AMD introduced the upcoming launch of its strongest AI chips thus far, the Intuition MI325X accelerators, on Thursday.
“Our goal is to drive an open industry standard AI ecosystem so that everyone can add their innovation on top,” mentioned Lisa Su, AMD chair and CEO, on the firm’s Advancing AI 2024 presentation in San Francisco.
The fifth era Epyc processor positions AMD as an underdog contender to NVIDIA’s Blackwell within the AI market. Throughout the identical presentation, AMD additionally unveiled a number of novel merchandise, together with a brand new server CPU designed for enterprise, AI, and cloud purposes.
AMD Intuition MI325X accelerators add capability to AI infrastructure
AMD Intuition MI325X accelerators pace up basis mannequin coaching, fine-tuning, and inferencing — the processes concerned in in the present day’s rapidly-proliferating generative AI — and have 256GB of HBM3E supporting 6.0TB/s. AMD’s CDNA 4 structure allows the brand new line.
The capability and bandwidth of those accelerators out-perform the foremost competitor, the NVIDIA H200, AMD claims. The tech firm additionally says that the Intuition MI325X accelerators can hasten inference efficiency on the Mistral 7B AI by 1.3x, on Llama 3.1 70B by 1.2x, and on Mistra’s Mixtral 8x7B by 1.4X when put next with the H200.
AMD primarily targets hyperscalers with this product. Particularly, hyperscalers need to increase their AI-capable {hardware} in information facilities and energy heavy-duty cloud infrastructure.
The Intuition MI325X is scheduled to go on sale within the final quarter of 2024. Within the first quarter of 2025, they’ll seem in units from Dell Applied sciences, Eviden, Gigabyte, Hewlett Packard Enterprise, Lenovo, and Supermicro. Following that, AMD will proceed to increase its MI350 sequence, with 288GB Intuition MI350 sequence accelerators anticipated within the second half of 2025.
The fifth Gen AMD Epyc server CPU contains as much as 192 cores
The newest era of AMD’s Epyc processors, code-named “Turin,” additionally debuted in San Francisco, that includes Its Zen 2 Core structure. AMD Epyc 9005 Collection processors are available in myriad configurations — with core counts from eight to 192 — and pace up GPU processing for AI workloads. AMD’s essential competitor on this space is Intel’s Xeon 8592+ CPU-based servers.
The efficiency density is a key benefit, AMD mentioned. Larger-capacity GPUs make it attainable to make use of an estimated 71% much less energy and about 87% fewer servers in a knowledge middle, the corporate mentioned. AMD supplies a disclaimer noting that environmental elements contain many assumptions if not utilized to a particular use case and site.
SEE: Safety researchers discovered some fraudsters revenue with the assistance of AI-generated video that may trick facial recognition software program.
All Epyc 9005 Collection processors had been launched on Thursday. Cisco, Dell, Hewlett Packard Enterprise, Lenovo, Supermicro, and main ODMs and cloud service suppliers help the brand new line of chips.
“With the new AMD Instinct accelerators, EPYC processors and AMD Pensando networking engines, the continued growth of our open software ecosystem, and the ability to tie this all together into optimized AI infrastructure, AMD underscores the critical expertise to build and deploy world class AI solutions,” mentioned Forrest Norrod, government vice chairman and common supervisor, Information Middle Options Enterprise Group, AMD, in a press launch.
Two new merchandise cowl front- and back-end tech for AI networking
For AI networking in hyperscale environments, AMD developed the Pensando Salina DPU (entrance finish) and the Pensando Pollara 400 NIC (again finish). The previous handles information switch, delivering information to an AI cluster securely and at pace. The latter, a NIC or community interface card, manages information switch between accelerators and clusters utilizing a Extremely Ethernet Consortium-approved design. It’s the trade’s first AI NIC to take action, AMD mentioned. The DPU helps 400G throughput.
The broader aim of this know-how is to allow extra organizations to run generative AI on units, in information facilities, or within the cloud.
Each the AMD Pensando Salina DPU and AMD Pensando Pollara 400 NIC will likely be usually obtainable within the first half of 2025, AMD expects.
Coming quickly: The Ryzen Professional 300 Collection laptops for business use
OEMs will start delivery laptops with AMD’s Ryzen Professional 300 sequence processors later in 2024. First revealed in June, the Ryzen Professional 300 sequence is a key part of AI PCs. Particularly, they assist Microsoft’s effort to place Copilot+ AI options ahead in its present and upcoming business units.
“Microsoft’s partnership with AMD and the integration of Ryzen AI PRO processors into Copilot+ PCs demonstrate our joint focus on delivering impactful AI-driven experiences for our customers,” mentioned Pavan Davuluri, company vice chairman, Home windows+ Gadgets, Microsoft, in a press launch.
Lenovo constructed its ThinkPad T14s Gen 6 AMD across the Ryzen AI PRO 300 Collection processors. Luca Rossi, president, Lenovo Clever Gadgets Group, talked up the chips within the press launch, saying, “This device offers outstanding AI computing power, enhanced security, and exceptional battery life, providing professionals with the tools they need to maximize productivity and efficiency.”
TechRepublic coated AMD’s Advancing AI occasion remotely.