Advanced Micro Devices Inc (NASDAQ:AMD, ETR:AMD) has unveiled a suite of next-generation graphics processing units in hopes of claiming a share of the artificial intelligence computing market from global leader NVIDIA Corp (NASDAQ:NVDA, ETR:NVD).
The flagship product is the AMD Instinct MI325X accelerator, designed specifically for data centre and AI applications.
AMD stated that this chip provides 1.8 times more memory capacity and 1.3 times more bandwidth than rival NVIDIA Corp (NASDAQ:NVDA, ETR:NVD)’s current flagship H200 chip.
According to AMD, the MI325X supports 1.3 times greater half-precision floating-point (FP16) and 8-bit floating point (FP8) compute performance compared to Nvidia's H200 GPU, making it highly competitive for intensive AI workloads like large language models (LLMs).
These new chips are due to ship in the fourth quarter of 2024, “and are expected to have widespread system availability from a broad set of platform providers, including Dell Technologies, Eviden, Gigabyte, Hewlett Packard Enterprise, Lenovo, Supermicro and others” starting in the first quarter of 2025.
“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,” said Forrest Norrod, executive vice president and general manager of MD’s data centre solutions business group.
AMD also previewed its next-generation MI350 chip series, which is intended to compete with Nvidia’s upcoming Blackwell chip.
Built on the next-generation CDNA 4 architecture, the MI350 is expected to deliver up to 35 times improvement in AI inference performance compared to the current MI300 series, which is based on CDNA 3.
These are expected to be commercially available in the second half of 2025.
While AMD is catching up to Nvidia’s technological edge, Nvidia remains ahead of the curve in its release schedule.
Despite delays, Blackwell has begun shipping to customers and Morgan Stanley (NYSE:MS) analysts predict billions of dollars in Blackwell revenue to start pouring in in the January quarter.
Earlier this week, OpenAI celebrated the arrival of its first Nvidia B200 Blackwell chip.
Look what showed up at our doorstep.
Thank you to @nvidia for delivering one of the first engineering builds of the DGX B200 to our office. pic.twitter.com/vy8bWUEwUi
— OpenAI (@OpenAI) October 8, 2024