Guide: The Jishu community and the E-learning platform jointly launched the Jishu newsletter, introducing high-quality content from industry media, technical communities, and consulting agencies, and regularly sharing industrial technology trends and market application hotspots.
core direction
Where is the PoC point for Arm Neoverse N2 and CMN-700 systems?
Point of Coherency (PoC) is located: ensure that all agents that can access memory in the system can see the same copy when accessing this address, no matter what its access memory type (memory type) and cache attribute (cacheable or non-cacheable) are . On many systems, it resides in the system's main memory. This paper analyzes the location of PoC points in a N2+CMN-700 system. (Source: Arm CPU Architecture Column of Jishu Community )
This article introduces FunASR, an open-source speech recognition toolkit from Alibaba DAMO Academy, which provides models trained on large-scale industrial corpora and can easily deploy them to practical applications. FunASR provides a variety of industrial models, including Paraformer-large models, FSMN-VAD models, and CT-Transformer models, etc. FunASR enables researchers to easily deploy them in real scenarios. (Source: Jishu Community Embedded AI Column)
Virtualization technology is the foundation of cloud computing business model, and it is also the core technology to promote cloud computing economy. Virtualization refers to the use of virtualization software to create an abstraction layer on computer hardware that divides the hardware elements of a single computer (including processors, memory, storage, etc.) into multiple virtual computers (VMs), each of which runs its own operating system (OS), similar to a stand-alone computer. Oversubscription is to control the maximum number of vcpu resources allocated. This article analyzes the relationship between virtual user oversubscription and virtual machine performance. (Source: Jishu Community Tianyi Cloud Community Column)
The compiler optimizes those things (9 ): Machine Outliner
In the embedded field, code size (code size) optimization can reduce the use of memory, which is very important to the competitiveness of products. For contemporary products, code size optimization can add more features to the product and enhance competitiveness; for next-generation products, code size optimization can improve product power consumption and cost competitiveness [1]. LLVM Machine Outliner is a compiler optimization technology that can identify repetitive code fragments, extract them and convert them into an independent function, thereby reducing the program code size. (Source: Jishu Community Embedded AI Column)
core observation
From June 2nd to 3rd, 2023, the 7th Jiwei Semiconductor Summit was held in Xiamen International Conference Center Hotel. Zhao Yongchao, head of ARM Technology's intelligent IoT and automotive business line, delivered a keynote speech entitled "High Performance Computing IP Platform Empowers the Automotive "Chip" Market" at the summit. In his speech, Zhao Yongchao introduced the technical advantages and market layout of Amou Technology in the field of automotive electronics, and promoted the innovation of automotive smart chips through the high-performance computing IP platform. (Source: Jishu Community Core Vision Column)
Recently, at the Apple conference, they brought a new Apple M2 Ultra chip and its hardware, completing the complete transformation of the Mac series from Intel to Arn. Apple says the M2 Ultra is the largest and most powerful chip Apple has ever made, built on a second-generation 5-nanometer process and uses Apple's breakthrough UltraFusion technology to connect the die of two M2 Max chips, doubling performance . The M2 Ultra is made up of 134 billion transistors -- 20 billion more than the M1 Ultra. A detailed introduction can be read in full. (Source: Jishu Community Semiconductor Industry Observation Column)
Thoughts on the development of AGI general intelligence: Is there a sufficiently general processor?
With the popularity of ChatGPT, AGI (Artificial General Intelligence, general artificial intelligence) has gradually seen the dawn of the outbreak. In just one month, all the giants responded quickly and "invested heavily in the AGI field, regardless of the cost." AGI is a general intelligence based on large models; in contrast, the previous intelligence based on small and medium models for specific application scenarios can be called special intelligence. In the field of chips, we have developed many dedicated chips for specific scenarios. Is it possible to develop a general-purpose chip similar to the development of AGI, which can not only cover almost all scenarios, but also have extremely powerful functions and performance? This paper explores this issue by introducing the development of AGI and the analysis of general-purpose and special-purpose processors. (Source: Jishu Community Software and Hardware Fusion Column)
One article to understand the car control chip (MCU )
This article introduces MCU chips in the four domains of body, chassis, power, and cockpit from the four dimensions of work requirements, performance requirements, industrial structure, and industry barriers. It also sorts out the application status of domestic MCU chips for reference by practitioners. (Source: Jishu Community Automotive Electronics and Software Column)
Related Reading
- 2023 Jishu Communication-In the new era of smart cars, domestic car cores will focus on the central computing architecture
- 2023 Jishu Communication - The penetration rate of smart cockpit is "hurricane", and car manufacturers, Tier1, and chip companies are deeply involved
- 2023 Jishu Communication - "Linglong" Codec Fusion Architecture Helps Diversified Video Demand
- 2023 Jishu Communication - Looking at the development trend of future computing chips from the rise of large models such as ChatGPT
For more content, please pay attention to the Jishu Communication Column