![AMD Targets the Agentic AI Era CPU Market with 6th-Gen EPYC [AMD ZEN 6] - IT조선](/api/images/proxy?url=https%3A%2F%2Fit.chosun.com%2Fnews%2Fphoto%2F202611%2F3000000001113_501251_4613.jpg&w=800&q=80&f=webp)
AMD Targets the Agentic AI Era CPU Market with 6th-Gen EPYC [AMD ZEN 6] - IT조선
AMD has unveiled its 6th-generation EPYC 'Venice' processors, featuring up to 256 cores and optimized for agentic AI workloads, with Microsoft set to become the first cloud provider to deploy them.
AMD is making an aggressive push into the data center CPU market for the agentic AI era with its 6th-generation EPYC 'Venice' processors. Featuring configurations of up to 256 cores, these processors deliver up to 2x the throughput of competing products and 2.8x the power efficiency for AI agent workloads.
At its 'AMD in Depth' event in Las Vegas on November 4-5, the company unveiled the core architectural highlights of the Venice lineup, which spans four distinct product families across three platforms. These address workloads ranging from cloud and enterprise to high-performance computing (HPC) and AI infrastructure.
Performance Optimizations
The 6th-gen EPYC 'Venice' is engineered to tackle escalating compute demands and diverse workloads while addressing data center constraints. It maximizes compute density with up to 256 cores per socket and enables organizations to consolidate up to 12 legacy servers into a single modern server.
AMD has structured the EPYC 9006 Series into four main product lines. The performance-centric SP7 platform-based family targets hyperscale clouds, while the SP8 platform-based Venice focuses on enterprise environments. The lineup also includes 'Venice-X' for HPC and technical computing, and the EPYC 9006 LP 'Verano' for performance and power efficiency.
According to AMD, EPYC 'Venice' achieves up to 3.3x higher throughput than rival processors in cloud-native workloads. For enterprise workloads, it delivers up to 2.6x the performance of competing products, while for storage and database workloads, it offers up to 3.3x the performance of competitors.
Memory Efficiency
AMD's MEXT technology, available on 6th-gen EPYC platforms, improves memory cost efficiency by offloading portions of underutilized high-capacity memory configurations onto SSDs. By tiering lower-priority 'cold data' to SSDs and using AI to predict memory usage, the technology boosts cost efficiency with minimal performance impact.
AMD states that MEXT can improve memory performance-per-dollar by up to 3.5x, helping data centers reduce memory cost burdens while maintaining high performance for critical workloads.
Microsoft Deployment
Microsoft will become the first cloud provider to launch instances powered by 6th-gen EPYC processors, rolling out three major service tiers. The 'HDv2' instance will deploy dual Venice CPUs with high-capacity memory and high-speed networking tailored for large-scale data preprocessing workloads.
The HDv2 instance delivers 80% higher SparkETL performance than comparable offerings currently on the market and recently opened for preview registration. Microsoft's adoption signals growing confidence in AMD's technology for demanding AI workloads.
Agentic AI Architecture
Ravi Kuppuswamy noted that inference workloads have surpassed training in volume, with agentic AI at the center of this transition.
AMD addresses the agentic AI pipeline through a three-tier architecture: GPU host nodes, agentic sandboxes, and general-purpose compute. The company offers different Venice CPU configurations for each tier, including a 96-core 'Zen 6' CPU for GPU host nodes and a 'Zen 6c'-based Venice CPU with up to 256 cores for agentic sandboxes.
AMD is deploying these CPUs across both its GPU-centric 'Helios' and CPU-centric 'Helion' rack-scale solutions, with the high-density liquid-cooled variant packing up to 36,864 cores per rack. The company has also introduced the 'ROCm Agentic Toolkit' to streamline the deployment of CPU sandbox environments.