OV Xiaomi core are selected flagship machine debut, OPPO Mariana X why more exciting?

OPPO has officially launched its first self-developed chip, The OPPO MariSilicon X (Mariana X), during the OPPO Future Technology Conference 2021 in the middle of this month. We are excited not only about its leading performance, but also about the upcoming first quarter of 2022. The Mariana X chip will be integrated into OPPO's new Find X series flagship devices, promising an even better user experience.

Domestic core heat continues to rise, Xiaomi Vivo performance?

In recent years, Huawei is undoubtedly the benchmark in the way of making core for domestic mobile phone brands. However, friends who are familiar with Huawei Kirin chip should know that Kirin SoC is not the peak of its debut, but gradually ranks among the TOP in the industry through continuous learning and catch-up. As for Xiaomi, the development of its Thepaperchip has been more tortuous. Thepapersoc has been absent for several years after its mid - and low-end models failed to test the water.

It was not until the first half of this year that the Paper C1, a video chip, returned to the public. Although it was installed in Xiaomi's ultra-high-end flagship MIX Fold, the paper C1 was not featured in video. Considering the reputation of the device after its launch, the paper C1 had a low presence. According to the official introduction, thepaper.cn C1 mainly focuses on AF autofocus, AWB white balance and AE exposure optimization.

Coincidentally, this year Vivo also made an image chip vivo V1, which is also used in its current flagship vivo X70 series. It is reported that this is the result of 24 months of in-depth cooperation between Vivo and mobile SoC manufacturers, involving more than 300 people. According to vivo's official description, the chip works with the main chip, not replacing the ISP of the main chip, but is a hardware-level overlay computing imaging algorithm. Compared with the software algorithm, the advantage is that the power consumption of the hardware circuit can be reduced by 50% when processing the same amount of computing tasks. Intriguingly, when the media interviewed the relevant personage of Vivo chip, the response was that Vivo V1 is not equal to ISP. Shortly after vivo V1 came out, there was a lot of comparison between the Paper C1 and Vivo V1 on the Internet, and even whether the two chips are really self-developed became a hot topic of discussion. But at present more consistent point of view is that the two domestic chips are mainly "auxiliary", can bring or high efficiency, or save electricity, or improve the performance of the related work of the image signal processor.

The ever more powerful OPPO Mariana X, more "transparent" self-developed chips

When it comes to the new OPPO Mariana X chip, OPPO has been generous in announcing that it is an NPU chip, or embedded neural network processor, made using TSMC's 6nm process, rather than being secretive about the details of some of its own chips. NPU first came into the public's attention thanks to Huawei Kirin. At that time, Kirin 970 was the first mobile phone in the world with BUILT-IN NPU. Soon after, Apple and Qualcomm added similar modules to SoC, and the application of AI in smart phones entered the fast lane. OPPO Mariana X directly selects apple A15 chip, the industry benchmark, for comparison. The AI computing power of OPPO prototype equipped with Mariana X reaches 18TOPS (18 trillion times), surpassing iPhone 13 Pro Max 15.8TOPS. The 11.6TOPS/W is also industry leading in energy efficiency, with OPPO even using the rare term "ever more powerful" to describe the MARIana X's AI computing energy efficiency. Ultra-challenging, multiple key indicators surpass the industry's top soCs

In fact, the above two key metrics are only part of the leading results of the Mariana X. In order to avoid "fatigue" by enumerating too many data in one go, here are some stories behind OPPO's development of the Mariana X. Jiang Bo, senior director of OPPO chip products, revealed that OPPO started to make this NPU chip in 2019, and the IP design was self-developed. The whole front-end and back-end of the chip was self-developed, including the software and hardware, and it was a successful stream. All these are very big challenges. "If you do a mediocre product, simple to do a chip, it is relatively easy, but the Mariana X entry point is a very steep curve, the standard of the target is very difficult at the beginning, some characteristics of the industry can not see, can only explore." Jiang Bo stressed many times when communicating with the media that the Mariana X team adheres to the concept of pursuing perfection, from the first day of the team establishment to every person who comes in. He believes that making chips requires investment of time, money and, more importantly, talent and team. Mariana is the deepest trench in the world, and the original office of OPPO NPU chip team is called Mariana, which is to explore the deep water of technological innovation.

OPPO Mariana X currently supports 20bit lossless real-time RAW computing, which even the industry's top SoC chips (mostly 18bit) cannot achieve. With the Help of the Mariana X, the terminal also delivers ultra-high specification computational photography capabilities, including 4K Ultra HDR video with a maximum light ratio of up to 1 million to 1 picture, 4K AI night view video (an industry first), and more.

What else can the Mariana X do to surpass the giants in its debut?

Speaking of this, some netizens may have a question. In fact, the comprehensive shooting effect of the flagship machine in the market is quite good. Is the capability of OPPO Mariana X excessive or unnecessary? In fact, I have some personal experience in this problem. Take the OPPO Find X3 Pro I have used for more than half a year as an example. This is the photography flagship of the listed models of OPPO. But for ultra-low-light scenarios, the flagship doesn't offer a WYSIWYG experience, even though it supports polar night. In addition, the overall performance of shooting video, to be honest, or more or less with the industry benchmark iPhone some gap.

With the Mariana X, all of these deficiencies are significantly improved, especially when it comes to bridging the gap between photography and video. OPPO used a prototype powered by the Mariana X chip to compare it to the Find X3 Pro. The 4K AI improved processing efficiency by up to 20 times and reduced power consumption by more than 50 percent, which is quite impressive.

When it comes to NPU, there is room for all ai-related application scenarios in theory. However, Jiang bo said that Mariana X is a dedicated NPU exclusively focused on image processing, and no other scenarios are considered at present. To be clear, it is not that this chip can not be used for other applications, but at this stage, focused image, dedicated core, which is exactly the difference between the MARIana X and SoC manufacturers built-in NPU. SoC manufacturers are more concerned with general NPU, which is a loose design in SoC, so it is difficult to focus on a certain application scenario, and the capability may not be the best play. OPPO, as a terminal manufacturer, will better understand the pain points of users and the value points of user scenarios.

Aiming at the image processing scene, Mariana X should make more suitable hardware architecture design and strive to improve the corresponding energy efficiency ratio. As for the computing power, Jiang bo said that it is easy for mobile phone NPU to increase the maximum computing power. It is good to gradually increase the chip area, but it is a great challenge to achieve such a high energy efficiency of the chip. The 11.6TOPS/W energy efficiency claimed by Mariana X is actually achieved under very specific and rigorous testing conditions, and is even higher in everyday scenarios. Compared with Xiaomi Paper C1 and Vivo V1, OPPO Mariana X chip obviously has more dry products. It has completed OPPO image vertical link integration, which can maximize the sensor capability. The Mariana X delivers industry-leading computational imaging performance through a remarkable set of metrics, a key breakthrough for OPPO Imaging at a time when industry leaders are recognizing the value of computational photography.