AMD发布新的AI芯片以与Nvidia竞争

At the Computex trade show in Taipei, AMD CEO Lisa Su revealed the MI325X accelerator, set for release in the fourth quarter of 2024.

The demand for advanced chips in AI data centers has surged due to the development of generative AI programs. AMD aims to compete with Nvidia, which dominates about 80% of the AI semiconductor market. Since last year, Nvidia has committed to an annual release cycle, a strategy AMD now adopts.

"AI is clearly our number one priority as a company and we have really harnessed all of the development capability within the company to do that," Su said. "This annual cadence is something that is there because the market requires newer products and newer capabilities... Every year we have the next big thing such that we always have the most competitive portfolio."

AMD also introduced an upcoming series of chips called MI350, which is expected to be available in 2025 and will be based on new chip architecture.

Compared to the MI300 series of AI chips, AMD said it expects the MI350 to perform 35 times better in inference, the process of computing generative AI responses.

Additionally, AMD plans to release the MI400 series in 2026, featuring the "Next" architecture.

Su also unveiled the Ryzen 9000 series desktop processors, featuring the Zen5 architecture. The initial lineup includes the Ryzen 9 9950X, Ryzen 9 9900X, Ryzen 7 9700X, and Ryzen 5 9600X, all set to launch in July. The fifth-generation EPYC series AI GPUs are expected to ship in the second half of 2024, while the Versal AI Edge Gen 2 series processors are available for early access now.

In her speech, Su highlighted the global nature of the semiconductor ecosystem. She emphasized the importance of collaboration with key suppliers like TSMC and other partners in Taiwan, underscoring AMD's commitment to a resilient and interconnected supply chain.

“AI is a phenomenal technology that will only improve over time, but it's not perfect yet. We need to continue enhancing capabilities, models, training, and data,” said Su.

“While we haven't fully achieved artificial general intelligence (AGI), AMD aims to provide the foundational hardware, software, and systems to help leading AI researchers reach these milestones,” she added.

Nvidia's CEO Jensen Huang said last Sunday that the company's next-gen AI chip platform, Rubin, set for 2026, will include GPUs, CPUs, and networking chips.

Investors, who have poured money into AI chip firms, are seeking long-term updates to gauge the sustainability of the booming AI sector. AMD shares remained steady, while Nvidia shares rose over 3% on Monday. AMD's stock has more than doubled in value since the start of 2023, though Nvidia's has increased seven-fold in the same period.

In April, Su said that AMD expects AI chip sales to reach roughly $4 billion in 2024, up $500 million from previous estimates. At Computex, AMD also mentioned that its next-gen central processor units will likely be available in the second half of 2024.

Although businesses prioritize AI chips for data centers, some of AMD's CPUs are used with graphics processors, albeit with a preference for GPUs. AMD also detailed its new neural processing units (NPUs), designed for on-device AI tasks in AI PCs.

Chipmakers are relying on enhanced AI capabilities to drive growth in the PC market, which is recovering from a prolonged slump.