Real-time streaming for the agentic era with NVIDIA

NVIDIA's new Vera CPU officially launches today. If you're running data infrastructure for AI agents, it's worth paying attention to. 

Redpanda worked with NVIDIA on early NVIDIA Vera CPU testing on workloads processing billions of messages daily and saw dramatically better performance than other systems we've benchmarked.

What 1.1 trillion messages a day looks like on Vera

Agentic workloads change the shape of datacenters from GPU centric to GPU and multi-core CPU centric workloads as agents fetch data from databases, stream events, connect to sinks and destinations. NVIDIA Vera is built to process more instructions, anticipate application behavior, and move data across large numbers of concurrent environments without becoming the bottleneck. That's what makes it interesting for the kinds of mission-critical workloads our customers are exploring. NYSE, a leader in financial services, is one of them.

“At the NYSE, our focus is to optimize the latency, throughput, and reliability of the systems underpinning our unrivaled infrastructure,” said Lynn Martin, president of NYSE Group. “The NYSE processes more than 1.1 trillion messages per day, and in collaboration with Redpanda, using NVIDIA Vera CPUs, we will be scaling our capacity while further optimizing latency to power a high-performance, resilient, and AI-ready market infrastructure.”

In addition, HPE announced collaboration with NVIDIA and Redpanda to power real-time data infrastructure based on HPE ProLiant Compute DL394 Gen12 and NVIDIA Vera CPU.

Why Redpanda over others?

We first ran the numbers when NVIDIA announced Vera at GTC in March, and you can read the results in our NVIDIA Vera performance benchmark blog. Vera represents a new direction in CPU architecture with higher per-core memory bandwidth, enabling our customers to move and process data at a scale that wasn't possible before, unlocking new AI and agentic applications.

We ran multiple sets of tests in the full benchmark, and here are the highlights: 

• Redpanda Streaming – 5.5x lower latency: The streaming test focused on P99 latencies — the longest response time for the fastest 99% of operations. P99 matters because it's where bottlenecks, lags, and SLA violations show up under load. In a cluster configuration, Vera handled real-time streaming workloads with 5.5x lower latency than competing CPUs, and, unlike other architectures, latency continued to improve as cores scaled.
  
  
• Redpanda SQL – no bottlenecks on the hardest join: Multi-table joins are a core operation in analytical workloads and AI evaluation pipelines, where you're joining model outputs against ground truth tables, metadata, and scoring dimensions. The Star Schema Benchmark Query #4 is one of the most join-heavy tests available, spanning all four tables in the schema. It's designed to expose bottlenecks. On NVIDIA Vera, Redpanda SQL found none.

Redpanda is built for production AI

Redpanda provides the governance infrastructure enterprise AI agents need to operate safely in production, purpose-built for the world's most mission-critical environments. It's all built on a battle-tested data platform that unifies data streaming, intelligent connectivity, and SQL analytics into a single system, delivering high throughput at low latency and unmatched value.

Want to govern mission-critical agents on the fastest streaming infrastructure available? Let's talk.