[FoRK] Mobile chipsets, Re: FoRK Digest, Vol 131, Issue 5

Stephen D. Williams sdw at lig.net
Sun Aug 10 18:06:27 PDT 2014


On 8/10/14, 12:29 PM, J. Andrew Rogers wrote:
> On Aug 10, 2014, at 12:25 PM, Stephen D. Williams <sdw at lig.net> wrote:
>> BTW, current mobile hardware, just considering the CPU and CPU cores, is theoretically capable of well over 150 gigaops / sec at shipping clock speeds.  Many of those can be float and/or double.  That is going to go up drastically soon.  Whether a particular application can take advantage of much of that is always a key question, but many could substantially.
>
> The challenge is not the operation rate, it is feeding the operators interesting data to operate on at that rate.
>

And that's where mobile processors right now are doing well, at least in some key ways.  Your desktop computer or laptop struggles 
with a USB 2 or 3 webcam while a mobile phone has multiple cameras, each with up to a 4 lane MIPI connection feeding hardware blocks 
then DMAing into memory.

Combine large and/or multi-sensor input with high frame rates, and computational imaging, then feeding AI interpretation, and it is 
interesting.  While memory bandwidth is an important bottleneck, it is getting better and there are multiple memory buses involved. 
Memory bandwidth is the main bottleneck in a lot of cases.  But little code makes efficient use of existing hardware too.

Stephen



More information about the FoRK mailing list