Richard (no@email.com) on 8/19/10 wrote:
---------------------------
>>>Do you realize that texture unit does interpolation on low-precision fix-point
>>>numbers? It's not good enough even for sonar/radar beam forming. Applicability to
>>>traditional HPC is extremely rare. Except, may be, final visualization phase, but
>>>that better handled by classic GPU APIs rather than GPGPU.
>
>>IIRC, 32-bit FP textures were in Nvidia's GPUs since series 6 hw.
>
>32-bit textures filtering wasn't available in hardware until the geforce 8 series.
>You could go around this by using a custom shader, but this would slow things down.
>
>The filtering hardware can be very useful for a certain class of problems. I have
>seen 100x speedups in ray casting, computer vision techniques, and custom graphics
>rendering (i.e. without using a graphics API). While GPGPU might not be applicable
>to all problems or even most traditional HPC problems at the moment, GPGPU certainly
>has its place and continues to evolve.
>
>While sonar/radar beam forming may not have been possible, displaying radar was
>done on Geforce6 series at high resolutions and high frame rates in 2005/2006. Although
>this was still graphics related, it was implemented using GPGPU techniques using
>shaders and was impossible on CPUs of the time. Cheap Geforces replaced expensive
>custom hardware (and the whole team designing it).

I certainly agree that 32-bit FP linear interpolation is applicable to wide class of computational problems, including many that have no relationship to graphics.