中文English
1, ARM three new core: performance increase exaggerated, next year's mobile phone chip will be big 2, TSMC 2nm chip or 2025 mass production! What does that mean? 3, AI chip war sp...

Contains the term Arm or mass production of AI chips by 2025

ARM three new cores simultaneous: Performance improvement is exaggerated, next year's mobile phone chip will be big

It can be said that these changes in X2 can effectively reduce the computing cycle and time, which in turn means further enhancing the computing performance of the chip. In addition, both front and back ends, X2 changes are not small, ARM claims that X2 compared to X1 integer performance improved by 16%, machine learning performance can be doubled, in addition to X2 level 3 cache capacity of 8MB, than the previous generation of X1 doubled.

The new chip is not only in performanceWhile Mali's graphics engine has 50% better processing performance and 20% lower power consumption, which is crucial for smooth VR applications. Mobile phone manufacturers will use ARM's new technology, such as Apple, Samsung and others may adopt these chips in new phones to meet the high requirements of Google Daydream platform, such as 4K video playback, low latency and powerful processors.

Performance overview: The big core is stronger, the middle core is more province, the GPU joins the hardware light to catch up to the point, we first directly bring you the performance related information of this new generation of ARM architecture. And it is very regrettable that ARM is no longer "clever" when announcing the performance comparison data of the new and old architectures, in fact, they have given what may be the most detailed performance comparison parameters in history. First, there's the new Cortex-X3 big core.

However, ARM's survival depends on continuous innovation, and if the funding chain goes wrong, the development of new architectures may be blocked, causing the entire industry to be in trouble. At the same time, chip manufacturers will improve the infrastructure according to their own advantages, which will prompt software manufacturers to optimize and further promote technological progress. But consider market stability and profitThe conflict between Qualcomm and ARM is more likely to be resolved by raising licensing fees. The new Snapdragon 875 chipset, codenamed "Lahaina", made an amazing appearance in the Antutu score test, with a score of 847,868 points, much higher than the 700,000 previously rumored. This makes the performance of Qualcomm Snapdragon 875 a performance improvement of about 25% compared to phones equipped with Snapdragon 865.

The Kirin 9000 uses a 1*13 GHz A77 super-large coreTSMC's 2-nanometer process will reach mass production in 2025, and the market expects it to surpass rivals Samsung and Intel. TSMC's advanced process is progressing well and the 3nm process is expected to enter mass production in the second half of this year. The upgraded 3-nanometer process (N3E) will enter series production in 2023, one year after the 3-nanometer process, and the 2-nanometer process is expected to enter series production in 2025.

2, but after the 2 nanometer process, involving the original process structure, materials, equipment updates, technological innovation, means that the industrial pattern or will be reshuffled, which is also many enterprises evenOne of the reasons why countries and regions have decided to bet on 2nm.

3, and "Intel", both will be tested several times before official use. Related technical personnel said that 2nm is under development, and there is confidence that it can maintain a leading position in technology. It is well positioned to support customer growth and will be ready to start production in 2025, with pre-production ready in 2024. In the 2nm chip, TSMC will use "GAAFET" for the first time to replace the previous "finFET".

4, even the world's most advanced chip manufacturingTSMC has only achieved the production and manufacturing of Mediatek's 4nm chips, and large-scale mass production still needs time to improve the process and improve the yield. The chip manufacturing process must be step-by-step, and it is impossible to cross the process era. Just because it can't be built now doesn't mean it can't be built in the future. According to TSMC's plan, the 2nm process will be mass-produced in 2025. Prior to this, 3nm will be mass produced in the second half of 2022.

5, according to the official description, TSMC will use the surround gate technology (GAA) -based MBCFET architecture in the 2nm process, which is planned to be carried out in the second half of 2023Risky trial production, mass production in 2024. Since last year, on the key node planning of the advanced process and the external notification of relevant breakthroughs, TSMC has performed quite "actively".

AI chip war spread, who is the "core king"?

In the case of Intel, Nvidia, arm company occupied the data center and mobile phone chip market, Chinese companies are trying to break through from the edge and terminal, using China's huge manufacturing hardware industry chain and scene advantages, trying to establish their own AI coreSlice ecosystem. Domestic chip companies are ready for the arrival of the new era of AI, so that many enterprises stand on the same starting line.

Hongsoft Technology. AI vision leader, serving smart phones, smart cars, Internet of things and so on. Saint Bond Shares. AI analog chip faucet, used in speech recognition, ultrasonic ranging, infrared obstacle avoidance and so on. Huichuan Technology. The automation servo system occupies the domestic leader with 8% share. Green harmonics. AI algorithm commercial landing manufacturers: iFlytek, tower.

In the next few years, a large part of Ford, Geely, and Great Wall's assisted driving landing models may be usedIt will be the EyeQ4 chip. Here's an example:? Gm Cadillac's SuperCruise, now using the chip is still EyeQ3, until the next generation platform will use the EyeQ4 chip. So, the pace of the industry is much slower. This is also why, Nvidia is not in a hurry, until 2022 to carry out L2 chip mass production landing.

In the enterprise services market, in addition to GPU leader Nvidia, Intel's old Allies - cloud service providers Amazon, Google, Microsoft, etc., are also competing to develop their own AI chips, such as Google'sTPU, Amazon AWS Inferentia, etc. Whether Intel can maintain its chip leader status is full of challenges in the AI era.

Products include mobile communication central processor, baseband chip, AI chip, RF front-end chip, RF chip and other communication, computing and control chips. Changjiang Storage was established in July 2016 by the merger of Unigroup Guoxin and Wuhan Xinxin Company.

Intel high-profile into the circle, and Qualcomm, Nvidia isDeclaration of war

According to the official news, the RTX4090 flagship product will be announced on October 12 - and Intel's ArcA770 completely colliding, the second flagship product RTX4080 (including 16G and 12G two model specifications) is expected to be sold in November. However, under the premise of improving performance, the price of Nvidia RTX40 has also been updated.

If Qualcomm and Veoneer can really join hands, Qualcomm's Snapdragon processor can partner with Veoneer's Guardianlike software, so as to directly communicate with Mobileye EyeQ and Nvidia Drive go head-to-head. There is one problem, though. Of course, the three protagonists we want to talk about today are the three semiconductor companies that exist like giants in the traditional consumer electronics field, they are Qualcomm, Nvidia and Intel.

Xiaopeng P7 equipped with Nvidia's Xavier chip has long been listed, of course, the Great Wall high-end car will be equipped with Qualcomm's Snapdragon Ride chip, NIO's ET7 will be equipped with Nvidia's Orin chip, and will be delivered in 2022.

Recently, Intel is very troubled. First, Apple announced at the 20th WWDC that it had broken up with a good friend who had cooperated for 15 years and turned to a self-developed chip based on ARM architecture. Then, since July, Nvidia's stock price has risen all the way, replacing Intel and becoming the most valuable chip manufacturer in the United States for the first time, and it has recently been reported that Nvidia wants to acquire ARM.

Baidu AI chip Kunlun mass production early next year, support domestic Feiteng CPU?

BaiduAI chip research and development results are about to usher in a mass production climax. Samsung Electronics and Baidu joined forces to announce that the Kunlun chip research and development work has been fully completed, and plans to mass production early next year, using Samsung's advanced 14-nanometer process, the cooperation marks the first deep cooperation between the two sides in the field of chip manufacturing. Baidu, a leading Chinese-language Internet search provider, and Samsung Electronics, a global leader in advanced semiconductor technology, today announced that Baidu KUNLUN, Baidu's first cloud-to-edge AI accelerator, has completed development and will move forward to mass production next year.

In addition, the Feiteng CPU also has a high degree of flexibility and scalabilityScalability, whether it is a single machine, cluster, cloud computing and other application scenarios can be supported. The emergence of Feiteng CPU has brought new development opportunities for China's high-performance computing field.

At present, domestic Feiteng CPU mainly includes Feiteng CT-10000 series and Feiteng CT-2000 series two series. The CT-10000 series, also known as the "Tengyun" series, is a high-performance CPU product from the company for the desktop and server markets.

Tianjin Feiteng Information Technology Company. Feiteng CPU is domestic Feiteng CPU server, domestic Feiteng CPU general server for short.

UOS system is an operating system based on Linux kernel developed by Tongxin Software, which supports notebook desktop all-in-one workstation servers of domestic chip platforms such as Loongson Fiiteng Mega Core Haiguang Kunpeng. It mainly focuses on desktop application scenarios, including a number of original applications of independently developed desktop environment and rich applications.