On the eve of the release of the Mate30 series, Huawei first announced the details of the latest processing chip Kirin 990 at IFA.

In addition to the upgrade of performance and power saving capabilities, it also includes AI computing capabilities and 5G introduction. The most important feature is that the Kirin 990 should be the first mass-produced and officially applied 5G mobile phone processor that integrates 5G data chips. (Although Samsung has also released the news of the Exynos 980, an integrated 5G chip, it will take next year to use it in products.)

This time, there are actually two Kirin 990s. The Kirin 990 5G has an integrated 5G chip, and the Kirin 990 is an LTE-supporting SoC. Others, such as the highest CPU clock and NPU architecture, are also slightly different. This is a simple comparison.

Kirin 990 can be viewed in three aspects: 5G, AI and performance.

Integrated 5G

For the 5G mobile phones currently launched, most of the BR32 is a processor group with a 5G data chip attached, such as Qualcomm S855+X50, or a combination of Exynos 9825+Exynos Modem 5100, while Kirin 990 5G is an All-in-One mobile phone chip .

The advantage of integrating a 5G chip is that it can reduce the size of the chip space. For example, it was mentioned on the stage that the Kirin 990 5G can reduce the main board area by 36% compared to the plug-in solution.

But there is doubt that it will increase the power consumption of the mobile phone.

In the future, or 5G that has already been launched, the challenges include weak signal a little far from the base station, resulting in power consumption of the device, and instability of the signal while moving and unable to exert speed.

Kirin 990 5G also proposes some solutions for these challenges. For example, in places where the 5G signal is weak, it adopts a smart upstream split design, which increases the upload speed by 5.8 times compared to the X50 solution.

In addition, Kirin 990 5G supports partial bandwidth technology (Bandwidth Part/BWP), which allows broadband devices to save power. The data of Kirin 990 is that after BWP is enabled, the power consumption of light loading is 44% less than that of Qualcomm S855+X50 solution. %.

In the mobile scene, through the adaptive reception of machine learning, the download speed is 19% faster than that of the X50 at a mobile speed of 120 kilometers per hour.

These supports enable the Kirin 990 5G to perform better in overall power consumption than Qualcomm and Samsung's existing solutions under high-speed transmission.

Kirin 990 5G also supports 5G+4G dual-card, SIM1 can receive 5G, SIM2 can receive 4G, and both can use VoLTE to make calls.

Using a 5G network to download large files is very fast. According to Huawei's data, the Kirin 990 5G has a maximum download speed (peak) of 2.3Gbps and a peak upload speed of 1.2Gbps.

I: large core + small core NPU

The Kirin 990 adopts the Da Vinci architecture NPU developed by Huawei itself. The difference is that it adopts a large-core + micro-core architecture for the first time. The large core is responsible for more complex operations, while the micro-core runs some AI operations that can use lower power (such as human faces). Recognition), component work, can maintain the exception of AI calculation performance, and also save power when some AI functions are operated.

What are the achievements of the NPU architecture with small and large cores?

According to Huawei's data, the performance is 6 times higher than that of competing products (A12 and S855), and the power used under the same operation is 8 times less. Using the micro-core NPU for face recognition, the power consumption can be reduced by 24 times compared with the large-core recognition.

The AI computing power of Kirin 990 can also develop more advanced functions. The scene mentioned a technology that can realize multi-instance segmentation. For example, when recording a multi-person video, you can distinguish and isolate multiple characters for special effects, such as changing the background of the video in real time, or deleting or moving some of the characters.

This is an official demonstration video. Even if the characters are in motion, they can be isolated and captured for various special effects.

Efficiency: 2+2+4 Cortex A76 architecture + 16-core G76 GPU

The Kirin 990 uses a 7nm+ EUV process, which allows the chipset to contain 10.3 billion transistors, increasing the processing speed.

This time, the Kirin 990 has two CPU architectures, two middle ones and four small ones, including two Cortex-A76@2.86GHz, two Cortex-A76@2.36GHz, and four Cortex-A55@1.95GHz. (Well, not the Cortex-A77 that everyone guessed earlier). The GPU is a 16-core Mali-G76.

Upgraded ISP 5.0

In terms of photography, the Kirin 990 5G adopts the new ISP 5.0, and implements the hardware noise reduction technology of BM3D (Block-Matching and 3D filtering) monocular camera on the mobile phone chip for the first time, which can make the low-light scene photography brighter and clearer.

In the test data, the single-core performance is 10% higher than that of the S855, and the multi-core performance is also 9% higher. The performance of large, medium and small cores also wins. The GPU performance is improved by 6%, but the power consumption is reduced by 20%. In the game, the Kirin 990 5G also uses less power than the S855 when running at 60fps.

Upgraded ISP 5.0

In terms of photography, the Kirin 990 5G adopts the new ISP 5.0, and implements the hardware noise reduction technology of BM3D (Block-Matching and 3D filtering) monocular camera on the mobile phone chip for the first time, which can make the low-light scene photography brighter and clearer.

In addition, the dual-domain film noise reduction technology is also implemented for the first time, which makes the film noise processing more accurate, and the low-light film has better noise reduction image quality; based on AI segmentation real-time video post-processing rendering technology, the color of the film screen is adjusted frame by frame, allowing the mobile phone to The short film has a cinematic feel.

ISP performance and AI algorithm recognition can bring a function of Face AR measurement of heart rate. The camera can directly capture the information of blood flow in the capillary of the face and convert it into heartbeat data. Generally, the face is relatively clean and there are not too many shaded parts, so choose to sweep the face to measure.

The first launch of the Kirin 990 is of course the Mate30 series that will be released on 9/19. So we can look forward to the power consumption of Mate30 and what upgrades AI computing brings to video recording.