[sdiy] Hardware convolution box?

[rsdio at audiobanshee.com]

Don't forget to consider the latency of sending an audio stream through video shaders for effect. Typically, these GPU solutions end up trading increased latency for faster overall throughput of certain types of calculations.

Brian


On Feb 13, 2017, at 7:43 PM, cheater00 cheater00 <cheater00 at gmail.com> wrote:
> Interesting... whatever the way is that those shaders are implemented,
> according to your numbers the Raspberry Pi 2 is less than twice as
> expensive per GMACS as the most cost effective TI DSP, and obviously
> has a gigabyte of ram rather than just a couple megabytes. And that's
> comparing a $200 dsp chip to a complete $40 raspberry pi board with
> everything on it so it's very unfair to the Pi. The question is
> whether this kind of performance (24 GFLOPs) can be achieved from some
> sort of realtime OS or only using their linux drivers.
> 
> On Tue, Feb 14, 2017 at 4:36 AM, cheater00 cheater00 <cheater00 at gmail.com> wrote:
>> Neil aren't the shader engines just in software on arm? Isn't it
>> better to just use the arm directly?
>> 
>> On Mon, Feb 13, 2017 at 9:49 PM, Neil Johnson <neil.johnson71 at gmail.com> wrote:
>>> If you can work with the slightly odd floating point format in single
>>> precision GLES shader language then the shader engines in the Raspberry Pi
>>> (1, 2 & 3) will give you about 24 GFLOPS.
>>> 
>>> Neil
>