CUDA Cores verus speed/Memory

[meshlife]

Your FAQ indicates that the number of CUDA Cores directly impacts the performance of Octane. Doing some research, I found the following:

GTX 690 (Kepler): 3072 CUDA Cores
GTX 1070 (Pascal): 1920 CUDA Cores
GTX 1080 (Pascal): 2560 CUDA Cores

and so on. Does this mean that four GTX 690s would be superior to four GTX 1070s? How does this work regarding the difference in memory? Should I just find the Kepler GPU with the most CUDA Cores?

Thanks again for all your help. I appreciate it.

[nuno1980]

Lastest version of OR supports GTX 10x0 but its got still bad optimization... But you wait to buy any GTX 10x0 for better optimization.

See Pascal cards support

[meshlife]

Does this mean that the 1080/1070, when optimized, will have better performance than GTX 690 even though it has more CUDA Cores?

[bepeg4d]

Hi meshlife,
unfortunately the number and the type of cuda cores are changing on every new architecture, so is not possible to compare Kepler with Pascal cards and so on. The comparison can be made only between the same series, eg 680 with 690 or 1070 with 1080.
For that reason, the best way to compare different cards from different architecture, is by using the OctaneBench db:
https://render.otoy.com/octanebench/res ... ingleGPU=1
At the moment Pascal cards are not supported yet, because Nvidia has not yet published CUDA 8, but with the non optimized version of OctaneRender 3.03.2, some users are reporting that at the moment 1080 has roughly 29% more speed compared with a 980:
viewtopic.php?f=11&t=55370
Since 690 scores almost similar to an 980, 1080 is already faster than 690 (~25%), and when Octane will be optimized, it should be even faster
ciao beppe

[meshlife]

OK great, thank you again!