Let's suppose you want to run VoIP on a TI DSP, but the processor isn't fast enough. What do you do? Sure, you could design your own DSP, but then you'd have the annoying problem of having to write all your own software.
AudioCodes has a better solution. They hacked the TI hardware to get a 20% performance boost. That way they can use TI's software* while getting the speed they need. Neat trick! Now if somebody could tell me how they did it… AudioCodes won't reveal how they pulled off this stunt, other than to say that they did it with TI's cooperation.
You might also be wondering: Why in the world does AudioCodes need so much performance? One reason is that the company is moving rapidly into HD voice. HD voice (aka wideband voice) looks like it's going to be a big deal in the future. This is a development worth watching!
What is this about? It is a little unclear.
Did they just write more optimal software than TI? A big fat yawn if they did.
Or did they re-implement the actual DSP hardware logic to improve some instructions performance?
What's the point of putting out an article about something when you can share the details of what they did. It would be nice is TI would share what was done so others could achive this performance increase also.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.