At Computex 2024, AMD highlighted the growing momentum of its AMD Instinct™ accelerators line during the opening keynote by President and CEO, Dr. Lisa Su. AMD unveiled an expanded and multi-year roadmap for AMD Instinct accelerators, promising annual leadership in AI performance and memory capabilities with each new generation.
The updated roadmap kicks off with the new AMD Instinct MI325X accelerator, set to be available in the fourth quarter of 2024. Following that, the AMD Instinct MI350 series will launch in 2025, equipped with the new AMD CDNA™ 4 architecture, offering up to a 35x increase in AI inference performance compared to the AMD Instinct MI300 series based on the AMD CDNA 3 architecture. The AMD Instinct MI400 series is expected in 2026, based on the AMD CDNA “Next” architecture.
“The AMD Instinct MI300X accelerators continue to see strong adoption by numerous partners and customers, including Microsoft Azure, Meta, Dell Technologies, HPE, Lenovo, and others, a direct result of the exceptional performance and value proposition of the AMD Instinct MI300X accelerator,” said Brad McCredie, Corporate Vice President of Accelerated Computing for Data Centers at AMD. “With our updated annual product cadence, we are relentless in our pace of innovation, providing the leadership capabilities and performance that the AI industry and our customers expect to drive the next evolution of AI training and inference in the data center.”
## AMD’s AI Software Ecosystem Matures
The open-source AMD ROCm™ 6 software stack continues to mature, enabling AMD Instinct MI300X accelerators to deliver impressive performance for some of the most popular LLMs. In a server using eight AMD Instinct MI300X accelerators and ROCm 6 running Meta Llama-3 70B, customers can achieve 1.3x better inference and token generation performance compared to the competition. With a single AMD Instinct MI300X accelerator with ROCm 6, customers can get 1.2x better inference and token generation performance than the competition on Mistral-7B. AMD also highlighted that Hugging Face, the largest and most popular repository of AI models, is testing 700,000 of its most popular models every night to ensure they run immediately on AMD Instinct MI300X accelerators. Additionally, AMD continues to work on popular AI frameworks like PyTorch, TensorFlow, and JAX.
## AMD Advances New Accelerators and Reveals Annual Cadence Roadmap
During the keynote, AMD unveiled an updated annual cadence for the AMD Instinct accelerator roadmap to meet the growing demand for more AI computing. This will help ensure that AMD Instinct accelerators drive the development of next-generation cutting-edge AI models. The updated annual roadmap for AMD Instinct highlights:
– The new AMD Instinct MI325X accelerator, which will bring 288 GB of HBM3E memory and 6 terabytes per second of memory bandwidth, use the same industry-standard universal server motherboard design used by the AMD Instinct MI300 series, and be generally available in the fourth quarter of 2024. The accelerator will have industry-leading memory capacity and bandwidth, 2x and 1.3x better than the competition, respectively, and 1.3x better computational performance compared to the competition.
– The first product in the AMD Instinct MI350 series, the AMD Instinct MI350X accelerator, is based on the AMD CDNA 4 architecture and is expected to be available in 2025. It will use the same industry-standard universal server motherboard design as other MI300 series accelerators and will be manufactured with advanced 3nm process technology, be compatible with FP4 and FP6 AI data types, and have up to 288 GB of HBM3E memory.
– The AMD CDNA “Next” architecture, powering the AMD Instinct MI400 Series accelerators, is expected to be available in 2026, providing the latest features and capabilities that will help unlock additional performance and efficiency for large-scale AI inference and training.
Lastly, AMD highlighted that the demand for AMD Instinct MI300X accelerators continues to grow with numerous partners and customers using the accelerators to power their demanding AI workloads, including:
– Microsoft Azure using the accelerators for Azure OpenAI services and the new Azure ND MI300X V5 virtual machines.
– Dell Technologies leveraging MI300X accelerators in the PowerEdge XE9680 for enterprise AI workloads.
– Supermicro offering multiple solutions with AMD Instinct accelerators.
– Lenovo powering hybrid AI innovation with the ThinkSystem SR685a V3.
– HPE using them to accelerate AI workloads in the HPE Cray XD675.