and noticed the discrepancy between the iRay and Octane results (between Titan and 780Ti) and the Blender Cycles results. Can somebody explain why the 780Ti doesn’t achieve better results with Cycles than the Titan? The only reason I can suppose is the use of double precision in Cycles (which is semi-locked in the 780Ti), but I’m not aware of this. Are there other possibile reasons? Or there is an error from the executor of the benchmark?
In my opinion is due to the fact that he used blender 2.66, which uses the old cuda. Probably with the lastest version results would be in line with octane’s.
Once again, Tomshardware fails at properly benchmarking Cycles. When they benchmarked the Titan, they adjusted the tile sizes, when they benchmarked the 780 they didn’t (even though they claim they did) and now with the 780 Ti they don’t mention tile sizes at all (but they didn’t adjust them, either).
Those are the results you get when running the Cycles benchmark with the stock Mike Pan blend file, where the tile sizes are not optimal for GPU utilization in newer Blender versions (because the tile size calculation has changed). Don’t trust Tomshardware for benchmarks, trust the community.
Also, Cycles doesn’t use double-precision and likely never will.
Titan is some ~10% faster than a standard GTX780, Ti is 10% faster than a Titan with fully unlocked GK110 like the new Quadro cards. That`s about 20% increase in performance over standard GTX780.