AMD Says EPYC Turin Already Crushes NVIDIA Vera by 2.37x in Agentic AI, With Zen 6 Venice Pushing the Lead Past 3.3x

AMD has posted new benchmarks of its EPYC CPUs, including upcoming Venice against NVIDIA Vera, showcasing a huge performance advantage.

AMD Says That Its EPYC CPUs Are Delivering Rack-Scale CPU Performance Today, Destroying Both NVIDIA Vera & Intel Xeon Within Agentic AI Workloads

Agentic AI is pushing all major manufacturers to accelerate their CPU development as they race to drive multi-GW AI factories. The three major manufacturers involved in this race include AMD, Intel, & NVIDIA. Others, such as Arm, Qualcomm, and Broadcom, are also working on their own Agentic AI CPUs, but they have yet to start supplying their latest chips to AI firms.

With that said, the competition between AMD, NVIDIA, and Intel is brutal so far. NVIDIA has recently said that they are all set to become the biggest CPU supplier in 2026, bringing massive revenue streams from its Grace and Vera lineups. At the same time, Intel's supply is fully locked in, and the company is even using its low-bin dies for AI. AMD, on the other hand, has seen massive success in the Agentic AI segment with its EPYC lineups and has already commenced mass production of its next-gen Venice line based on the Zen 6 architecture.

So there's a lot of action going on, and all companies are sharing their own benchmarks to showcase their respective performance advantages. AMD has shared its latest benchmarks based around Agentic AI work flows, and the company is claiming some big gains over the competition.

AMD has shared rack-level performance benchmarks modeled on a 100kW rack scenario. For testing, AMD used two of its own platforms against the competition. The two AMD platforms were based on AMD's Turin-based EPYC 9965 CPU with 192 Zen 5 cores and a next-gen Venice-based EPYC CPU with 256 Zen 6 cores. These were compared to Intel's Xeon 6980P (128-Core) Granite Rapids-AP chip & the NVIDIA Vera CPU with 88 Olympus Cores.

The test methodologies included:

• General-purpose computing: SPEC CPU 2017 Integer Rate
• Server-side Java: a SPECjbb2015-derived workload measuring throughput and latency-sensitive business-logic execution
• Web serving: NGINX with the WRK tool, under sustained concurrent request load
• Key-value store: redis-benchmark, for high-speed in-memory operations
• In-memory caching/analytics: Memcached with memtier_benchmark
• Relational databases: TPROC-C, a TPC-C-derived OLTP proxy, on MySQL

In the test results, AMD estimates that the Turin EPYC lineup, which is already shipping to AI firms, delivers a 2.37x gain over the NVIDIA Vera CPUs and a 1.6x advantage over Intel's Granite Rapids-AP chip. These are AMD's existing EPYC CPUs.

For its next-generation AMD EPYC lineup based on the 256 Core "Zen 6" configuration, the Red Team is estimating a mammoth 3.30x gain over NVIDIA's Vera CPUs. Even Intel's existing solution provides a 1.46x gain over NVIDIA Vera.

AMD says that as core density rises within the fixed power envelope, throughput sees a rise. This is crucial for workflows such as transactional, web-serving, and middleware tiers.

Recent closed-door benchmarks by Phoronix performed at NVIDIA showcased some decent performance figures for NVIDIA Verano CPUs, but obviously, those were conducted under NVIDIA's observation, and the power/thermal characteristics were just some of the many tests that were missing. NVIDIA Vera CPUs are a threat to x86 offerings and are seeing massive adoption, but x86 CPUs still have advantages, and AMD is showing them in full force with its own internal benchmarks.

Rack-Level Density & Single-Core Performance Advantages

Also, AMD's advantage not only lies within performance but also in its shipping-ready Rack-Scale AI solutions, which are massively dense. AMD offers over 27,000 cores and is projected to offer over 36,000 cores with its next-generation Venice lineups. NVIDIA's Vera CPUs will pack 22,500 cores in Rack-Scale solutions. This gives AMD the density and value advantage, offering 2.18-2.90x cores per socket, at 1.18-1.41x system power.

But it's not just density; AMD is also promising superior single-threaded performance, which remains a critical consideration for niche workloads. AMD is expected to deliver a 27% performance per core advantage with its 64-core Venice CPUs against NVIDIA's Vera 88-core configurations. Even the high-core count 96-core Venice models are projected to drive single-core performance gains by 11% over Vera.

AMD is also working towards Inference-Optimized CPU solutions such as Verano, which is going to be cost-optimized and will utilize LPDDR5X memory to drive adoption of EPYC CPUs further in the Agentic AI era.

Follow Wccftech on Google to get more of our news coverage in your feeds.