Tue Oct 15 14:10:00 UTC 2024: ## AMD Unveils New AI-Focused Hardware and Software: A Challenge to Nvidia’s Dominance
**Taipei, Taiwan** – AMD, the chip giant, is aggressively pursuing the burgeoning AI market with a new suite of hardware and software designed to meet the growing demands of enterprise and cloud computing. At its Advancing AI 2024 event last week, AMD announced the launch of its 5th Generation EPYC processor (codenamed “Turin”), the MI325X AI accelerator, and a pair of new networking solutions: Pensando Salina and Pollara.
AMD’s focus on AI is a strategic move, recognizing the shift in the market from theoretical models to practical AI implementation. The company believes that the unique performance requirements of AI necessitate specialized hardware and software. To this end, AMD has crafted its new EPYC processor to deliver exceptional performance in handling AI data pipelines, while the MI325X GPU tackles the crucial tasks of AI training and inference.
The EPYC processor, available in two versions – Zen 5 and Zen 5c – caters to a wide range of datacenter workloads. The Zen 5, built on a 4nm process, excels at workloads such as database, data analytics, and AI, while the Zen 5c, utilizing a 3nm process, focuses on efficiency for scale-out cloud and virtualized workloads.
AMD’s EPYC processors have earned a reputation for performance leadership, with over 950 cloud instances already based on them. The company’s strength lies in the high core count and performance of its processors, allowing cloud providers to pack more virtual machines on each server, leading to greater monetization.
The new Turin EPYC processor boasts significant performance gains, delivering double-digit percentage increases in instructions per cycle (IPC) over its predecessor. While a head-to-head comparison with Intel’s Xeon 6P is yet to be performed, early benchmarks suggest that EPYC and Xeon are closely matched in terms of core count, memory channels, and I/O, with EPYC holding a slight edge in some areas.
AMD’s MI325X GPU, based on CDNA 3 architecture and boasting 256GB of HBM3E memory, promises significant performance leaps over its previous generation and aims to challenge Nvidia’s dominance in the AI space. AMD also unveiled ROCm 6.2, a software stack that aims to enhance performance and support for a broader range of AI models.
To complement its hardware and software offerings, AMD introduced Pensando Salina, a third-generation DPU designed to accelerate AI and ML functions, and Pensando Pollara 400, a new NIC purpose-built for AI workloads. Both are currently being sampled with cloud service providers, with commercial shipments expected in the first half of 2025.
AMD’s strategy goes beyond individual components, aiming to create an end-to-end AI ecosystem. With the acquisition of Silo AI, a leading European AI lab, AMD has added expertise in building and deploying AI solutions. The recent acquisition of ZT Systems further enhances its ability to deliver rack-scale integration, bringing its AI factory closer to the customer.
AMD’s latest announcements mark a significant step in the company’s ambition to challenge Nvidia’s dominance in the AI space. By offering a comprehensive range of AI-focused hardware, software, and expertise, AMD is positioned to play a key role in the rapidly evolving landscape of enterprise and cloud computing.
First Piper 