Qualcomm Explains How Snapdragon 820 Chips Improve Low-Light Photography

Not open for further replies.


Sep 26, 2013
I don't see any explanation anywhere around in this article?

Camera pixel size & IPS are that what they are.
QDSP 680 & SIMD block (Spectra) are improvement concerning QC design but doesn't represent leading solution in any way + they are closed source non licensable nor customizable solution. There are licensable solutions that are better in many ways that integrate SIMD block way back.
The usage for various tasks including lo light enchantment depends on quality of algorithms that are used for processing.
First to have a better efficiency along with speed programmable or semi programmable adjustable set of instructions is needed so that can be adapted to new & better optimized algorithm along with optimized to hardware architecture enchantment. QDSP's lack this ability.
Second part is how good, fast and wide is the external DRAM communication as it's not how strong is the animal but how good you feed it. All speaks about better hyperwissor, internal DRAM coherence & better QoS, trading are for litle children. Certainly all of those will help but won't resolve problem of a bigger task that simply can't fit in internal cache. This is another thing QC is far behind competition.
Third & most important part is easy to program DSP. For this particular purpose it's needed that Open imagining & vision standards are supposed & well backed up by large community. In the case of QC this brings us back to first part that design is not customizable, scalable nor it have a good documentation along with software (including open standards) support so no one serious wouldn't work on it. Along with it QC isn't known as someone who is keen backing up Open source community nor standards, they never whare able to push any of their hardware design specifics in larger scale optimized software nor they will ever be able to do it judging by the way how they work. All do to be fair there are some projects from QCDC recently that are interesting but I don't hear much about how they are progressing.

At the end just to mention how final result also much depends on how good & up to date are implemented software solutions by OEM's & do they care & work on improving & implementing new & better stuff.

I find both Ceva & Tensilica licensable solutions better. In comparison to QDSP 680 (+Spectra) Cevas MX4 & Tensilica Vision P5.
Why? Because of all 3 thing I wrote + as they are licensable IP's witch gives them much wider possibility of use.


The explanation "is" in the video. It is just a automatic image manipulation that reduce the noise. I still trust more on big sensors and good objects and physical stabisator...
Not open for further replies.