New 344.48 WHQL driver released.
384 bits would still be far short of the 512 bits memory bus of the GTX280!
New 344.60 WHQL driver released.
I have played around with CUFFT on both a titan black and gtx980, and the 980 is quite impressive.
I found some software called Nukadafft which includes a measure of time on gpu vs wall clock, you might want to check it out.
I am doing some work to improve the 3D FFT needed for molecular dynamics calcs, and so far the 980 is proving much faster than the titan black.
They expected 980Ti/Titan2 in Q4 2014.
I don’t know which rumors to believe, but there was one that hit the blogs a couple of days ago, and according to it, 980Ti/Titan2 shouldn’t be expected until 2016!
New 344.65 WHQL driver released.
New 344.75 WHQL driver released.
Could you share some more details on this? Im also very interested in 2d 32bit fft performance, but I see varying results. Some claim the smaller memory bus is limiting in cufft for Maxwell, resulting in much reduce performance (see: https://devtalk.nvidia.com/default/topic/782314/cuda-programming-and-performance/cufft-speed-on-gtx980/). But perhaps it depends a lot on the specifics?
I don’t have a Maxwell part yet, but could you share your findings in more detail?
New 347.09 beta driver released with CUDA 7.0 driver.
347.09 WHQL certified driver released.