Originally Posted by trumpet-205
Sad news. While Nvidia has double precision for CUDA 2.x, it is capped at 25% performance for GeForce line. Seems these GPU manufacturers don't consider GPGPU (BOINC) to be that helpful.
Umm, that is a little over simplified if you ask me. Sure Nvidia has designed various hardware with different capability, but FERMI GPU's do in fact have more capability. The older CUDA capable chips have fewer DPFPU's but still have full compliment of SPFPU's. Remember the Tesla line is the real HPC enabled professional compute engine that doesn't have rendering or gaming as it's main target... while Geforce is game directed.
A network search of CUDA double precision hardware turns up a lot of information about these differences... it has nothing to do with BOINC... BOINC being relatively new to GPU computing (started around late 2008 IIRC after CUDA was released).
One can't compare a Beetle with a Corvette and say the Beetle is capped at 50% of the Corvette... It isn't an apples vs. apples comparison and is misleading at least. How can something be "capped" if the physical components aren't there?
Geforce is fast, cheap, and wasn't designed for HPC, while Tesla is so designed. Also, Quadro and Tesla at least triples available memory capability... They are simply different animals these Tesla vs. Geforce products. At least, according to everything I've read and cared to research on the web...