NVIDIA Vera CPU Outperforms Intel and AMD in Early Benchmarks, Phoronix Reports
Chip giant NVIDIA has unveiled impressive early benchmark results for its new Vera CPU, a processor specifically designed to power next-generation agentic AI workloads and AI factories. According to testing published by Phoronix, the NVIDIA Vera CPU delivers significant gains in performance, memory bandwidth, and power efficiency compared to traditional x86 processors from Intel and AMD.
The Vera CPU is built around 88 custom NVIDIA Olympus cores and offers up to 1.2TB/s memory bandwidth through its LPDDR5X memory subsystem. NVIDIA says the processor is optimized for AI factory operations that rely on branch-heavy runtimes, orchestration tasks, database queries, code compilation, and large-scale software coordination.
NVIDIA Olympus Cores Deliver Major Performance Boost
At the core of Vera are NVIDIA’s custom Olympus CPU cores, fully compatible with the Armv9.2 architecture. The processor uses a monolithic die design alongside NVIDIA’s second-generation Scalable Coherency Fabric to keep data flowing efficiently across all 88 cores.
Benchmark testing by Phoronix on a single-socket Vera CPU, rated at 450 watts TDP, showed strong performance across several demanding workloads, including:
- Linux kernel compilation
- File compression
- Video transcoding
- Python and Java workloads
- Database management
Michael Larabel, founder of Phoronix, described Vera as “the most formidable competition to Intel and AMD x86_64 processors ever realized.”
The testing also showed Vera compiling a default Linux kernel in just 20 seconds, making it the fastest processor tested by Phoronix for that workload.
Massive Memory Bandwidth Gives Vera an Edge
Memory bandwidth is becoming increasingly critical for AI infrastructure, especially with multiple AI agents and sandboxed workloads running simultaneously.
NVIDIA says Vera’s LPDDR5X memory subsystem delivers up to twice the peak memory bandwidth of traditional CPUs while consuming significantly less power. The company claims the memory subsystem uses less than 30 watts, compared to more than 100 watts for conventional DDR5-based systems.
In STREAM TRIAD testing, Vera reportedly sustained 90% of its peak memory bandwidth, one of the highest efficiency levels ever recorded by Phoronix. The CPU also achieved more than four times the memory bandwidth per core compared to traditional x86 processors.
AI research company Prime Intellect also found that Vera maintained low and consistent memory latency even as parallel workloads increased, an important requirement for agentic AI systems.
Vera Shows Big Leap Over NVIDIA Grace CPU
The new processor also demonstrated a substantial generational improvement over NVIDIA’s previous Grace CPU architecture.
According to Phoronix benchmarks, Vera delivered a 1.6x geometric mean performance increase over Grace and a 1.5x overall advantage against a latest-generation 128-core x86 processor.
Larabel noted that Vera achieved around 10% better overall performance than AMD’s EPYC 9575F high-frequency processor in geometric mean testing.
NVIDIA Expands Vera Ecosystem Support
At NVIDIA GTC, NVIDIA announced broad ecosystem support for Vera from AI companies, cloud providers, supercomputing centers, and infrastructure partners.
The company has already delivered early Vera CPUs to several leading AI firms and cloud providers ahead of broader partner availability planned for the second half of the year.
Vera systems will be available in both single- and dual-socket configurations with air-cooled and liquid-cooled deployment options for enterprise data centers and large-scale AI factory infrastructure.
More details about the processor are available on the official NVIDIA Vera CPU page