Source: EEWORLD
The hotness of ChatGPT continues to give birth to the hotness of applications including artificial intelligence and big data. Behind the emerging new applications, more complex computing methods are required, and the concept of heterogeneous computing was born. Decomposing different computing tasks and using different architectures and instruction sets including GPU, CPU, and NPU to handle different computing requirements is the key to achieving efficient computing.
"In the post-Moore era, when the computing power and power consumption can no longer follow the linear change of Moore's Law, the heterogeneous architecture has become the development direction of the chip computing architecture." Liu Guojun, general manager of Imagination China, recently attended the first session held by Alibaba Pingtouge Said at the Xuantie RISC-V Ecological Conference.
In Liu Guojun's view, Imagination can provide various products including GPU + CPU + NNA (Neural Network Accelerator), which are very suitable for the needs of different scenarios of heterogeneous computing.
RISC-V does bring an important addition to Imagination's heterogeneous computing strategy. In 2017, Imagination sold the MIPS department, and by the end of 2021, the company officially announced to re-enter the CPU IP market, launched the Catapult series processors, and released them.
In addition, Imagination has become a senior member of the RISC-V Foundation in 2022. In the RISC-V Foundation, Imagination can provide support for the ever-growing RISC-V ecosystem. In addition, it can also provide GPUs and even NNAs for RISC-V products, thereby continuously expanding Imagination's ecosystem.
A typical successful case is the Yingying 1520 launched by Pingtouge last year, which uses Imagination's GPU and NNA, combined with the 4-core black iron C910 core with a maximum frequency of 2.5GHz, to achieve 4K video processing capabilities, and contains rich Heterogeneous computing and high-speed IO resources, as well as 4TOPS@int8 AI computing power.
While Imagination is bringing new vitality to heterogeneous computing for RISC-V, RISC-V is also injecting new vitality into Imagination.
Today, as the demand for heterogeneous computing continues to increase, what preparations has Imagination made? Liu Guojun gave some of his understanding.
Liu Guojun, General Manager of Imagination China
Heterogeneous Computing Requires Broad IP Portfolio
Liu Guojun said that Imagination's IP portfolio basically supports multi-core architecture, so that it can cover different needs from low power consumption and small computing power to parallel large computing power, from entry-level to high-end, and all product portfolios will launch corresponding vehicle-level certification .
As far as the CPU is concerned, as mentioned above, Imagination is reshaping its position in the CPU field through RISC-V. At the same time, it is also helping RISC-V's ecological partners to use more advanced products.
For example, in terms of GPUs, the company has launched GPU IPs at a rate of one generation per year for four consecutive years, namely A/B/C/D series GPUs. Among them, the B-series GPUs have landed in many fields including automotive ADAS and domestic graphics chips. The C-series GPU is the industry's first hardware ray tracing product on the mobile phone platform.
The IMG DXT series GPU launched by the company in 2021 has been further optimized and improved in 2022, providing a 50% increase in performance and computing power. The standard configuration has reached 2.25T floating-point computing power and 72GB pixel filling capacity. 9TB of AI reasoning capability. At the same time, the D4 ray tracing technology is configurable and scalable, and it can be used together with ray tracing. "The pioneering ray-tracing GPU will bring the most advanced graphics technology to all mobile device users, including a wide range of markets such as mobile, automotive, and desktop data centers, bringing a new experience." Liu Guojun said.
In terms of AI accelerators, Imagination's single-core NNA can cover a computing power of 1.5Tops to 12.5Tops, and can quickly expand to hundreds of Tops through a multi-core architecture. At present, Imagination's NNA has been successfully commercialized in fields including visual processing, speech recognition, and automatic driving.
Imagination's comprehensive product portfolio
And as shown in the shadow 1520, the combination of Imagination's GPU and NNA's IP brings unlimited AI and visual possibilities to Pingtouge's products.
Liu Guojun said that based on the combination of Imagination's three core computing IP technologies, combined with advanced interconnection and storage technologies, Imagination can support customers to build flexible multi-core heterogeneous architectures to meet different needs.
For example, in automotive smart cockpits or ADAS scenarios, the combination of GPU+NNA can provide powerful computing power and support the fusion of computing and display rendering. In the field of autonomous driving, the large-scale heterogeneous system composed of GPU+CPU+AI can meet the large-scale, high-concurrency and low-latency computing requirements at various stages from perception, positioning to path planning, and can also be used for data center computing and Smart NIC and other scenarios. The combination of NNA+CPU can meet a wide range of intelligent scenarios, from AIOT to consumer electronics, from smart industry to smart city. For remote rendering of cloud games, heterogeneous computing of GPU+CPU can be applied.
All in all, with the support of a rich IP product portfolio, Imagination can provide customers with more and more comprehensive processor IP solutions in one stop.
ever-expanding ecosystem
The reason why RISC-V grows rapidly is that one of the most important consensuses is open source, and Imagination is also building various ecosystems with an open source attitude.
Liu Guojun said that since joining Imagination in 2015, the company's Chinese employees have grown from less than 10 to more than 60. Through local R&D and services, Imagination's Chinese customer base has achieved rapid growth.
On the GPU, Imagination not only provides a standard GPU software stack, but also provides comprehensive open source software, including firmware, kernel space driver, and user space driver, to better support customer software customization needs. "We give customers the flexibility to customize the GPU software stack. Customers can choose closed source or open source according to their needs." Currently on Chromebook, Imagination has released a demo based on the open source Mesa Vulkan Driver.
As shown in the figure, the pink part is the open source protocol stack
In terms of AI, Imagination is also actively expanding the circle of ecological friends and jointly building an open AI ecosystem. By providing an efficient and flexible NC-SDK software protocol stack, Imagination realizes a software + IP tightly coupled software and hardware solution. In 2022, Imagination and Baidu Flying Paddle jointly established a model zoo to adapt and deeply optimize a representative network model in the industry on Imagination's IP, and it has been released on github.
Exploring the four major competitiveness of chips in the future
Liu Guojun said that the competitiveness of computing chips in the future will be mainly reflected in the following four aspects: scalability and reusability, programmability, high efficiency, and flexibility.
First of all, in terms of scalability and reusability, it is necessary to support a core-based computing architecture and expand in multiple dimensions to meet the computing power requirements of matching workloads.
In terms of programmability, on the premise of supporting industry standard frameworks, languages and APIs, the chip computing architecture is highly optimized. At the same time, it should have complete visualization tools, debugging tools and modeling capabilities. Empowering OEMs to develop at the system level.
In terms of efficiency, advanced sparse and compiler optimization methods are required to reduce computing overhead. And in the highly integrated computing IP, fine-grained assignment and scheduling of tasks.
In terms of flexibility, chips can build SoCs by combining, connecting and programming processor IP to support dynamic structure, sparseness and create execution units. And should adopt advanced interconnection and memory technology, thus reduce system bandwidth and reduce power consumption.
In fact, Imagination's product development trajectory is progressing along these four directions. The RISC-V ecosystem is also becoming more and more robust at the moment when the demand for heterogeneous computing is getting stronger and stronger because of the participation of more commercial companies like Imagination who are good at insight into the needs of the industry.
To quote RISC-V CEO Calista recently, it's an "inevitable trend."
END
Welcome to join Imagination GPU and artificial intelligence communication group 2
Join the group, please add the editor WeChat: eetrend89
(Please add company name and title)
recommended reading
Dialogue with the Chairman of Imagination China: Using GPU as the fulcrum to strengthen software and hardware collaboration to facilitate digital transformation
[Video] Technical Director of Imagination China fully interprets IMG DXT GPU
Imagination Technologies is a UK-based company dedicated to the research and development of chips and software intellectual property (IP). Products based on Imagination IP are used in the phones, cars, homes and workplaces of billions of people around the world. For more information on cutting-edge technologies such as the Internet of Things, smart wearables, communications, automotive electronics, and graphics and image development, welcome to Imagination Tech!