Advancing AI 2024: AMD's vision for the future of AI and data centres

Note: This feature was first published on 11 October 2024.

AMD held its Advancing AI 2024 event earlier today, where CEO Dr Lisa Su delivered a keynote outlining how the tech company is positioning itself at the forefront of data centre and AI solutions. To say that the tech world is going all-in on artificial intelligence (AI) is an understatement and in this fast-evolving landscape, where AI continues to reshape business operations and infrastructure, AMD unveiled a new line-up of processors, accelerators, and networking products aimed at setting new standards for data centres in the AI era.

Here are the key announcements.

The new AMD EPYC 9005 series

The AMD EPYC 9005 series processors represent a significant leap forward for Team Red, which is based on the latest Zen 5 architecture. With up to 192 cores (for the EPYC 9965 CPU), these processors promise record-breaking performance. Designed to meet the varied demands of data centres – from AI to mixed workloads – AMD says the new EPYC processors will be featured in a wide range of platforms from leading OEMs and ODMs such as Cisco, Dell and Lenovo. There’s also These new processors are available immediately.

By modernising to a data centre powered by these new processors, AMD claims customers can achieve 391,000 units of SPECrate2017_int_base general-purpose computing performance. This impressive performance comes alongside an estimated 71% reduction in power usage and approximately 87% fewer servers needed. This flexibility gives companies’ CIOs the ability to either benefit from the significant space and power savings or to increase performance for daily IT operations while still delivering strong AI capabilities.

AMD Instinct roadmap revealed

Since its launch, the MI300X has seen widespread deployment by cloud and enterprise partners, supporting models ranging from ChatGPT to Meta's Llama and numerous open-source AI models. The breadth of adoption points to AMD's growing relevance in AI computing and is proving to be a serious competitor to NVIDIA’s leadership in this segment.

At the event, Lisa spotlighted its year-on-year advancements in AI accelerators with the launch of the AMD Instinct MI325X. Targeted at demanding AI workloads, the MI325X aims to lead in both performance and memory capacity. These accelerators are currently on track for production shipments in Q4 2024 and are expected to have widespread system availability from a broad set of platform providers, including Dell, Gigabyte, Hewlett Packard Enterprise, just to name a few, starting in Q1 2025.

Beyond that, AMD revealed future developments too.

The MI350 series accelerators for instance, is set to debut in 2025 and is based on the AMD CDNA 4 architecture. AMD Instinct MI350 series accelerators are designed to deliver a 35x improvement in inference performance compared to AMD CDNA 3-based accelerators like the MI325X. The AMD Instinct MI350 series will have an incredible memory capacity of up to 288GB of HBM3E memory per accelerator. The AMD Instinct MI350 series accelerators are on track to be available during the second half of 2025.

Then there’s the MI400 series expected in 2026 but AMD isn’t quite ready to reveal information about these accelerators just yet.

Networking solutions for AI Infrastructure

Another significant aspect of AMD's announcements involved networking – a crucial part often overlooked in discussions about AI infrastructure. The introduction of the AMD Pensando Salina DPU and Pollara 400 NIC aims to support the evolving, and increasingly high, requirements of AI systems, enhancing CPU and GPU performance by improving overall networking efficiency. Similar to NVIDIA and their Spectrum-X products, which features the NVIDIA Spectrum SN5600 Ethernet switch and the NVIDIA Bluefield-3 SuperNIC, AMD is not just delivering powerful chips; it is also thinking holistically about the infrastructure that supports modern AI workloads.

Industry Collaboration and Ecosystem Expansion

AMD's keynote wasn't just about showcasing new products – it also highlighted the extent of its partnerships across the tech industry. Google, Oracle, Microsoft, and Meta are all integrating AMD hardware into their operations in varied but impactful ways. Google, for instance, announced that EPYC 9005-based VMs would be available in 2025, while Oracle's Cloud Infrastructure is leveraging AMD CPUs and accelerators to support high-profile clients like Uber and PayPal.

Microsoft's CEO Satya Nadella also joined the keynote (albeit in a pre-recorded video) to reaffirm Microsoft's ongoing collaboration with AMD, emphasising the “strong performance of the MI300X” on Azure's AI workloads. Satya also revealed at Microsoft's plans to utilise the upcoming MI350 series and beyond, focusing on improving AI performance efficiency –a significant point given the increasing importance of power and cost metrics in AI compute.

Meta also discussed its collaboration with AMD, revealing that its Llama AI models are now fully operational on AMD EPYC CPUs and Instinct accelerators. These collaborations have allowed AMD to grow its market share in this segment from just a small 2% in 2018 to an incredible 34% by the first half of 2024.

Ryzen AI Pro 300 series for enterprise

Not all of AMD's announcements were limited to data centres. The company also launched its Ryzen AI Pro 300 series processors, which will power the first AI-enabled laptops for enterprises. These processors, also built on Zen 5 and AMD XDNA 2 architectures, boast 55 NPU TOPS (exceeding Microsoft Copilot+’s requirements) of AI performance and enterprise-grade security.

Partners like HP and Lenovo will be putting these processors into their commercial offerings, with over 100 laptop models expected by 2025.

Final Thoughts: AMD's AI Vision

AMD's Advancing AI 2024 event was an impressive showcase of the company’s ambitions and its drive to make a significant impact in the AI and data centre sectors. From what I’ve seen, AMD isn't just trying to keep up with NVIDIA and Intel – it seems to be making a genuine push to redefine this industry. The EPYC 9005 processors, Instinct MI325X accelerators, and Pensando networking solutions all reflect a thoughtful and comprehensive strategy in an AI-driven world.

But what stood out to me most was AMD's commitment to an open AI ecosystem, which the company painstakingly pointed out. It’s undoubtedly a strategic decision to differentiate itself from its competitors, but it’s also a refreshing change from the usual closed systems we see from the biggest player here – NVIDIA. By expanding the ROCm stack and forging partnerships with the likes of Microsoft, Google, and Meta, AMD is making a compelling argument for itself as a major player – not just a follower. It’s not just about making powerful hardware; they're working to make AI accessible, which could be pivotal for smaller developers and enterprises wanting to dive into AI without the constraints (and costs) of proprietary platforms.

Right now, AMD looks set to play a major role in shaping the future of AI and data centres, and I'm excited to see how this unfolds in the coming years.

Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.