GTX470 benchmarks / comparison

[Skeletor]

Club3D GTX460 1Gb = 1.62 Mb/s
8800GT = 0.59 Mb/s


Both ????
Cant make it to work together. Waiting for 2.3.

Is the 460 really not faster than that? I mean 470 can do 2.69 according to buda. In games the difference is alot smaller

Not sure if I should get a 460 2GB or a 470

[radiance]

according to wikipedia: http://en.wikipedia.org/wiki/GeForce_400_Series

the 460 is 907 GFLOPS
the 470 is 1088 GFLOPS

so the difference should not be that big.
maybe something went wrong with the benchmarks, i think a GTX460 should be able to do approx ~2.3 or ~2.4 Megsasamples/sec on the bench scene

Radiance

[Skeletor]

Thanks for clearing that up. I just ordered the 460 2GB

[timbarnes]

Ok got it to us both gpu's but the 'add' button seems to be greyed out sometimes. seems it has to be set nbefore loading a scene.
Also shouldnt the memory read 3000 as there are two cards instead of 1503?

I hope this changes in a future release: I notice that as soon as a model is loaded into the GPU it becomes impossible to change the configuration of GPUs. As I understand it, when you use a plugin, the model is immediately loaded into the GPU, so if you haven't set the right GPU setting in the plugin, you can't change it later.

So a feature request: could we have the ability to change which GPU is used for rendering after the model is loaded? I understand it would require reloading the GPUs and re-starting the render. It's important to me because I have a Quadro FX 580 (intended for screen display) and a GTX 470 (for rendering), and if I accidentally let Octane use the Quadro, my system is pretty much unusable.

An alternative solution: a configuration file that persists between sessions in which I can tell Octane not to use the Quadro.

Thanks for listening,

tim

[radiance]

Hi,

2.3 solves this with a floating account profile.

Radiance

[Kitsune]

This is the new GF104 with superscalar architecture, but only 224 of the 336 cores are used according to the device manager.

Nvidia's GF100-series had 32 CUDA-cores per Streaming Multiprocessor (SM), the new GF104 has 48 CUDA-cores per SM. It is entirely possible, that a software coded for the older hardware (old 400-series) must be modified before it can use all those CUDA-cores in GTX 460!

If so, the rendering speed of Geforce GTX 460 should increase 1.5x after the fix which allows using all the CUDA-cores.

[timbarnes]

2.3 solves this with a floating account profile.

Thank you—that's great. A lot to look forward to .

tim

[crazyhorse666]

i have been looking extensively at the gfx460 cards today

...looks like we need a code optimised version for the gf104 chip
and this will effectively replace the gtx470 at 3/4 the price (2gb version) and the 2GB version has 50% more memory space !!

+runs cooler and TDP 160W max

Crazy

[radiance]

This is the new GF104 with superscalar architecture, but only 224 of the 336 cores are used according to the device manager.

Nvidia's GF100-series had 32 CUDA-cores per Streaming Multiprocessor (SM), the new GF104 has 48 CUDA-cores per SM. It is entirely possible, that a software coded for the older hardware (old 400-series) must be modified before it can use all those CUDA-cores in GTX 460!

If so, the rendering speed of Geforce GTX 460 should increase 1.5x after the fix which allows using all the CUDA-cores.

Does anyone with the new GTX460 notice any change in compute model ?
is it 2.0 or 2.1 ?

Radiance

[peterd]

Mine Club3D GTX460 1Gb says compute21