SAPEON, a global AI semiconductor company, has announced the launch of the X330, an AI semiconductor for data centres with four times higher performance than its predecessor, the X220.
The X330 is produced using TSMC’s 7nm technology. SAPEON plans to conduct X330 prototype testing and customer reliability verification for major customers and to begin mass production in the first half of 2024.
Since the X330 offers enhanced performance and power efficiency, it represents an ideal choice for Large Language Model (LLM) support while improving overall total cost of ownership (TCO). The X330 forms the core of SAPEON’s plan to proactively focus on AI service model development companies and data centre markets. X330-equipped servers will support performance optimisation of software stacks based on open neural network exchange (ONNX), while AI inference platform software and software development tools will also be offered.
Compared to its X220 predecessor, the X330 inference NPU’s application range has been significantly expanded based on standard technology. It is an NPU that can be efficiently utilised in a broader range of fields and industries.
The X330 boasts more than 4x the computational performance and more than twice the power efficiency of the existing X220 AI semiconductor. This represents roughly twice the computational performance and 1.3x better power efficiency than competitors’ 5nm products launched in 2023.
The X330 has a built-in video codec and video post-processing IP to improve the processing speed of video-related programmes. Additionally, 4-channel 4K 60fps video input processing is possible through the built-in hardware IP.
The existing SAPEON X220 was the first in Korea to demonstrate excellent performance in BERT inference, a high-performance AI language model, and the new X330 achieved LLM implementation based on transformer architecture, which is the original technology underpinning ChatGPT.
Armed with innovative AI semiconductor technology that boasts outstanding performance, SAPEON is actively targeting new service markets by strengthening cooperation with SK Telecom. For example, in December 2022, SAPEON built a large-scale NPU Farm with a throughput of 7.6 Peta OPS using a server equipped with SAPEON’s X220 AI semiconductor in SK broadband’s Gasan IDC; a project to commercialise and operate the service in an actual data centre environment was promoted in earnest.
Although there have been prior examples of these processes and services on a small scale, this is the first case in Korea of building a large-scale NPU Farm with the goal of commercialising actual services.
Above all, SAPEON confirmed the possibility of expanding its business model by successfully conducting commercialisation tests in various fields, such as image analysis, natural language processing and image quality improvement through the NPU Farm project.
Along with the recently released X330, SAPEON plans to introduce AI NPUs for high-performance Edge devices such as IP (semiconductor design assets) for self-driving cars and CCTV.
Click below to share this article
