Sign in with
Sign up | Sign in
Your question
Solved

CUDA on ATI

Last response: in Graphics & Displays
Share
October 20, 2009 10:05:37 PM

I want to buy the HD 4770 card,
http://www.newegg.com/Product/Product.aspx?Item=N82E168...
but I want to be able to use CUDA/Stream apps with it.
(mostly password recovery and video editing)
Is this possible?
also interested in seeing if Anyone has got ATI cards to run CUDA
thanks in advance

More about : cuda ati

Best solution

a b U Graphics card
October 20, 2009 10:59:59 PM

Cuda is designed for nV hardware, not good for ATi hardware, you don't make CUDA for for ATi, you make the application work for Brook+, Stream or OpenCL/Direct Computer (compute shader model).

Love the 'password recovery' , yeah , ok. :lol: 

As for video editing, look at your app it either supports it or it doesn't an 9/10 it's OpenGL based and has nothing to do with CUDa.
Share
October 21, 2009 12:17:20 AM


I know CUDA is ment for Nvidia. I also know that bandit had been able to get CUDA programs to work on ATI radeon cards, due to some driver changing. Has anyone else been able to do this?
(yes brute forcing passwords/building rainbow tables)
m
0
l
Related resources
a b U Graphics card
October 21, 2009 5:12:44 AM

You could possibly get CUDA to work on ATi hardware, you can even run it on CPUs with alot of tweaking (which is easier than getting driver support from ATi), but Bandit/Regen NEVER made public his work (and retreated into a hole at NGOHQ after people being sucked in for page hits) it's more of a 'look what I think I can do' than a practical use of the hardware, his results wouldn't be what I call a success so much as proof of potential concept he never followed through on. He didn't even know you could run both ATi and nV drivers side by side and do PhysX on nV and graphics on ATi at the time, so my faith in his abilities to do what others did in other apps and scholastic works is low. :pfff: 
Even if it were made available in generic ATI drivers, it's far better to work in a more native or even open environment. If you need CUDA buy a cheap second hand GF8 card, it'll outperform any implementation of it running on the HD4770 as an unsupported work-around layer or layer upon a layer.
m
0
l
October 21, 2009 5:25:12 AM

TheGreatGrapeApe said:
You could possibly get CUDA to work on ATi hardware, you can even run it on CPUs with alot of tweaking (which is easier than getting driver support from ATi), but Bandit/Regen NEVER made public his work (and retreated into a hole at NGOHQ after people being sucked in for page hits) it's more of a 'look what I think I can do' than a practical use of the hardware, his results wouldn't be what I call a success so much as proof of potential concept he never followed through on. He didn't even know you could run both ATi and nV drivers side by side and do PhysX on nV and graphics on ATi at the time, so my faith in his abilities to do what others did in other apps and scholastic works is low. :pfff: 
Even if it were made available in generic ATI drivers, it's far better to work in a more native or even open environment. If you need CUDA buy a cheap second hand GF8 card, it'll outperform any implementation of it running on the HD4770 as an unsupported work-around layer or layer upon a layer.



so then I should just buy a Nvidia card.
Thanks for the help.
m
0
l
a b U Graphics card
October 21, 2009 6:56:49 AM

StrongWinds said:
so then I should just buy a Nvidia card.
Thanks for the help.


If supporting CUDA's all you wanted to do with them, sure, no doubt there.
I work with both at work, and I don't see the point to your thread if that was your only concern especially as it applies to your 2 supposed uses (which work on both platforms and will work better under OpenCL on the newer cards). You made it seem as if there was some other reason bringing you to the HD4770 (like price, performance, efficiency), but obviously that's not the case.

Like your 'password recovery' I question your motives. :pt1cable: 
m
0
l
a c 106 U Graphics card
October 21, 2009 5:53:17 PM

Well, cuda apps would have to be recompiled to run on radeon cards, so you could probably do dynamic recompilation, but it would slow things down and some programs may simply not run as nVidia cards support some features for which the Radeon cards have no equivalent so it would just depend on if the program contained said features.
m
0
l
!