I recently used the (new) BMW-benchmark from blender.org with blender 2.78b (yeah, I know: already outdated) to test two of my older graphics cards: i rendered the benchmark with a GTX 550Ti and afterwards with a GTX 650Ti on the same system (Win7, 4GB RAM, SSD, Intel i7 950).
For some reason the newer GTX 650Ti is slower than the older GTX 550Ti (20 seconds!) - does anybody know why?
Concerning to the Datasheets the newer one should be faster - and not vice versa…
The 4xx/5xx architecture is simply superior for compute tasks, it was designed to do complex GPGPU well. The 6xx architecture was designed for gaming efficiency (which it did deliver) at the cost of GPGPU performance (which was disappointing).
different architecture, the shader clock was separate (and faster) on the 5xx series(and lower?). this meant more performance per cuda core i think. i have a 560 ti in my backup computer and that thing still keeps up good.
Hey, thanks for all your responses!
Already thought that it might be something like this but didn’t believed it - it sounded too strange that older hardware should be better than newer one (at least in a specific area)…