GTX470 benchmarks / comparison

[pedrofelipe]

I think we need to do more tests when we've figured out the issues with the new fermi cards.
the fact that the device manager says it's a compute 1.0 (eg 8000 series) cards with only 110 cuda cores kinda makes these benchmarks not %100 reliable.

I'm hoping to get a reply from nvidia about this soon.

Radiance

If it did around 2.0 with only 110 cuda cores squeazing juice then when all cuda cores would be working... shish! that would be great!

[pixym]

My results:

[pixie]

The results from my GTX 285:
1.91 Megasamples/px

[pixym]

This is the result I was waiting for!
A gtx 285 gives 1,92 Msamples/sec and a fermi gtx 470 gives only 2,05…

[n1k]

Octane doesn't yet detect fermi cards right so the results are not quite valid. It makes no sense that card with 2x more cuda cores have similar performance as gtx 285.

[ycarry]

Good old GTX 260:

[radiance]

This appears to be a bug in cuda 3.0, i'm awaiting help from nvidia regarding it.

Radiance

[DavidT]

Perhaps this screenshot will be of some use. It shows the GPU Load at 95%. Now that I have set the fan from auto mode to smart mode the GPU temperature stays around 66C.

[radiance]

thanks for that, that's very helpfull.
i'm working on a fermi compatible version for our upcoming 2.1 release.

Radiance

[justix]

I figured you were interested in the final render time, so I let Octane finish the render.

Here is a screenshot for while it is rendering and a screenshot when it is finished, showing the final time (01:15:38)

It looks like my GTX 470 averaged about 2.06 megasamples/sec. I didn't try any overclocking.

I may very much wrong but render time looks slightly high or am I just writing rubbish?
(ok ok the scene has his complexity I understand, just pushing a bit..)

Here a screenshot of my old 8800GT in action...LOL 0.63 megasamples/sec only....