What I want to thank the most is not the European and American giants... Are domestic chips really going to go viral? Smell the Wind Direction Model from the World Artificial Intelligence Conference | World Artificial Intelligence Conference | Hanbo Semiconductor | Tiantian Intelligent Chip | Kunlun Chip | Shanghai | Domestic | Chip
Today, the 2023 World Artificial Intelligence Conference kicks off. The exhibition of this year's conference gathers numerous highlights, especially around large models. On site, there will be exhibitions of large model application chips such as the Mu Xi Si N100, Hanbo SG100, Suiyuan's Deep Thinking 2.0 and Deep Thinking 2.5, Kunlun Core 2nd generation, Haifei Science Compass C10, and Tiantian Smart Core Smart Armor 100. In yesterday's early exploration, the reporter felt that the biggest change in the chip industry is that with the explosive rise of ChatGPT, domestic computing power has gone from surplus to insufficient, and various types of application chips, especially domestic chips, have truly begun to show their skills.
【 Tiantian Zhixin Universal GPU Inference Product "Zhihai 100" 】
Tiantian Zhixin, established in 2016, is one of the few "Shanghai Xins" that can run from design, casting to mass production, and application closed-loop. Tiantian Zhixin is quite ambitious. Its Tiangai 100, released in March 2021, is a high-performance cloud general-purpose parallel computing card based on GPGPU architecture chips. It is independently designed and developed from the bottom hardware to the upper software, without taking the shortcut of purchasing foreign IPs, ensuring complete independent intellectual property rights. Lin Hang, the product manager of Tiantian Zhixin, told reporters that it is precisely based on the Tiangai 100 that domestic customers have gradually accumulated knowledge and trust in domestic chips. "Fortunately, customers of top ranked security monitoring companies in China are willing to give us the opportunity to switch from mainstream code to Tiantian Zhixin code, which makes us deeply grateful."
Tiangai 100
The actual use of "Tiangai 100" in domestic customer scenarios has gradually established a foothold for domestic chips, and has also become the biggest confidence for Tiangai Smart Chip to further promote "Zhihai 100" to the market. The Smart Armor 100, which was unveiled at the 2023 World Artificial Intelligence Conference, has been mass-produced and is mainly aimed at the cloud inference market. It has three major characteristics: high computing performance, wide application coverage, and low usage cost. Compared to mainstream products in the market, the Smart Armor 100 can provide 2 to 3 times the actual usage performance, suitable for many scenarios such as smart cities, smart ports, smart transportation, and intelligent manufacturing. It also continues the ecological compatibility strategy of the Tiangai 100 and can complete incremental training work without additional purchase costs, effectively reducing user cost expenditures.
Hanbo Semiconductor will promote the second generation of graphics processing and rendering chip SG100
In 2018, the World Artificial Intelligence Conference opened its first year and is also recognized as the year of the rise of domestic AI enterprise entrepreneurship. A large number of enterprises chose Shanghai as the place where their dreams began. In December of that year, Hanbo Semiconductor was established. At last year's World Artificial Intelligence Conference, Hanbo gave the public a preview of its domestically produced 7-nanometer cloud GPU chip SG100. SG100 has industry-leading graphics rendering performance, providing deep optimization for key metaverse application scenarios such as cloud games, cloud phones, cloud desktops, and cloud computing.
Wang Xiaoning, the product manager of Hanbo GPU, gave an on-site demonstration to the reporter. The reporter saw that after encoding the image, Hanbo had significantly less information loss. In addition, after AI processing, the video can achieve intelligent color enhancement and super-resolution intelligent optimization. Wang Xiaoning sighed that Hanbo's domestic chips have been able to "cheat" all the way because they constantly gain confidence from domestic customers. He admitted that the customer's concerns when they first started purchasing domestically produced chips can be imagined, "When we claim to be better than others, the customer will ask us to prove it to them. We continue to prove our abilities and bring value to the customer in the constant demand and new scenarios. Trust is a step-by-step accumulation process."
It is reported that Hanbo Semiconductor will officially release the second generation SG100 and a number of brand-new products at this conference, providing complete solutions for AI large models, graphics rendering and high-quality content production.
Kunlun chip showcases the second generation of AI chips
The new products showcased by Kunlun Xin Technology mainly include second-generation AI chips and R200 series accelerator cards based on second-generation AI chips.
Among them, the Kunlun Core second-generation AI chip is the first general-purpose AI chip in China to use graphics memory, which has important value in promoting the research and commercial implementation of AI chip technology in China. Compared to the first generation product, this chip achieves performance, energy efficiency, and ease of use improvements through optimization of underlying core technologies such as chip architecture and instruction sets, and increases the computing power of general-purpose computing cores by 2 to 3 times.
To unleash the computing power of a chip, an accelerator card is needed. The Kunlun Core R200 series AI accelerator card adopts Kunlun Core second-generation AI chips, which can fully support various artificial intelligence tasks, including natural language processing, computer vision, speech, and traditional machine learning. At present, the Kunlun Core AI accelerator card R200 has been widely deployed in various application scenarios such as the Internet, smart finance, smart parks, and intelligent transportation.
【 Hai Fei Ke Compass C10 chip 】
The Compass C10 chip showcased by Haifei Technology is the company's first generation GPU chip. This is the industry's first GPU chip with a graphics memory of up to 128GB, and its biggest feature is the ability to deploy billions of models on a single card.
"A large model like ChatGPT requires support from tens of thousands of chips and acceleration cards, but in reality, the effect of chips stacked together is not 1+1 equal to 2, but less than 2. Therefore, in terms of cost-effectiveness, if one chip can solve the problem, then the cost-effectiveness must be the highest." The technical personnel at the Haifei Technology booth told reporters.
The technician also stated that from the current commercial applications in China, domestic chips can meet the requirements, without the need to compete for the highest computing power, but rather for actual efficiency. Therefore, cost-effectiveness will be an important factor affecting decision-making.