AMD has unveiled significant advancements in AI infrastructure, signaling a robust push to challenge competitors in the AI and data center markets. The company’s introduction of the Instinct MI325X and the MI355X AI accelerators, alongside new data processing units (DPUs), reflects its strategic focus on delivering enhanced performance for AI and generative models, particularly against industry leader NVIDIA.
The Instinct MI325X is a follow-up to AMD’s successful MI300X accelerator, offering key upgrades, including expanded memory capacity and enhanced power management. The MI325X boasts 256GB of HBM3E memory and over 6TB/s of peak bandwidth, making it a formidable option for inference workloads and AI training models. Despite a reduction in the announced memory footprint, AMD claims the design optimizations yield superior real-world performance, with multiple AI models benefiting from the improved memory bandwidth and compute capacity.
While the MI325X is expected to ship within this quarter, the company also revealed its successor, the Instinct MI355X, which is slated for release next year. The MI355X will feature a next-gen CDNA 4 architecture, built on a 3nm process, and is expected to deliver up to 1.8x improved AI inference performance. With a memory configuration of 288GB of HBM3E and support for newer FP4 and FP6 data types, the MI355X is expected to further elevate AMD’s standing in AI acceleration.
