[Insight] Why Nvidia Is Bringing Rival AI Chips Into Its Ecosystem
U.S. AI chip startup d-Matrix said on Sept. 10 that its next-generation Raptor XPU will connect directly to Nvidia MGX rack-scale infrastructure through NVLink Fusion. Raptor is expected to complete its final design by the end of 2026, with initial MGX-based systems targeted for the fourth quarter o
AI chipmaker Nvidia is expanding its ecosystem by incorporating rival AI chips into its data center infrastructure. This move comes as a response to the growing demand for processors optimized for specific inference workloads, especially in commercial AI deployments. The startup d-Matrix, which developed the Raptor XPU, is partnering with Nvidia to integrate its processor into Nvidia's MGX rack-scale infrastructure via NVLink Fusion.
Raptor is designed with a memory-centric architecture, combining DRAM memory and SRAM compute dies in a three-dimensional design. This architecture aims to reduce bottlenecks associated with data movement between memory and processors, making it particularly suitable for latency-sensitive decoding in generative AI inference workloads. Nvidia's MGX infrastructure, which supports a mix of GPUs, CPUs, DPUs, and networking technologies, will now also incorporate Raptor, forming a modular and interoperable system.
The collaboration between Nvidia and d-Matrix allows for the division of AI inference workloads between different processors, with Nvidia's Vera Rubin platform handling the compute-intensive prefill stage and Raptor taking care of the latency-sensitive decoding stage. This division enables Nvidia's GPUs and Raptor to work in tandem on different stages of the same AI service, thereby increasing the overall efficiency and performance of AI systems.
As the AI market continues to evolve, this partnership signifies the increasing importance of specialized processors and optimized data center architectures. Samsung Electronics and SK hynix, two prominent Korean semiconductor companies, are also well-positioned to benefit from this trend. With the growing variety of AI processors and their specific memory requirements, these companies can expand their market share by providing high-performance memory solutions tailored to these processors.
The shift towards modular and interoperable data center systems presents new opportunities for chipmakers and memory suppliers alike, reshaping the competitive landscape of AI infrastructure.
Written by urgent.news from Korea IT Times's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
