At the Advancing AI event in San Francisco, AMD unveiled its latest technological marvels, aiming to dominate the AI and high-performance computing (HPC) landscape. With the introduction of the 5th Gen AMD EPYC processors and the Instinct MI325X accelerators, the company is positioning itself to challenge the reign of competitors like Intel and NVIDIA.
These new chips promise to deliver top-notch performance, increased energy efficiency, and more accessible AI capabilities for enterprises and consumers.
AMD’s CEO, Dr Lisa Su, led the showcase, highlighting how these advancements are pushing the limits of computational power and enhancing AI workloads. Notably, big-name players such as Microsoft, OpenAI, Meta, and Google Cloud have shown keen interest in AMD’s new products, reinforcing the significance of these launches.
New EPYC processors: more cores, more power
The new AMD EPYC 9005 series processors, built on the advanced “Zen 5” architecture, take center stage. These CPUs boast record-breaking performance and energy efficiency, with early adopters including major tech companies like Dell, Lenovo, and HPE.
AMD’s EPYC chips come in various configurations, from an 8-core, low-power option to a mighty 192-core processor designed for supercomputers. The aim is precise: to capture a larger slice of the data center market, where Intel’s Xeon processors have long been dominant.
AMD has made significant strides in this space, with its data center sales doubling over the past year to $2.8 billion. However, while Intel still controls the majority, AMD is steadily closing the gap, accounting for around 34 percent of the money spent on data center CPUs. The EPYC 5th Gen chips are mainly geared towards optimizing AI workloads, as most GPUs require a CPU to boot up and manage data flows efficiently.
Instinct MI325X: Aiming to undercut NVIDIA in the AI race
The big news, however, comes from the AI front. AMD introduced the Instinct MI325X accelerator, a chip that delivers leading AI performance and memory capacity. This is AMD’s play to take on NVIDIA, which has dominated the data center GPU market with a 90 percent share.
While NVIDIA has enjoyed massive demand and high profit margins, AMD hopes to shake things up by offering a competitive alternative with its MI325X. The chip is set to go into production by the end of 2024 and, if well received, could push NVIDIA to adjust its pricing strategies.
AI demand continues to surge, especially with the rise of advanced generative AI like ChatGPT, which requires substantial GPU power for data processing. Although AMD didn’t announce any new major cloud partners for its Instinct GPUs during the event, the company has already secured Meta and Microsoft as customers. AMD aims to capture a big piece of the market, which it estimates could be worth $500 billion by 2028.
Other announcements
AMD’s ambitious product roadmap also caught attention. The company plans to release new AI chips yearly, with the MI350 expected in 2025 and the MI400 in 2026. This move mirrors NVIDIA’s aggressive pace, with its upcoming Blackwell chips set to hit the market early next year.
Aside from the big chip announcements, AMD revealed new tools for enhancing AI infrastructure. The Pensando Salina DPU and Pollara 400 AI NIC are designed to optimize data pipelines, ensuring smooth GPU communication for high-performance AI systems. Additionally, the Ryzen AI PRO 300 series processors, based on the Zen 5 architecture, will offer leading performance and battery life for AI-driven laptops, particularly those aimed at enterprises.
To round things off, AMD hosted its first developers’ conference, bringing together AI experts from Microsoft, Meta, and Google DeepMind, showcasing how AMD’s software ecosystem can further unlock AI potential.
With this suite of new releases, AMD is making its play to become a severe contender in the AI and data center space, going toe-to-toe with industry giants in what promises to be an exciting battle for tech supremacy.
