Home > News content

The nightmare of Arm Mali GPU: Samsung and Huawei have turned to self-research!

via:博客园     time:2019/8/9 8:31:56     readed:2762


Currently in the desktop CPU market, it is the world of Intel and AMD. On the desktop level of GPUs, especially the discrete graphics market, it is dominated by Nvidia and AMD. The data shows that Nvidia occupies 66.3% of the discrete graphics market, and AMD's share is 33.7%. Although Intel is preparing to launch a self-developed discrete graphics card, it is not easy to compete with Nvidia and AMD.

In the mobile CPU market, Arm's Cortex family of CPUs is unique, and other MIPS, RISC-V, and x86 architectures are difficult to compete with. However, in the mobile GPU market, Arm's Mali series GPUs are not strong.

Although, after Apple announced in 2017 that it would abandon Imagination Technologies' GPU technology within two years, Arm's Mali series of GPUs began to lose a strong competitor. Moreover, the surge in shipments of Huawei mobile phone chips using the Arm Mali series of GPUs in recent years has also helped Arm to advance its position in the mobile GPU market. However, Qualcomm's Adreno GPU still occupies a significant portion of the market share (Qualcomm's Adreno GPU is derived from ATI's imageon, and AMD's acquisition of ATI has sold mobile device assets to Qualcomm).

Data from unity shows that as of July 2016, the mobile GPU market (including Android, iOS, WP devices) market rankings are: Arm (35.9%), Qualcomm (32.4%), Apple (16.0%), Imagination (11.2) %), Vivante (1.9%), Broadcom (1.1%), NVIDIA (0.7%), Intel: (0.4%).


In recent years, with the further upgrade of the mobile market competition represented by smart phones, in order to enhance the competitiveness of their mobile phone products, Samsung, Apple, Huawei and other headphone smartphone manufacturers have increased their investment after building their own mobile phone SoCs. Self-developed core devices such as CPU, GPU, NPU and baseband chips.

In June of this year, Samsung and AMD reached a cooperation agreement, Samsung will use AMD's latest RDNA architecture GPU to design its own mobile GPU core. In addition, since the “ZTE event” and the US sanctions against Huawei, Huawei has further increased its investment in self-developed core chips. Recently, there have been rumors that Huawei is developing independent GPU chips in addition to the self-developed CPU. This will undoubtedly pose a threat to Arm's position in the mobile GPU market.

apple:CPU/GPU/baseband will be fully self-developed

The iPhone 4 is a very important product in the development of Apple's mobile phone, because Apple has adopted its own A series processor for the first time since the iPhone 4. Prior to this, all third-party processors were used. The first generation of Apple A-series processors was based on Arm's public Coretx-A8 core, and subsequent generations were still Arm-based Cortex cores. Until September 2013, Apple released its first self-developed 64-bit ARM v8 dual-core processor — — Apple A7, which is the world's first 64-bit processor. After that, Apple's A-series processors have been using self-developed CPUs.

On the GPU side, since the launch of Apple's A-series processors, Imagination's PowerVR family of GPUs has been used, making it the largest customer of Imagination. Until April 2017, Apple announced that it would abandon all of Imagination's technology, including patents, intellectual property, confidential information, and so on, within two years. Apple made the decision because of a breakthrough in its self-developed GPU.

In fact, as early as 2013, Apple has begun to develop independent GPUs. Apple then convened a number of GPU senior talents to form a team to design the GPU, and AMD's former graphic design chief technology officer Raja Koduri was among them. However, at the time, Apple did not go well in the development of its own GPU. This can be seen from the return of Raja Koduri to AMD. As a result, Apple renewed its multi-year licensing agreement with Imagination in 2014, covering all of Imagination's then and future years of PowerVR graphics and video technology patents. However, Apple has not given up on the development of its own GPU.

In the first half of 2016, Apple further expanded its GPU R&D department and recruited GPU R&D talents. At the same time, it is constantly digging people from the partner Imagination. Perhaps because Apple is its biggest customer, Imagination seems to be “something to dare to say” for Apple's digging.

Although Apple introduced the A11 in September 2017, the media said that its integrated GPU is self-developed by Apple, but in fact, it still seems to be based on Imagination's IP customization. Because its kernel design is very similar to the previous Imagination Rogue. It is worth mentioning that Apple also integrated the self-developed NPU core for the first time on the A11 processor.


Image credit: TechInsights Apple A12 Die Shot, ChipRebel Apple A11 Die Shot

Apple's A12 GPU, introduced in 2018, still supports PVRTC (PowerVR Texture Compression), a proprietary format that means its GPU may still be associated with Imagination's IP.

In conjunction with Apple's announcement in 2017, "all technologies that will use Imagination will be abandoned in the next two years", Apple should use self-developed GPUs on the A13 processor released in September this year.

In addition, it should be pointed out that Apple has been developing its own baseband chips in recent years. Especially after spending RMB 1 billion to acquire the Intel baseband chip business this year, it is expected that the research and development process of Apple's self-developed mobile phone baseband chips will be further accelerated. It will be commercially available in the next two years.

Samsung:AMD is expected to launch a self-developed GPU within two years

In 2015, Samsung released the Exynos 8890 eight-core processor with 14nm FinFET technology in South Korea. The special feature of this processor is that its CPU adopts four "mezzle" architecture based on the Arm v8 instruction set. The large core, and the small core of the four Arm Cortex-A53. After that, Samsung's flagship processor began to continue to upgrade the self-developed "cats" CPU architecture.

After several years of speech, Samsung's self-developed CPU architecture has reached the fourth generation. In 2018, Samsung's Exynos 9820 processor was made up of two fourth-generation self-developed cores, two Cortex-A75s and four Cortex-A55s. It also added the self-developed NPU for the first time.


It is worth mentioning that in August 2018, Samsung also officially released the first self-developed 5G baseband chip Exynos Modem 5100. The Korean version of the Samsung Galaxy S10 5G version released earlier this year uses the Exynos Modem 5100 baseband.

At the same time as the full rise of Samsung's self-developed CPU, NPU, and baseband chips, Samsung has already begun the development of self-developed GPUs. As early as 2017, there was news in the industry that Samsung was developing a self-developed GPU. In 2018, Samsung dug up Nvidia veteran, GPU technology expert Dr. Chien-Ping Lu, who will lead the GPU R&D team.

According to the data, Lu Jianping participated in the development of nForce chipset integrated with GPU in NVIDIA, and he led the IGP integrated graphics chipset. It is worth mentioning that after NVIDIA, Dr. Lu Jianping joined MediaTek to lead MediaTek's self-developed GPU program, but this self-developed GPU was later cancelled for various reasons.

Perhaps the development of self-developed GPUs has suffered setbacks. After all, the GPU market is quite mature. If you are completely independent, you need to avoid the patents of other GPU vendors. This is not easy. As a result, Samsung began to seek cooperation with GPU manufacturer AMD. In June of this year, Samsung officially announced a cooperation agreement with AMD. AMD licensed the latest RDNA architecture GPU to Samsung, allowing Samsung to design its own GPU core. It is worth noting that one month before the two sides reached cooperation, Lu Jianping left from Samsung.


According to outside estimates, Samsung will launch mobile GPUs integrating AMD Radeon graphics processing technology in the next two years, and the new mobile GPUs will be expected to significantly enhance the graphics processing capabilities of Samsung mobile phones.

Huawei:In addition to the self-developed mobile phone SoC / baseband / NPU, self-developed CPU / GPU is also in progress?

Huawei has always insisted on independent research and development on many core devices. In addition to Huawei's recent exposure to high-definition smart TVs, the Huawei HiSilicon chip family has been Kirin, Barong, Kunpeng, and Shengteng. , Tianzhu, Ling Hao and other product lines.

As early as 2004, Huawei established HiSilicon, which is engaged in research and development of related mobile phone chips. By 2009, Huawei's first mobile phone processor K3V1 was officially released, but this chip is mainly for the low-end market, and it was not successful at the time. Then Huawei launched the quad-core Coretex-A9 processor K3V2 in 2012, which received high attention in the market. Huawei also carried this processor on the flagship Ascend D at that time and achieved certain results. . However, in the past two years, Huawei has not launched a new chip, and it has been relying on the K3V2 to hit the market. Therefore, it has also been rumored by the outside world, "Wan K3V2".

In fact, during this period of time, Huawei made in-depth improvements to the problems that occurred on K3V2. In 2014, the first “Kylin” processor & mdash;— Kirin 910 was successfully launched. Although the Kirin 910 is only an upgraded version of the K3V2, it solves some of the problems of the previous K3V2, and also integrates the self-developed Balong 710 LTE multimode baseband chip for the first time. Prior to this, Huawei's mobile phone chips were all external basebands.

In fact, while Huawei began to develop mobile phone processors, Huawei also started the development of baseband chips. In September 2008, Huawei officially established the LTE UE development department to launch the development of LTE chips. In early 2010, Huawei successfully launched the industry's first Balong 700 chip supporting LTE FDD and TD-LTE dual mode. In 2012, Huawei released the industry's first multimode LTE terminal chip Baron 710 supporting LTE Cat.4, which was successfully integrated into the Kirin 910 series processors.

In the following years, with the rapid outbreak of the smartphone market, Huawei is also a rapid upgrade iteration on the Kirin processor and the Baron baseband chip. In September 2017, the successful release of the Kirin 970 based on the 10nm process is a milestone for the Kirin chip. Its overall level has reached the same level as the Qualcomm Xiaolong flagship processor for the first time, and even has advantages in many aspects. For example, the Kirin 970 integrates the NPU core for the first time, and its integrated Baron baseband supports LET Cat.18 (4.5G, Pre 5G), supports 5-carrier aggregation, 4× 4 MIMO and 256QAM, which can maximize data transmission efficiency. It surpassed the Gigabit baseband chip Snapdragon X16 equipped with Qualcomm Snapdragon 835, which achieved the same 1.2 Gbps download rate as the Snapdragon X20 released by Qualcomm at that time.

Then in February 2018, Huawei first released its first 5G commercial chip & mdash; — Barong 5G01 (Balong 5G01) and 5G commercial terminal & mdash; & mdash; Huawei 5G CPE. At the beginning of this year, Huawei also pre-empted Qualcomm to release the first SA/NSA dual-mode 5G baseband chip Baron 5000.

It is also worth mentioning that the integrated NPUs of Huawei Kirin 970 and Kirin 980 are based on the IP of the NPU of Cambrian Technology. In July this year, the Kirin 810 processor released by Huawei was the first to integrate Huawei's self-developed new The DaVinci architecture's NPU, and its powerful AI performance, made the Kirin 810 a leader in the rankings of the AI-Benchmark.


It can be said that the Huawei-designed Kirin processor, the self-developed NPU and the Baron baseband chip have reached the top level in the world. However, it should be pointed out that the CPU core and GPU core of Huawei's Kirin processor are all from Arm. In May of this year, the United States listed Huawei as an export control "list of entities", after which Arm interrupted cooperation with Huawei, which also means that Huawei's development of new mobile chips in the future may not be possible due to the lack of Arm latest. IP licensing will encounter difficulties.

Fortunately, Huawei has previously obtained a permanent license for the ARMv8 architecture. That is to say, Huawei does not need the IP cores that have been designed by Arm, but directly obtains the instruction set authorization for designing these IP cores, so that Huawei can IP. The kernel has undergone a major transformation and can even extend or shrink the Arm instruction set. The current mainstream Arm mobile processor CPU core is basically based on the ARMv8 instruction set design.


Therefore, Huawei can design its own processor based on the ARMv8 instruction set and has complete intellectual property rights, which is not subject to the US ban. For example, Apple's A-series processors, some of Qualcomm's high-end flagship processors, and Samsung's meerkat-based CPUs are all designed independently based on the Arm instruction set. In addition, in January of this year, Huawei released the high-performance server processor —— 鲲鹏 920 is Huawei's self-developed based on the ARMv8 instruction set.

Therefore, Core Intelligence (ID: icsmart) judged that Huawei will inevitably design its own mobile phone CPU core based on the ARMv8 instruction set. Although Huawei can also choose the open source RISC-V architecture to design autonomous CPU cores, the performance RISC-V architecture may not be able to compete with Arm in the mobile phone market, and there may be problems in software compatibility. More suitable for the Internet of Things market. Therefore, Huawei has a high probability to design an autonomous mobile phone CPU core based on the ARMv8 instruction set.

Recently, according to Taiwan media reports, relevant supply chain sources pointed out that “His is currently developing a variety of chips, from a series of chips used in mobile devices to multimedia display chips and CPUs and GPUs used in computers. Moreover, the technologies used by the HiSilicon chips are all concentrated on TSMC's advanced process technology below 7nm, while at the same time, the capacity of the downstream packaging and testing plant in Taiwan and the downstream PCB industry are packaged.

For this news, Core Intelligence (ID: icsmart) believes that in addition to designing autonomous mobile CPUs, Huawei may indeed design its own GPUs. Because of the US ban, Huawei will not be able to use Arm's new Mali GPU. In addition, for smartphones, the performance of the GPU directly affects the screen display and picture fluency of the mobile phone, especially in the large-scale games, AR/VR experience, and even affects the capabilities of the mobile phone AI (GPU AI). The computing power is much higher than the CPU). Therefore, in order to further enhance the competitiveness of mobile phone processors, self-developed GPU is also inevitable.

Therefore, we can see that the first-line headphone brands such as Apple and Samsung are developing their own GPU cores. Qualcomm's flagship processor is popular with mobile phone brands, and one of the key factors is the outstanding performance of its self-developed Adreno GPU.

However, developing your own GPU is not an easy task. Since its inception in 2013, Apple has not been commercially available yet, although it may be close to success, but it is based on its long-term cooperation with Imagination and the hiring of a large number of GPU talents. Samsung has been developing for several years and has suffered setbacks. Finally, it chose to cooperate with AMD. It is not difficult to see that the difficulty of GPU development is high. It is also not easy for Huawei to launch a self-developed GPU.

However, Core Intelligence believes that Huawei can actually choose to cooperate with Imagination to start GPU joint customization. After all, in September 2017, mobile GPU maker Imagination was successfully acquired by the Chinese-language private equity fund, Canyon Bridge, for £550 million. Imagination is now a Chinese-funded company and is also aggressively exploring the domestic market.

Previously, Liu Guojun, general manager of Imagination China, said in an interview with the media that the Chinese market accounted for about 10% of Imagination's total revenue. The first task was to capture the market share of the entire China region.

Core Intelligence believes that Huawei can even consider buying Imagination directly from Canyon Bridge. Of course, this stage is unlikely. After all, the US ban exists, and Huawei's acquisition of Imagination may not be approved by US regulators.

In addition, the previous news said that Huawei wants to design the CPU and GPU used by the computer, which is probably nonsense. Everyone knows that the current CPU/GPU competition in the computer market is extremely mature. The remaining players have long monopolized the market for many years (as mentioned in the opening article), and the ecology is already very mature. Huawei is involved. There is no chance at all. Of course, similar to the Arm-based Win10 notebook based on the Arm architecture, there are some opportunities (ie, the self-developed CPU based on the ARM v8 instruction set and the Windows notebook based on the Kirin processor of the self-developed GPU), but the current feedback of the Snapdragon Win10 notebook in the market. In general, and currently Microsoft is only working with a Qualcomm chip manufacturer.


From the perspective of the smart phone market, Samsung, Apple, Huawei and other head-end smartphone manufacturers have chosen to develop their own CPU, GPU and NPU, which will inevitably affect Arm's authorized business, even if these CPUs are still based on The ARM v8 instruction set is licensed, but it is a one-time payment authorization. The GPU does not need to say, if Samsung and Huawei are turning to self-developed GPUs, it will undoubtedly cause a heavy blow to Arm's GPU licensing business, because only the remaining mobile phone chip manufacturers that can contribute to the Arm GPU licensing business are MediaTek and Violet. The exhibition is sharp, and the Violet Show also uses the Imagination GPU.


Image source: Counterpoint

Earlier, Counterpoint analyst Parv Sharma said, “Because of the US ban on Huawei, Arm and Huawei and its subsidiary HiSilicon will be affected, so Arm’s market share in mobile GPU will drop significantly, and it is expected in the next few years. The share may fall below 30%. ”

From the perspective of the Internet of Things market, Arm's CPU licensing business is being challenged by RISC-V. Although the impact is still limited, the ecology of RISC-V is currently growing rapidly, and many first-line chip designers are entering the market. As a result, Arm is also beginning to be forced to respond by downgrading the licensing fee and launching a new form of licensing.

In addition, with the rise of the AI ​​market, many AI chips have sprung up. Samsung, Apple, Qualcomm and Huawei have all begun to integrate the self-developed NPU core in their mobile phone chips. Huawei even launched the self-developed AI cloud chip. This also makes Arm's related AI core IP suddenly lack the support of the head phone manufacturer.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments