What Vera Is and Why Nvidia Says It Beats x86
Nvidia’s Vera CPU is a server-class ARM processor built around 88 custom Olympus cores and LPDDR5X memory, designed to run agentic AI, databases, and data analytics significantly faster than existing x86 data center CPUs by combining high instructions per clock, massive internal bandwidth, and low-latency memory access in a single-socket platform. In its keynote, Nvidia claimed that Vera can “fetch and execute 10 instructions per clock cycle” and delivers 1.8x the performance of the highest-performing x86 CPU in agentic sandbox benchmarks. This is the basis of the headline claim that Vera CPU performance is about 80% higher than leading x86 rivals. Because Vera is ARM-based, it also pushes the ARM processor vs x86 debate deeper into the server space, positioning Nvidia server processors as central parts of its AI-focused data center CPU benchmark story.
88 Olympus Cores, 176 Threads and Mesh Bandwidth Architecture
At the heart of Vera are 88 custom Olympus cores using the ARM instruction set, each supporting two threads for a total of 176 threads per socket. Every core includes 2MB of dedicated L2 cache, while all cores share a large 164MB L3 cache. Instead of splitting cores across chiplets, Nvidia uses a single large mesh, which it says avoids cross-chiplet latency and raises internal bandwidth to 3.4TB/s. According to The Elec, this design yields three times higher per-core bandwidth and twice the total bandwidth of conventional x86 CPUs. Thermal design power spans 250W to 450W, confirming Vera as a high-performance data center CPU. These choices show Nvidia optimizing for tightly coupled, thread-heavy AI and data workloads where fast core-to-core communication and cache access can matter more than raw clock speed in data center CPU benchmark comparisons.
LPDDR5X Memory and the 1.2TB/s Bandwidth Advantage
A defining feature of Vera CPU performance is its memory system. Vera is the first server CPU to use LPDDR5X, a memory type more common in laptops and phones than in racks of servers. Each Vera socket can be paired with up to 1.5TB of LPDDR5X RAM, delivering a quoted 1.2TB/s of memory bandwidth between CPU and memory. Nvidia also claims a 40% reduction in maximum memory latency compared with existing x86 processors. That combination of high bandwidth and lower latency supports workloads like AI inference, in-memory databases, and real-time analytics, which can stall when memory feeds cores too slowly. For data centers weighing an ARM processor vs x86 purchase, these numbers indicate that Vera’s edge is not only its many cores, but also how quickly those cores can move data in and out of memory under heavy multi-tenant server loads.
Real-World Workloads: Agentic AI, Databases and Market Data
Nvidia positions Vera as a CPU for “agentic AI” — AI agents that call tools, run code and respond at nanosecond-level timescales. Benchmarks cited from Phoronix show Vera leading in agentic workloads such as code compilation, Python, Java and database processing, with SQL performance up to three times higher than previous systems based on older CPUs. In a more concrete scenario, Vera demonstrated up to a six-fold improvement in real-time stream processing for the New York Stock Exchange, which processes more than 1.1 trillion messages per day. These figures suggest that in data center CPU benchmark terms, Vera’s design is tuned for latency-sensitive, parallel workloads rather than conventional, human-facing transactional tasks. For operators of trading platforms, recommendation engines, or AI agent farms, that could mean fewer servers to hit the same throughput and faster response times when pipelines are under peak load.
Vera in Nvidia’s ARM Server Roadmap and the Shift to Custom Silicon
Vera is not a standalone experiment; it is the CPU half of the Vera Rubin platform, paired with Rubin GPUs over an NVLink-C2C interconnect rated at 1.8TB/s between CPU and GPU. Nvidia has also designed a Vera CPU Rack holding 256 CPUs for 22,528 cores and 45,056 threads, targeting hyperscale AI “factories.” Customers including Anthropic, OpenAI, SpaceX AI, ByteDance, CoreWeave and Oracle Cloud Infrastructure plan to adopt Vera, while server makers such as Dell, HPE, Lenovo and Supermicro will build standalone Vera CPU systems. This push expands Nvidia server processors beyond GPUs into full ARM-based platforms, deepening the ARM processor vs x86 contest in cloud and enterprise data centers. As more operators favor custom silicon optimized for specific workloads, Vera’s aggressive bandwidth-focused design signals how quickly general-purpose x86 dominance is being challenged in high-end server deployments.
