Huawei kirin 970 or smarter capabilities

As artificially intelligence creeps its way inkhổng lồ our điện thoại thông minh experience, SoC vendors have sầu been racing to improve neural network & machine learning performance in their chips. Everyone has a different take on how to power these emerging use cases, but the general trkết thúc has been to include some sort of dedicated hardware lớn accelerate common machine learning tasks lượt thích image recognition. However, the hardware differences mean that chips offer varying levels of performance.

Bạn đang xem: Huawei kirin 970 or smarter capabilities


Neural Networks(NN) and Machine Learning (ML) were two of the year"s biggest buzzwords in mobile processoring. Huawei’s HiSilicon Kirin 970, the image processing unit (IPU) inside the Google Pixel 2, & Apple’s A11 Bionic, all feature dedicated hardware solutions …
Last year it emerged that HiSilicon’s Kirin 970 bested Qualcomm’s SnapLong 835 in a number of image recognition benchmarks. Honor recently published its own tests revealing claiming the chip performs better than the newer Snapdragon 845 as well.

Related: best Snaprồng 845 phones you can buy right now

We’re a little skeptical of the results when a company tests its own chips, but the benchmarks Honor used (Resnet & VGG) are commonly used pre-trained image recognition neural network algorithms, so a performance advantage isn’t to lớn be sniffed at. The company claims up khổng lồ a twelve-fold boost using its HiAI SDK versus the Snaprồng NPE. Two of the more popular results show between a 20 & 33 percent boost.


Regardless of the exact results, this raises a rather interesting question about the nature of neural network processing on điện thoại thông minh SoCs. What causes the performance difference between two chips with similar machine learning applications?

DSPhường vs NPU approaches

The big difference between Kirin 970 vs SnapLong 845 is HiSilicon’s option implements a Neural Processing Unit designed specifically for quickly processing certain machine learning tasks. Meanwhile, Qualcomm repurposed its existing Hexagon DSPhường design to lớn crunch numbers for machine learning tasks, rather than adding in extra silicon specifically for these tasks.

With the Snapdragon 845, Qualcomm boasts up lớn tripled performance for some AI tasks over the 835. To accelerate machine learning on its DSPhường, Qualcomilimet uses its Hexagon Vector Extensions (HVX) which speeds up 8-bit vector math commonly used by machine learning tasks. The 845 also boasts a new micro-architecture that doubles 8-bit performance over the previous generation. Qualcomm’s Hexagon DSP. is an efficient math crunching machine, but it’s still fundamentally designed lớn handle a wide range of math tasks và has been gradually tweaked khổng lồ boost image recognition use cases.

The Kirin 970 also includes a DSP (a Cadence Tensilica Vision P6) for audio, camera image, & other processing. It’s in roughly the same league as Qualcomm’s Hexagon DSPhường, but it is not currently exposed through the HiAI SDK for use with third-party machine learning applications.


The Hexagon 680 DSPhường from the SnapLong 835 is a multi-threaded scalar math processor. It’s a different take compared to mass matrix multiple processors for Google or Huawei.

HiSilicon’s NPU is highly optimized for machine learning and image recognition, but is not any good for regular DSPhường. tasks like audio EQ filters. The NPU is a bespoke chip designed in collaboration with Cambribé Technology & primarily built around multiple matrix multiply units.

You might recognize this as the same approach that Google took with its hugely powerful Cloud TPUs and Pixel Vi xử lý Core machine learning chips. Huawei’s NPU isn’t as huge or powerful as Google’s VPS chips, opting for a small number of 3 x 3 matrix multiple units, rather than Google’s large 128 x 128 kiến thiết. Google also optimized for 8-bit math while Huawei focused on 16-bit floating point.

The performance differences come down to lớn architecture choices between more general DSPs and dedicated matrix multiply hardware.

Xem thêm: Download Lmht Tốc Chiến Apk + Obb 1, Tải Tốc Chiến Apk

The key takeaway here is Huawei’s NPU is designed for a very small phối of tasks, mostly related lớn image recognition, but it can crunch through the numbers very quickly — allegedly up lớn 2,000 images per second. Qualcomm’s approach is khổng lồ tư vấn these math operations using a more conventional DSP, which is more flexible and saves on silinhỏ space, but won’t quite reach the same peak potential. Both companies are also big on the heterogeneous approach lớn efficient processing và have dedicated engines to manage tasks across the CPU, GPU, DSPhường., và in Huawei’s case its NPU too, for maximum efficiency.


Qualcomilimet sits on the fence

So why is Qualcomilimet, a high-performance mobile application processor company, taking a different approach lớn HiSilicon, Google, và Apple for its machine learning hardware? The immediate answer is probably that there just isn’t a meaningful difference between the approaches at this stage.

Sure, the benchmarks might express different capabilities, but the truth there isn’t a must-have sầu application for machine learning in smartphones right now. Image recognition is moderately useful for organizing pholớn libraries, optimizing camera performance, và unlocking a phone with your face. If these can be done fast enough on a DSPhường, CPU, or GPU already, it seems there’s little reason lớn spend extra money on dedicated silicon. LG is even doing real-time camera scene detection using a Snapdragon 835, which is very similar to lớn Huawei’s camera AI software using its NPU và DSP..

Qualcomm"s DSPhường is widely used by third-parties, making it easier for them to lớn start implementing machine learning on its platform.

In the future, we may see the need for more powerful or dedicated machine learning hardware lớn power more advanced features or save sầu battery life, but at the moment the use cases are limited. Huawei might change its NPU thiết kế as the requirements of machine learning applications change, which could mean wasted resources & an awkward decision about whether lớn continue supporting outdated hardware. An NPU is also yet another bit of hardware third-tiệc ngọt developers have to lớn decide whether or not khổng lồ support.


Baông xã at the start of 2017, Arm announced its first batch of dedicated machine learning (ML) hardware. Under the name Project Trillium, the company unveiled a dedicated ML processor for products lượt thích smartphones, along with a …
Qualcomm also has a history of dismissing novel or nibít ideas only lớn quickly adopt similar technologies of its own once the market moves in that direction. Cast your minds baông chồng to lớn the company dismissing 64-bit Smartphone application processors as a gimmick.

Qualcomilimet may well go down the dedicated neural network processor route in the future, but only if the use cases make the investment worthwhile. Arm’s recently announced Project Trillium hardware is certainly a possible candidate if the company doesn’t want lớn thiết kế a dedicated unit in-house from scratch, but we’ll just have sầu to wait & see.


Does it really matter?

When it comes to Kirin 970 vs Snapdragon 845, the Kirin’s NPU might have sầu an edge, but does it really matter that much?

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

Xem thêm: Tăng Tiết Mồ Hôi Tay Chân: Cách Chữa Trị Hiệu Quả, Cách Khắc Phục Chứng Ra Mồ Hôi Tay


Neural Networks và Machine Learning are some of this year"s biggest buzzwords in the world of smartphone processors. Huawei"s HiSilicon Kirin 970, Apple"s A11 Bionic, and the image processing unit (IPU) inside the Google Pixel …
As the machine learning market place evolves và more applications break through, smartphones with dedicated hardware will probably benefit — potentially they’re a bit more future proofed (unless the hardware requirements change). Industry-wide adoption appears to lớn be inevitable, what with MediaTek & Qualcomm both touting machine learning capabilities in lower cost chips, but it’s unlikely the speed of an onboard NPU or DSPhường is ever going to lớn be the make or break factor in a smartphone purchase.

Chuyên mục: Tin Tức