HoloLens secret sauce: A 28nm customized 24-core DSP engine built by TSMC

Re:HoloLens secret sauce: A 28nm customized 24-core DSP engi... by yosefk on Tuesday August 23, 2016 @03:53AM (#12342065)

I'm not personally opposed to Tensilica "dying", especially if it doesn't involve Cadence dying, since they are a (somewhat indirect) competitor from my point of view, but ARM is not a substitute for Tensilica. You can't extend its ISA unless you license the arhictecture for $30M. A DSP like Tensilica's is also much more efficient than ARM on a range of tasks, and in particular, having local memory instead of caches is done for a reason. (The least efficient accelerator of all and the favorite of academics who have easy access to it, GPUs, also have this.)

As to their compiler licensing - that's what happens when you develop for a small niche, you get more expensive tools which are worse than the free ones used by the majority. But it doesn't mean that the thing doesn't have its uses. I hear that a recent chip by AMD had 40 Tensilica (smallish, inaccessible to most software) cores.

The same is true about CEVA (which was mentioned in a sister thread), more or less.

[ Parent | on HN ]

Re:HoloLens secret sauce: A 28nm customized 24-core DSP engi... by n00b101 on Tuesday August 23, 2016 @08:43AM (#12343206)

The least efficient accelerator of all ... GPUs

Interesting, could you please elaborate on why you think this? Any data you know of on the subject? What are better alternatives in you view?

[ Parent | on HN ]

Re:HoloLens secret sauce: A 28nm customized 24-core DSP engi... by yosefk on Tuesday August 23, 2016 @09:34PM (#12349062)

In terms of performance/power or per dollar, GPUs are only good at graphics - if you're doing deep learning, for instance, GPGPU will lose to almost any other accelerator, Google's TPU is one example, FPGAs are another, a DSP designed for deep learning by CEVA, Tensilica or Synopsys are a third, and by "lose" I mean factors of 5-10x at least, assuming the same process node and operating environment, especially when deploying networks as opposed to training them. But deep learning is just one example, it'll be the same story elsewhere. The strength of GPGPU is high availability and a relatively simple & convenient programming model relatively to other accelerators.

[ Parent | on HN ]

Re:HoloLens secret sauce: A 28nm customized 24-core DSP engi... by samfisher83 on Tuesday August 23, 2016 @10:30AM (#12343984)

Cadence is an EDA company. They have a lot of other software they sell.

[ Parent | on HN ]

Re:HoloLens secret sauce: A 28nm customized 24-core DSP engi... by _yosefk on Tuesday August 23, 2016 @09:29PM (#12349043)

Precisely, and I use it, so wouldn't want Tensilica to die as a part of Cadence dying.

[ Parent | on HN ]