Recently, at Computex 2024, AMD (NYSE:AMD) made a number of important announcements, including the new Instinct MI325X accelerator, which will be available in the fourth quarter of this year. In addition, the MI350 series based on the CDNA4 architecture will be launched next year, and the new MI400 series with the CDNA 'Next' architecture is planned for 2026.

AMD has committed to ensuring continued improvements in AI and high-performance computing (HPC) performance with new product releases every year, including enhanced instruction sets and higher memory capacity and bandwidth.

For the AMD Instinct MI325X, the product will come with up to 288GB of HBM3E memory with a memory bandwidth of 6TB/s, providing a 1.3x improvement over Nvidia's H100 in terms of inference performance and token generation. Notably, the MI325X will compete with Nvidia's H200 and even B100/B200 accelerators.

The MI350 series is based on the AMD CDNA 4 architecture and is expected to launch in 2025, offering a 35x improvement in AI inference performance over the current MI300 series. The series will use 3nm-class process technology and support new data formats (FP4 and FP6) and instructions to boost AI performance and efficiency.

AMD said the AMD ROCm 6 software stack plays a key role in maximising the performance of the MI300X accelerator. According to AMD's benchmarks, systems using eight MI300X accelerators outperformed Nvidia's H100 by a factor of 1.3x in Meta Llama-3 70B model inference and token generation, and individual MI300X accelerators outperformed the competition by a factor of 1.2x in Mistral-7B model tasks.

Adoption of the AMD Instinct MI200 and MI300 family of products by cloud service providers and system integrators is also accelerating. Microsoft Azure uses these accelerators for OpenAI services, Dell integrates them into PowerEdge enterprise AI machines, and Lenovo and HPE use them for servers.

AMD Instinct MI300X accelerators continue to receive strong purchase intent from Microsoft Azure, Meta, Dell Technologies, HPE, Lenovo and many other partners and customers, which is a direct reflection of the strong performance of the AMD Instinct MI300X accelerators. Our increasing pace of new product releases each year will deliver AI industry-leading capabilities and performance to support customers in driving data centre AI and inference training.