smicha wrote:Tutor wrote:It's two, two, two tools in one: an excellent benchmarking tool and an excellent assist tool for goal oriented GPU tweaking. Just as the best tool to assist in tweaking CPUs to handle Cinema4d CPU rendering speedily is Cinebench, the best tool to assist in tweaking GPUs to handle Octane GPU rendering speedily is now OctaneBench. Octanebench will help my system to peak again further to the right and simultaneously help me to establish overclock presets for various rendering tasks. Compare this - live in development - [ https://render.otoy.com/octanebench/sum ... GTX+780+Ti ] with Score Distribution pic, below - first large hump is where my system's performance had stalled before I had realized that there could be other uses to OctaneBench; lower humps to right is where OctaneBench's assistance came into play during tweaking of this air-cooled GPU system. Now, my ultimate goal is to get an OctaneBench score of, at least, 960 (or about 120 pts per 780 Ti ACX SC OC) on this system, while keeping all GPU temperatures below 85 degrees (C); my interim goal is to get a score of, at least, 912 (or about 114 per GPU).
The results you've posted are indeed impressive, especially these are air-cooled cards. According my calculations 8x780Ti may score around 1060 (watercooled) on stock bios. My GTX 780 6GB which has 2304 cores - scores 106 (1340 Mhz on core and 7000 MHz on memory). Hence a 780 Ti with 2880 cores has 25% more of computing power giving 1.25*106=132.5. Such results require overclock over 1300mhz and I am not sure how well the stock fan would cope up with heat dissipation. Hope to see you breaking 1000 score limit. Good luck
BTW I managed to get 0.044375 score per core on a 9600 cored system (0.046 for a single 780).
Thanks for the 780 TI projection and other information. Like minds often think alike and sometimes nearly simultaneously. Guess what my project for this weekend involves? Lets just say it involves a significant no. of GTX 780 TI ACX OC SCs and H20. I'm also thinking outside of the box - literally, because there'll so many of 780s that the box can't contain them even if I wanted the security of the inner womb of the case.
At least one observation that I've made from studying the various OctaneBench scores, is that, although Octane scales linearly for each additional, identical GPU card, the real world performance addition for each additional, identical card seems to be less and less than what I would have expected. Try it with the scores for the Titan Xs by looking at the performance for single card systems; then compare with that of dual Titan X systems, etc. on to four or five Titan X card systems. Interesting what each addition of an accelerator does, especially when it also adds more heat and/or act to block maximal air flow for its neighbor.