Article: Parallelism at HotPar 2010
By: Gabriele Svelto (gabriele.svelto.delete@this.gmail.com), August 18, 2010 6:11 am
Room: Moderated Discussions
Michael S (already5chosen@yahoo.com) on 8/18/10 wrote:
---------------------------
>Do you realize that texture unit does interpolation on low-precision fix-point numbers?
Actually in DX10-level hardware it should work on textures with anything from 8-bit fixed-point to 32-bit floating-point components. However I don't know if the rounding mode, precision and such are mandated in the standard or up to the implementer to choose. If you cannot rely on obtaining the same results from different cards/vendors or if its default behavior doesn't match your requirements it would be of little use anyway.
>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.
Not necessarily, one of the interesting applications of GPGPU is that it enables accelerating visualization techniques and algorithms which couldn't be implemented on vanilla graphics API and were often done off-line on the CPU (volumetric visualization, global-illumination solvers, ray-tracers, etc...). Being graphics-related those applications can also make very good use of the specialized hardware even though they are not implemented using traditional graphics APIs.
---------------------------
>Do you realize that texture unit does interpolation on low-precision fix-point numbers?
Actually in DX10-level hardware it should work on textures with anything from 8-bit fixed-point to 32-bit floating-point components. However I don't know if the rounding mode, precision and such are mandated in the standard or up to the implementer to choose. If you cannot rely on obtaining the same results from different cards/vendors or if its default behavior doesn't match your requirements it would be of little use anyway.
>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.
Not necessarily, one of the interesting applications of GPGPU is that it enables accelerating visualization techniques and algorithms which couldn't be implemented on vanilla graphics API and were often done off-line on the CPU (volumetric visualization, global-illumination solvers, ray-tracers, etc...). Being graphics-related those applications can also make very good use of the specialized hardware even though they are not implemented using traditional graphics APIs.
