AMD has released fresh performance data for its 6th Gen EPYC “Venice” server processors, positioning the new Zen 6-based CPUs as a major step forward for Agentic AI, enterprise cloud workloads, and high-performance computing. The company’s latest benchmark figures compare the new EPYC lineup against competing platforms from Intel and NVIDIA, with AMD claiming strong advantages in per-core speed, total throughput, memory bandwidth, and rack-level efficiency.
The 6th Gen EPYC Venice processors were introduced during AMD’s Advancing AI 2026 event and are designed around the company’s new Zen 6 architecture. AMD says the new cores deliver higher per-core performance, stronger overall platform throughput, and improved power efficiency, making them better suited for modern data center tasks that blend traditional CPU workloads with AI-driven operations.
A major focus of AMD’s new performance disclosure is Agentic AI. These workloads rely on systems that can reason, plan, retrieve information, execute tasks, and interact with users or other tools in real time. Because of that, CPU performance still plays a major role, especially in latency-sensitive stages such as inference orchestration, database access, live translation, user-facing services, and decision-making pipelines.
In per-core testing using SPEC 2026, AMD claims its 6th Gen EPYC Venice CPUs deliver a 20% uplift over NVIDIA’s 88-core Vera CPU. Per-core performance is especially important in AI inference and enterprise applications where faster response times matter just as much as raw throughput.
AMD also highlighted results from SPECrate 2026 Integer testing. In those benchmarks, the EPYC 9996 server processor reportedly delivers 1.2 times the per-core performance of a Vera-based platform and 2.24 times the overall platform-level performance. AMD says the advantage grows even larger in enterprise and cloud-native workloads, including server-side Java, OpenSSL, MongoDB, Redis, NGINX, and transaction processing, where the company reports gains ranging from 2.4x to 3.7x.
Against Intel’s Xeon 6980P, AMD claims the EPYC 9996 delivers performance advantages between 1.8x and 3.13x across workloads such as molecular dynamics, materials modeling, and weather forecasting. These are compute-heavy tasks where CPU throughput, memory bandwidth, and platform efficiency can significantly impact time to results.
AMD’s broader SPEC 2026 comparisons include several major data center processors, including NVIDIA Vera, Amazon’s AWS Graviton5, and Intel’s Xeon 6980P. According to AMD, its 5th Gen EPYC processors with 192 cores deliver a 37% gain over NVIDIA Vera, while the 6th Gen EPYC Venice processor with 256 cores increases that lead to 2.37x.
For Agentic AI-specific benchmark testing, AMD used a 96-core EPYC Venice processor against NVIDIA’s 88-core Vera chip. In those tests, AMD reports a 12% gain in per-core performance overall, with the a4 subtests averaging around a 20% improvement. Some individual benchmarks showed much larger leads, including up to 2.01x in tests such as stockfish_r, while several other workloads reportedly exceeded a 30% advantage.
Memory bandwidth is another area where AMD says EPYC Venice pulls ahead. This is critical for generative AI, model inference, simulation, graph analytics, and large enterprise database systems. Using the STREAM Triad benchmark to measure sustained memory bandwidth, AMD reports that EPYC Venice delivers 18% higher bandwidth than NVIDIA Vera.
The advantage comes from EPYC Venice’s 16-channel memory design, which supports RDIMM modules at speeds up to 12,800 MT/s. AMD says this allows the platform to reach 1.6 TB/s of memory bandwidth, compared with 1.2 TB/s on NVIDIA Vera using LPDDR5x memory. Even with AMD’s higher 96-core count versus NVIDIA’s 88 cores in the tested comparison, AMD says each Venice core still gets access to roughly 8% more memory bandwidth.
AMD also shared results for enterprise and cloud-native workloads. When normalized against an Intel Xeon 6980P system with 128 cores, the EPYC 9996 reportedly delivers 2.4x higher platform-level performance in server-side Java, 2.5x in OpenSSL, 3.5x in MongoDB with YCSB, 2.9x in Redis Benchmark, 3.7x in NGINX with WRK, and 2.6x in transaction processing workloads based on a TPC-C kit.
High-performance computing is another major target for the new EPYC lineup. AMD says its 5th Gen EPYC processors already provide up to a 2x performance gain over Intel’s latest Xeon 6 family in select HPC workloads, while 6th Gen EPYC Venice can extend that advantage to more than 3x in certain tests. These workloads include scientific simulation, weather modeling, molecular research, and other compute-intensive applications commonly used in research institutions and large-scale enterprise environments.
AMD also emphasized rack-level performance under a 100-kilowatt power budget, an increasingly important metric for data centers constrained by power availability and cooling capacity. In AMD’s modeled 100kW rack scenario, 5th Gen EPYC reportedly delivers a 2.3x uplift over NVIDIA Vera, while 6th Gen EPYC Venice raises that figure to 3.4x. AMD says this reflects the platform’s ability to deliver more work within the same power envelope, which can help reduce infrastructure costs and improve data center density.
The arrival of EPYC Venice marks an important moment for AMD’s server CPU strategy. With Zen 6 cores, higher core counts, expanded memory bandwidth, and stronger efficiency claims, the 6th Gen EPYC family is being positioned not only for traditional enterprise servers but also for the next wave of AI-heavy data center workloads.
While these figures come from AMD’s own testing and should be evaluated alongside independent benchmarks once systems become widely available, the company’s message is clear: EPYC Venice is designed to compete aggressively in Agentic AI, cloud computing, enterprise databases, and HPC. As AI workloads continue to evolve beyond GPU-only acceleration, AMD is betting that high-performance CPUs will remain a critical part of the modern data center.
