By: fastpathguru (fastpathguru.delete@this.yahoo.com), November 1, 2006 2:30 pm
Room: Moderated Discussions
Stephen Goss (x@x.com) on 10/21/06 wrote:
---------------------------
>Accelerator apparently can target either GPUs or multi-core (when they write it)
>but IMO the technique will have niche appeal whilst GPUs don't implement IEEE 32bit
>floating point or 64bit for that matter. What needs to be done is for GPUs and CPUs
>to be aligned on FP behaviour and/or CPUs to contain lots of simple but standards
>complying FPU cores. At the end of the day the trend is surely for the application
>software to not need to know exactly what hardware is running under the hood to extract the maximum performance.
ATI's Xenon GPU in the Xbox360 apparently implements full IEEE 32b FP throughout.
http://www.beyond3d.com/articles/xenos/index.php?p=07
fpg
---------------------------
>Accelerator apparently can target either GPUs or multi-core (when they write it)
>but IMO the technique will have niche appeal whilst GPUs don't implement IEEE 32bit
>floating point or 64bit for that matter. What needs to be done is for GPUs and CPUs
>to be aligned on FP behaviour and/or CPUs to contain lots of simple but standards
>complying FPU cores. At the end of the day the trend is surely for the application
>software to not need to know exactly what hardware is running under the hood to extract the maximum performance.
ATI's Xenon GPU in the Xbox360 apparently implements full IEEE 32b FP throughout.
http://www.beyond3d.com/articles/xenos/index.php?p=07
fpg