I"m currently working on a code that requires double precision for many of its computations. I have a CUDA version of the code, but of course NVIDIA's current crop of GPUs has abysmal performance for double precision. Because of the vastly better double precision performance numbers, I'm interested in the ATI 5000 series cards. If the ATI Stream SDK supported double (even it not fully optimized), I'd pick up several 5870s for my application.
Can anyone in the know hint at a timeframe for GPU double precision via OpenCL?
Thanks,
Mike