AMD unveils new GPU architecture designed to directly compete with Nvidia's Rubin flagship AI accelerator.
AMD has introduced the Instinct MI455X accelerator, its flagship processor designed for large-scale AI training and inference, directly challenging Nvidia's Rubin GPU. Unveiled during AMD's Advancing AI 2026 event, the MI455X represents the new flagship of the company's Instinct MI400 family and will power its Helios rack-scale AI platform. Despite trailing Nvidia in market share, AMD has demonstrated competitive parity in GPU technology.
The MI455X boasts 320 billion transistors, 432GB of HBM4 memory, and up to 40 petaflops of FP4 AI compute. In several key specifications, it matches or surpasses Nvidia's Rubin. While Rubin is expected to deliver approximately 50 PFLOPS of FP4 performance, it offers only 288GB of HBM4 memory—50 percent less than the MI455X. AMD also claims memory bandwidth of up to 23.3 TB/s, slightly exceeding Rubin's 22 TB/s.
Built on AMD's new CDNA 5 architecture, the MI455X represents the company's most significant accelerator redesign in years. Manufactured using TSMC's 2nm and 3nm process technologies, the chip uses an advanced chiplet design connected with 3D hybrid bonding. The architecture introduces improvements in tensor processing, cache bandwidth, memory movement, and execution efficiency aimed at accelerating large language models and agentic AI workloads.
With up to 40 PFLOPS of FP4 performance and 20 PFLOPS of FP8 compute, the MI455X roughly doubles the AI compute capability of the current Instinct MI350 generation. It also substantially increases HBM memory bandwidth from approximately 8 TB/s on the MI350 to more than 23 TB/s on the new accelerator.
Both Nvidia and AMD have evolved beyond positioning their GPUs as standalone products. Instead, they market them as components within comprehensive compute systems that include CPUs, networking, security, and software.
The Helios rack-scale platform—AMD's answer to Nvidia's DGX platform—links 72 GPUs into a unified system. A fully configured Helios rack features Instinct accelerators, EPYC processors, Pensando networking, and the ROCm software stack. AMD claims it provides up to 31TB of HBM4 memory and as much as 2.9 exaflops of FP4 AI compute.
The MI455X narrows AMD's hardware gap with Nvidia considerably. On paper, the accelerator offers competitive AI compute alongside substantially more memory than Rubin—a combination designed to appeal to customers training ever-larger AI models.
The Instinct MI455X will begin shipping at the end of the third quarter of 2026, with volume ramping through the fourth quarter and into the first half of 2027.