named gtx630. one is based on kepler arch,the other two are older parts. no knowing which you will get. so its better to get a gtx640. atleast you will be sure you are getting the kepler architecture and 28nm goodness

 

[pepe2907]

How exactly you will make 6970 and an Nvidia card work together?
I am quite surprised Premiere CS6 works only with CUDA yet and I am sure it will change with some upgrade soon

 

[thently]

How exactly you will make 6970 and an Nvidia card work together?
I am quite surprised Premiere CS6 works only with CUDA yet and I am sure it will change with some upgrade soon

This is not a problen in Windows 7 was in Vista though. If you wait the 6970 might get support the 6770m is supported on macs only. Very odd but I bet they cover PC's over time and newer cards.. Just FYI the kepler cards are not on the supported list for AP cs6 should work just sayin.

Thent

 

[pepe2907]

"This is not a problen in Windows 7 was in Vista though."
Thanks for the info, I'll note that.
Just a note on using Keplers for GPGPU - I think they are slower as older Fermis, because they are really badly crippled down. If I am not wrong they are allowed to achieve only 1 double precision instruction in 16 cycles per calculating unit in GPGPU mode while Fermi based are artificially limited down only to 1 per 4 cycles or something like that effectively causing them to be 4 times less capable of GPGPU high precision calculations per unit per cycle.
/While Teslas although being basically the same hardware, are not artificially limited that way./