optimization for GTX460/GF104

[waterthinking]

Hi,

Looks like performance on 460 stays the same as with pre2.3 versions; I know that has allready been pointed out. I just wanted to add a link to some discussion on the CUDA forums about the subject. Don't understand much of what they are saying, but looks like it could be of some help.

Anyway, here's the link:

http://forums.nvidia.com/index.php?show ... 096&hl=460

Pd. If you don't want to go through all the thread, just look at what "seibert" has to say; looks like he knows something about the GF104 chip

[pixelrush]

His replies seem to indicate drivers with cuda 3.2 will be required and Octane code would need to be optimised to take full advantage of the presently lazy cores.
Ok so in a timeframe of a few months and probably more, depending how everything is going at Nvidia and Refractive, there is some potential to get more work out of a 460 - maybe also a 475 by then too.

[radiance]

as said we need to wait until we get cuda 3.1 working with octane.
We've been working hard on that, nvidia made some changes that completely break octane,
and we're communicating with them to get them resolved.

All GTX460 users should have a bit of patience...

Radiance

[Ayertosco]

I think we'll have to wait till cuda 3.2 hoping the Nvidia fix this problem. As far as i understand from the link posted, even with Cuda 3.1 the problem remains

[radiance]

I think we'll have to wait till cuda 3.2 hoping the Nvidia fix this problem. As far as i understand from the link posted, even with Cuda 3.1 the problem remains

Can you tell me where you read this information ? any link ?

Radiance

[cpember]

http://forums.nvidia.com/index.php?show ... 096&hl=460

cp

[Kitsune]

Can you tell me where you read this information ? any link ?

You probably want to read this, Radiance:

http://forum.beyond3d.com/showthread.php?t=58077

Theoretically GTX 460 could be faster than GTX 470 in CUDA calculations, but reaching that theoretical maximum speed seems to be extremely difficult.

[radiance]

I say we wait till 3.2 is out and try it.
Since this is a new architecture and brecht starts in 2 weeks in need of a computer for the office, i'll get some of these cards.

i'm hope nvidia will have the relevant documentation about modifying code to run fast on compute 2.1 cards.

Radiance

[hypergrafica]

hi, CUDA 3.2 is out!
Shall we install this last toolkit?
or 3.0 is better for now?
Thanks

[mib2berlin]

AS windows user you don´t need the toolkit.

mib