AUS
NZ
UK
CA
USA
IND
Highlights:
- Enhanced AI Performance: AMD's Instinct MI325X offers 1.8 times more memory capacity and 1.3 times more bandwidth than Nvidia’s H200 chip.
- Widespread Availability: The MI325X is set to ship in Q4 2024, with broad platform support anticipated by early 2025.
- Competitive Edge with MI350: The upcoming MI350 series aims for a 35-fold AI performance improvement, positioning it against Nvidia’s Blackwell chip.
Advanced Micro Devices Inc (NASDAQ:AMD) has introduced a new series of high-performance graphics processing units (GPUs) aimed at competing in the fast-growing artificial intelligence (AI) computing market, which is currently dominated by Nvidia Corp (NASDAQ:NVDA). The latest product line is centered around the AMD Instinct MI325X accelerator, specifically designed for data center and AI applications, offering significant improvements over existing competitors.
The Instinct MI325X aims to provide enhanced capabilities for handling demanding AI workloads, including large language models (LLMs). AMD claims that this new accelerator offers 1.8 times more memory capacity and 1.3 times more bandwidth than Nvidia's flagship H200 chip. This increase in memory and bandwidth is essential for processing larger datasets and executing complex AI calculations more efficiently. Additionally, the MI325X boasts 1.3 times greater half-precision floating-point (FP16) and 8-bit floating-point (FP8) compute performance, making it well-suited for intensive AI tasks.
AMD plans to start shipping the MI325X in the fourth quarter of 2024, with widespread availability expected from major technology partners such as Dell Technologies, Eviden, Gigabyte, Hewlett Packard Enterprise, Lenovo, and Supermicro in the first quarter of 2025. This broad support from platform providers highlights AMD’s strategic efforts to integrate its new GPUs into a wide range of AI and data center infrastructures, allowing customers to scale AI deployments more effectively.
Forrest Norrod, AMD’s executive vice president and general manager of its data center solutions business group, emphasized the company's commitment to delivering on its product roadmap, stating, “AMD continues to deliver on our roadmap, offering customers the performance they need and the choice they want, to bring AI infrastructure, at scale, to market faster.”
In addition to the MI325X, AMD provided a preview of its upcoming MI350 chip series, designed to directly compete with Nvidia’s next-generation Blackwell chip. Built on the new CDNA 4 architecture, the MI350 series is expected to deliver up to a 35-fold improvement in AI inference performance compared to the current MI300 series, which uses CDNA 3. AMD anticipates that these chips will be commercially available in the second half of 2025.
Despite AMD's advancements, Nvidia maintains a technological lead, particularly with its Blackwell series, which has already started shipping to customers. Analysts from Morgan Stanley (NYSE:MS) predict that Nvidia’s Blackwell series could generate significant revenue starting in the January quarter, underscoring the intense competition between the two tech giants in the AI space.
