HiSilicon Kirin 970 NPU discussion thread

[Digit-Brand]

Participate in our Honor discussions, post your opinions, get likes on your posts and stand to win an Honor 10 worth ₹33,000 & up to ₹70,000 PayTM cash!

*lh5.googleusercontent.com/wLD_ZR8crJqlOxP54tgd8CDHpE51cMHfsgNGKQqRmNrw7JBMTmDrPFaqEW426qVlhdTxp1nX4ONvmHBuMI9CJTqPXg_C0fo9GFVKHg_EMd89t6ziCeIt__Z6qkTmyUaJt0r2CX8C​

The Honor 10 is a AI and camera-focused smartphone, and a lot of that can boil down to its chipset. The smartphone is powered by a HiSilicon’s flagship chipset, the Kirin 970 SoC that sports a dedicated NPU or Neural Processing Unit.

As the name would suggest, the NPU is used by the phone to process AI-based tasks. This includes things such as image recognition, voice-controlled personal assistant and more.

By processing all this separately from the CPU, GPU and DSP, the company says that it can deliver AI-related tasks such as scene recognition much faster. The company claims that over one billion of images have been learned in the lab to deliver 25 times better performance and 50 times higher efficiency.

The NPU in the Kirin 970 is optimised to recognise 22 specific scenes such as landscapes, food, sports and more. On top of this, it also helps with the Honor 10’s overall efficiency and power consumption.

So, what do you you think about the implementation of the NPU in the Kirin 970 chipset? Got any ideas on how to improve it? Let your thoughts be known below.

Like we said earlier, post your comments and stand to win an Honor 10 worth ₹33K and up to ₹70,000 PayTM money. For more info, check here in a new tab for more details.

 

[Nerevarine]

Staaaaaaaaaahppppppppp

You want honor to succeed, release kernel codebase, promote first party support for development, build actual community around after market development.

 

[SaiyanGoku]

Staaaaaaaaaahppppppppp

You want honor to succeed, release kernel codebase, promote first party support for development, build actual community around after market development.

No, they want to be like Apple. They want people to prefer form over functionality and show off the design of the phone irrespective of its performance or ui or updates.

 

[PrashantK007]

Would this NPU create problems if someone clicks a private pic? Would any of my pics sent to Honor for research purpose? What about videos as well?

 

[Charchit Sharma]

Would this NPU create problems if someone clicks a private pic? Would any of my pics sent to Honor for research purpose? What about videos as well?

No, the thing about the NPU is that it kind of works on the device rather than server based so it is kind of more secure.

 

[Charchit Sharma]

As said in the previous discussion thread, Last year it emerged that Kirin 970 beats Snapdragon 845 in a number of image recognition benchmarks. Honor published its own tests revealing claiming the chip performs better than the newer Snapdragon 845 as well which was 20-33 percent faster run on classified pre-trained models. Because of the NPU, its performance for ML has increased and performs better than Snapdragon DSP design.

And I think it is good to have a competition and it helps in creating a good platform, the thing is that only consumers have benefitted from there competition, just like Intel and amd. So, we should support it.

Got any ideas on how to improve it?
There’s no must-have use case for smartphone machine learning or “AI” yet. Even large percentage points gained or lost in some specific benchmarks isn’t going to make or break the main user experience. All current machine learning tasks can be done on a DSP or even a regular CPU and GPU. An NPU is just a small cog in a much larger system. Dedicated hardware can give an advantage to battery life and performance, but it’s going to be tough for consumers to notice a massive difference given their limited exposure to the applications.

But, still, improving the matrix multiplication.

 

[pankajudhas]

This npu is helping honor phones take much much best pics