MJY1000
Based on Samsung SF4 · Next-generation AI accelerator
A high-performance AI computing solution supporting both LLM training and inference.
MJY1000: The Core of Next-Gen AI Computing
MJY1000 goes beyond being a simple chip to set a new standard in generative AI and accelerated computing. With outstanding performance, efficiency, and scalability, it meets the demands of modern AI workloads and accelerates enterprise AI innovation as a key infrastructure connecting cloud training and edge inference.
Explore the Inner Workings of Innovation
High-Performance Computing
MJY1000 is designed to support a wide range of precision computing required by modern AI models. It delivers powerful performance with 260 TFLOPS for FP16/BF16 and 520 TFLOPS for FP8/INT8, and even supports ultra-low precision like FP4. Optimized for matrix and tensor operations in large-scale models such as LLMs, it shortens training time significantly.
Memory & Interface
Equipped with HBM3E-ready memory interfaces, MJY1000 ensures stable data transmission in high-bandwidth environments. PCIe Gen5 support enhances scalability and compatibility in server and data center environments. Its efficient NoC architecture optimizes data flow between compute units and memory, minimizing latency. High-speed link technology maintains scalability even in multi-chip systems.
Energy Efficiency & Power Design
Designed with a maximum load TDP of approximately 500W, MJY1000 achieves excellent operational cost efficiency with 0.5T/W for FP16/BF16 and 1T/W for FP8/INT8. Built on advanced low-power processes and thermal control architecture, it maintains stable performance under prolonged high-load conditions.
Scalability & System Integration
MJY1000 is designed not only for single-chip performance but also for large-scale training systems through multi-chip configurations. Its high-speed link-based scalability makes it adaptable to next-generation training clusters and LLM pipelines. Backed by Samsung’s HBM supply chain and advanced packaging technology, it ensures stability in mass production and system integration.